Cross Platform DAW Multiprocessor X-Scaling Performance :

[TAFKAT]

Hey All,

I have spent probably more time and energy than most on the x-scaling issues that we have been navigating on Cubendo ever since the Octocores hit the wild, so I thought I would give a quick summary of where exactly the state of play is at the end of seq4. I am not interested in regurgitating the Soap Opera that played out over at Fredo.com, anyone interested and with a few spare hours and some brain cells to kill can search that thread out, I will however quote some relative information forwarded by the developers.



Just wanted to give short update to the situation:

We have looked into the reported problems and were able to reproduce them on a Clovertown system. It turned out that projects on multiprocessor machines which are run at low latency and contain high numbers of output objects indeed result in an unexpected low level of performance. Further analysis has shown that the synchronization between dependend audio signal paths takes more overhead the more processors are being used. It is planned to optimize this behaviour for the next Nuendo release currently scheduled for Q3 2007.

Lars
Steinberg Media Technologies



This was the first official statement acknowledging the 2 issues that have been collectively rolled into the "x-scaling bug"

The first was the accumulative primary loading when using high I/O, which was further complicated as the core count rose - ASIO Serialisation , and the higher than expected increase in diminishing returns at lower latency - Synchronization Delays.

This was confirmed in greater detail at a later date -



Actually, I can give you more news about this issue.

From development: "The problem is a serialization of the ASIO driver, which eats upto 40% of the processing time. Together with the other synchronization delays only 25%-30% of the 1.5 millisecond timeslice can be used for processing. This is not very efficient.

In addition on OSX there seems to be an incredible increase of the synchronization delays on 8 cores compared to 4 cores."

Fredo

Now the second statement didn't ring so strongly until after the release of N4 and numerous updates had still left the x-scaling across both platforms lacking, until I discovered and disclosed that a recent microcode update on the original 5000 Series Xeon systems had in fact alleviated a lot of the issues with the diminishing return at lower latencies, well on XP at least.

In short Steinberg were digging in the wrong place , the issue was not in fact at the application level as they had official acknowledged, but in fact a hardware arbitration/synchronisation issue that Intel had resolved. The performance improvements however did not translate to the OSX platform unfortunately, so my disclosure was simply dismissed , and the official line that x-scaling was still broken at low latencies across the board, has been maintained to this day .

Steinbergs Official Knowledge Base article can be found Here. (http://knowledgebase.steinberg.de/278_1.html)

Now compare the information offered , with the graph below.

http://dawbench.com/images/dawbenchdsp-xp-v-osx.jpg

It is clearly evident that the x-scaling performance at very low latencies is not consistent across the 2 platforms, it is quite good on XP/Vista , absolutely abysmal on OSX, so Steinbergs blanket statement that it is still broken at latencies below 256 across the board, is more than a little inaccurate.

I can not understand why its is so hard for Steinberg to acknowledge that fact , and give the end users clear and accurate assessment instead of some corporate spin that is of no benefit.

To highlight further just how efficiently Cubendo is scaling with multiple cores/threads on XP, check out the graph below with the new Intel i7 , where the arbitration/synchronisation bottlenecks of the previous architecture have been further improved.

http://www.dawbench.com/images/dawbench-core2 v i7-4.JPG

This is all good news , right.. ?

Whats even harder to understand, is that they can't even acknowledge that they actually fixed the serialisation of the ASIO driver in regards to the accumulative loading with high I/O. Its not even mentioned in the fixes for N4.1, unless of course acknowledging the fix is actually admitting there was an issue in the first place.. ??

Either way, I'll be interested what Seq5 brings to the table..

Update : 12 June 2009 :

http://www.dawbench.com/images/blofeldsdsp2-xp-v-osx-0609.JPG


Thanks to Scott and the team over at ADK who have graciously offered their time and energy in testing a new Mac Pro - Dual Nehalem Xeon E5520 2.26 GHZ- on both OSX and Vista64 ( Bootcamped)

The above results are confirming that Cubendo seq5 still sucks on OSX compared to Windows , so I can't see the geniuses at Steinberg moving off their official stance of it being broken across the board on 8 cores and above, let alone their continuing ill informed stance on Hyperthreading on the new i7 architecture.

We can only hope someone there has the capability to start addressing the information in a more accurate manner.

* Ongoing testing and reports have been expanded to include multiple DAW hosts on both OSX and Windows , this thread is no longer focussed exclusively on Cubendo*

[psvennevig]

Thanks for the post Vin.

I can't wait for the Xeon i7 workstations. They will rock.

I just want to tell you guys about something I finally discovered with OSX and proved a few days ago.

History:
Since I started using OSX (intel) It has always been my belief that with Cubendo the new Intel MACs were doing a lot of CPU Speedstepping (throttling) even at quite high CPU loads. And as we all know, CPU throttling at low latencies kill performance and stability and effectively kills drop-out-free audio.

Today:
I finally proved on my Latebook PRO that this problems IS indeed CPU throttling.
Using Vins latest universal test (the same as used in the graphs above) I was able to only put around 8 MB5 dynamics in at 64 samples latency (internal soundcard).
When I finally were able to lock the CPU freq. at full speed (used the cool app. CoolBook (http://www.coolbook.se/CoolBook.html) )
.
Now I could easily drop in over 30 MB5 compressors and Cubendo was 100% stable.
VST meter was still. I could load it up to close 100%, no drop outs.
I have NEVER been able to load Cubendo this much under OSX at all.
To me this shows that the same CPU throttle is a problem with the 8 core MAC pro as well. But I do not have a tool to lock the CPU freq. on the MAC pro.

Great news? I think so. I have notified Steinberg about my findings just in case they were not aware.

Waiting for Seq5.

Pål

[TAFKAT]

Hey Pal,

Wow Man, you may have just found the holy grail in regards to why the scaling on OSX has been so neutered compared to XP. It also explains why the microcode updates didn't deliver the same scaling improvements that I reported on the Xeon systems, despite the architecture being very similar. Whereas the Xeon / XP platform has the facility to access and disable EIST and C1 halt state settings in the BIOS , the Apple EFI system is nowhere near as accessible, therefore the CPU throttling needs to be controlled either by or within OSX.

Do you know if the behavior is consistent with other audio applications .. ?

Peace

[leggy]

What a shame this isn't working on the Mac Pro. Wonder if it would be worth dropping the developer a line to see if he could come up with something?


Rob

[psvennevig]

Logic do not have this problem in OSX.
I have to test to be sure. But I'm pretty sure the Apple Devs. have told OSX in the Logic code to not accept CPU throttle.

I have read a lot of threads at the Apple dev. forum, and I think cpu throttling can be controlled from the Carbon/Cocoa API on OSX (of course not being a dev. I might be wrong).
But at least Logic do not show the problem. Also Logic use a whole other kind of buffering, so it might be several reasons.

P

[TAFKAT]

Well if Logic isn't displaying the issue, that means hopefully that the Steini devs have an avenue to the same solution..

Then again, we are talking about Apple... :-(

[Animus]

That sounds like encouraging news! thanks Pal!

[Thonex]

Pal..... You're the MAN!!!!!!

[Softy]

Pal..... You're the MAN!!!!!!

Yeah! +1!

[psvennevig]

Thanks, I'm blushing.

P

[TAFKAT]

Thanks, I'm blushing.

P

..is that what that is, I was meaning to ask you what the tinge was... :-)

[TAFKAT]

And so it continues.. :willy_nilly:




.. Regarding Octo-Core and further improvements: later on this year we will release an update regarding this.

More information can be found here:

http://knowledgebase.steinberg.net/297_1.html
http://knowledgebase.steinberg.net/278_1.html

Gr,

Chris

[psvennevig]

...and here I was thinking you wouldn't play the "soap opera" anymore???!!

Bad Cop

[TAFKAT]

...ohh didn't you hear, we have been signed for another season ..:D

Seriously tho, don't you think its about time that they actually got their shite together and just gave the end users some accurate information, I mean it isn't that hard, and its actually good news.. , well on Windows at least :icon_yes:

[Paxophobe]

In regards, I think that it would be easy to test this C1 state theory for OSX on a Hackintosh as that option would then be available in the BIOS.

I am really curious how well Logic scales. It would strike me as funny if the low latency performance was equally abysmal on the "OS for musicians" ;)

[TAFKAT]

..its a little more than C1 Halt, its actually the EIST- Speed Stepping, oh, and its more than a theory.. :-)

More detail soon..

Good idea re the Hackintosh, as that has more control at BIOS level than EFI.

O.K who's up for it.. ?

[Paxophobe]

Whoops, thought it was C1.

Funny I don't get an EIST option in my BIOS if I do any sort of overclocking. At stock its available, though. I am going to assume that EIST is probably automatically disabled as it perhaps causes instability.

Anyways, awesome. =)

[Paxophobe]

..its a little more than C1 Halt, its actually the EIST- Speed Stepping, oh, and its more than a theory.. :-)

More detail soon..

Good idea re the Hackintosh, as that has more control at BIOS level than EFI.

O.K who's up for it.. ?


I'm up for the Hakintosh. Sent you a PM.

[TAFKAT]

Just blowing the dust off this thread just to see how long it takes Steini to actually deliver some viable information in regards to this whole soap opera..

Apart from the knowledgebase articles listed previously, there is also this added note in the publicly viewable FAQ for Cubase 5 :


What is the status of Octocore support in Steinberg products?
The upcoming generation of Cubase 5 and Nuendo 5 will offer improved support for these processors. Cubase 5 will see an update later this year that will further improve support for 8 core CPUs, while Nuendo 5 will contain those improvements on its release. We will keep you informed of developments through our website and the Cubase and Nuendo forums.

I am currently running a new series of x-scaling shoot outs across numerous audio apps, so I'll be sure to do a head to head with C5.01 and what ever the upcoming generation delivers..

Anyone on OSX is welcome to chip in at any time, considering the above information is specifically related to the scaling on OSX far more so than XP/Vista , and from further information I have , its going to take way more than some optimisations at Steinbergs end to sort the scaling on OSX.. :eusa_whistle:

[Daryl]

If they don't hurry up we'll all be using 12 and 16 core machines, and that's with hyperthreading switched off...!

D

[unanimity]

It's mainly to do with the fact that Intel processors are "non-native".

[Sam]

It's mainly to do with the fact that Intel processors are "non-native".

What the??? you sure you aint messiniac?? ;)

[Softy]

Just blowing the dust off this thread just to see how long it takes Steini to actually deliver some viable information in regards to this whole soap opera..

Apart from the knowledgebase articles listed previously, there is also this added note in the publicly viewable FAQ for Cubase 5 :



I am currently running a new series of x-scaling shoot outs across numerous audio apps, so I'll be sure to do a head to head with C5.01 and what ever the upcoming generation delivers..

Anyone on OSX is welcome to chip in at any time, considering the above information is specifically related to the scaling on OSX far more so than XP/Vista , and from further information I have , its going to take way more than some optimisations at Steinbergs end to sort the scaling on OSX.. :eusa_whistle:

You know what? I'd be willing to go through the whole hackintosh ordeal, if someone would hand-hold me through the process. I could spend the time and follow directions from somebody who knows what he's doing. Being a mainly PC guy who's switching to the Mac, it would make sense for me to delve into that. Maybe I'd find out that I actually can't switch Nuendo to Mac OS X.

[MattiasNYC]

You know what? I'd be willing to go through the whole hackintosh ordeal, if someone would hand-hold me through the process. I could spend the time and follow directions from somebody who knows what he's doing. Being a mainly PC guy who's switching to the Mac, it would make sense for me to delve into that. Maybe I'd find out that I actually can't switch Nuendo to Mac OS X.

I believe this is a good resource to get you started...

http://wiki.osx86project.org/wiki/index.php/Main_Page

[TAFKAT]

What the??? you sure you aint messiniac?? ;)

Took you long enough Sam.. :D

Now where is that ignore button.

[TAFKAT]

You know what? I'd be willing to go through the whole hackintosh ordeal, if someone would hand-hold me through the process. I could spend the time and follow directions from somebody who knows what he's doing. Being a mainly PC guy who's switching to the Mac, it would make sense for me to delve into that. Maybe I'd find out that I actually can't switch Nuendo to Mac OS X.

Hey S,

The Hackintosh maybe a work around to some of the scaling issues due to being able to actually get to the EIST and C1 Halt state settings.

On the real iFruits tho, its not as simple as some believed , and it deals with custom kernel scripts, etc.

The results have been close , but no cigar.

Sam here has a lot of experience in Hackintoshing one of my Quadcore systems, and will be able to give you a direct heads up on comparisons. I think it was still not anywhere near the XP install.

BTW: Believe it or not I have been integral in helping some high end Macnoids connect into a brain trust to try and sort the scaling issue on OSX due to the O.S/Hardware not being capable of maintain the full clock speeds , simply because I wanted Nuendo on OSX to get closer to what we are achieving on XP/Vista , so our genius friends at Steinberg can finally "officially" acknowledge that Cubendo is scaling well on 8 cores and above , not in any way to help the iFruit campaign for obvious reasons..:wink:

With a Hackintosh you will be constantly chasing a ghost tho, as Snapple will plug any holes at each release, and you will then need to go hacking all over again.

[Sam]

With a Hackintosh you will be constantly chasing a ghost tho, as Snapple will plug any holes at each release, and you will then need to go hacking all over again.

Not with mine ;)

An emulated EFI inside the hidden 200mb partition on a GUID formatted drive is the way to go...I installed my OSX straight of my Leopard disk....then updated it over the net from apple update :)

If your hardware requires you to run a patched kernel of some sort (AMD processors for instance) it is a bit more involved....but if your hardware is very compatible like mine - this way is the best option....

[TAFKAT]

Ohh, O.K thanks for the heads up..

Do you know whether that will hold true with the newer systems. i.e i7.

Either way, the comparative performance is still not on par from my understanding.

Do you have any info there you can share in regards to the direct performance comparison between N4 on OSX/XP on the rig.

[Sam]

Not sure about the newer stuff Vin, I have'nt kept pace - I just got mine to work...and then put it to work ;)

Yeah - Cubendo on winxp is still a clear winner for me...i dont have graphs - just intuition and uneasy experiences with anything under 256 buffers on OSX...

[TAFKAT]

I have added this info on the first post of the thread as well..

http://www.dawbench.com/images/blofeldsdsp2-xp-v-osx-0609.JPG

Thanks to Scott and the team over at ADK who have graciously offered their time and energy in testing a new Mac Pro - Dual Nehalem Xeon E5520 2.26 GHZ- on both OSX and Vista64 ( Bootcamped)

The above results are confirming that Cubendo seq5 still sucks on OSX compared to Windows , so I can't see the geniuses at Steinberg moving off their official stance of it being broken across the board , let alone their continuing ill informed stance on Hyperthreading on the new i7 architecture.

We can only hope someone there has the capability to start addressing the information in a more accurate manner.., should I hold my breath.. ?

[psvennevig]

Yes! Hold your breath...
You might be surprised.

/Pål

[TAFKAT]

Yes! Hold your breath...
You might be surprised.

/Pål

At what, whether I can hold it for more than 3 minutes without passing out... :eusa_whistle:

[psvennevig]

he he,

That would be fun.
But I was thinking more in lines of Cubendo and OSX.

P

[TAFKAT]

:D

What I find really intersting in all of this is that the issues related to the scaling in Cubendo in OSX is directly related to the inability to sustain the CPU's at full clockspeed, which can be attributed to EIST and C1 Halt , however in XP/Vista it is far less detrimental, simply changing a power profile will at least ensure the EIST portion of the equation , whereas in OSX it seems that the stepping needs to be controlled at the application level, which I find ludicrous..

This is evidenced in the fact that Logic scales substantially better than Cubendo as you know, ( a little inner circle coding there doesn't hurt ) , mind you its hard to do a direct scaling comparison at latency settings due to Logics hybrid playback engine which always plays back at a far higher setting , unless record arm / input monitoring is activated.

A little smoke and mirrors there for the iFruits who don't know any better until of course they have a heavy loaded session at low latency , and they wonder why it collapses when they track arm/input monitor a few tracks , shifting the playback engine into a lower latency state :willy_nilly:

Anyhow,

Steini still have a mile to go in getting comparative cross platform performance , I just wish they could bloody acknowledge how well its scaling on the Windows platform , instead of polishing the fruit- so to speak.. :eusa_whistle:

[psvennevig]

I too would like if they acknowledged how the scaling in fact works on multicore and Windows.

But try to load a 1 hour 96khz project with a lot of audio events.
Then it suddenly doesn't matter that much.
What it proves is that with heavy plugin load Cubendo in fact scales very very well today in Windows.

I still think there is room for improvement on other load matters of Cubendo even on Windows. On OSX we need it in all matters of Cubendo load.

/Pål

[leggy]

Do you think Snow Leopard and Grand Central will make much difference? I'm hanging on till September to see what happens but then I'm pretty sure I'll be back to PC land unless something spectacular happens with the new OS.


Rob

[psvennevig]

I don't think Grand Central will help much for specialized apps like Cubendo.
But will make wonders for many other apps.

I have high hopes on the other hand for Nuendo 5 (and new Cubase 5.x) on SNow Leopard.

Pål

[TAFKAT]

Just spotted this posted on the 25 August 2009, I am making a note of the date just to highlight how idiotic and ridiculous that this misinformation is still being peddled by the official drones !!!


No, Hyper-Threading and real time audio just don't get along too well as it is now and thus there won't be much improvement on Hyper-Threading.
Nevertheless, we will continue our research on this topic in cooperation with Intel but based on the results we have, our recommendation to disable Hyper-Threading will still be valid with upcoming versions of Cubase and Nuendo.
_________________
Support Manager
Steinberg Media Technologies

What absolute , total and utter misinformed crap !!!!!!

So Steinberg are claiming on the research and results that they have , in co-operation with Intel , that HT needs to be disabled on i7's, well how about sharing that information !!

This was posted in the Mac Forum of Cubase, Here (http://www.cubase.net/phpbb2/viewtopic.php?t=122320) , so I have to assume it was directed to OSX, however how the fark do they expect to disable HT on a MAC , ohh, hang on that pearler is in relation to a PC , where HT is working perfectly for the vast , vast majority of Cubendo end users.., except of course the geniuses at HQ !

Not much improvement with HT , well I suppose better scaling at lower latencies than going from 4 to 8 physical cores of the previous gen, could somehow be construed as "won't be much improvement"

Even more ridiculous that their recommendation is still valid for future versions of Cubendo, so how in farks name are they claiming an 80% improvement in N5 , this a just a running joke.

And this folks, is from the " Support Manager "

:willy_nilly:

[psvennevig]

I agree that the information sounds like bullshit.
And one head stating 80% performance gain and the other stating that HT must be turned off, is not a good example of great info nor correct.

I how ever do not agree with the language and the way this is presented here.
Everyone knows you disagree with this erroneous info, but why attack this way?

Best,
Pål Svennevig

[TAFKAT]

Pal,

What Language ?

How many times do we have to go thru this before someone at HQ gets a clue and amends the information to be accurate to the real state of play with these systems.

I am sick to death of having to explain to prospective clients that in fact the i7 HT systems work very well , actually incredibly well with Cubendo, when the idiots at HQ continue to officially peddle the opposite.

This is just a ridiculous position to be in where I am having to sell the scalability of the application, in direct opposition to what the official stance is.

You explain to me how I should approach this if you feel attacking them is inappropriate, throw me a bone here Mate.. !!

A moderate approach doesn't work, they simply turn up their nose's in their self absorbed arrogance ignoring any an all quantifiable information , and continue to shoot themselves in the foot by not embracing the true potential of the scalability and peddling this bullshite.

[psvennevig]

Attack is not the solution. We will for sure end up loosing.

I agree. HT on Nehalem surely works. It even works on my MAC pro with OSX.
All my friends with i/ run with HT on, really good.

So, where is the scientific facts?
:pop_corn:

P

[TAFKAT]

Pal,

I would love to know what the actual solution is to try and get thru to these guys, seriously, I am way , way out of ideas , this is just my frustration spilling over again..

I'm with you, I can't wait to see the quantifiable facts that they have collated between theirs and Intels collective gene pool..

Pass that Popcorn, and here have a few beers while we wait !

:pop_corn::emote_beerchug:

[MattiasNYC]

I agree that the information sounds like bullshit.
And one head stating 80% performance gain and the other stating that HT must be turned off, is not a good example of great info nor correct.

Seems like par for the course for SB.... unfortunately....


I how ever do not agree with the language and the way this is presented here.
Everyone knows you disagree with this erroneous info, but why attack this way?

Best,
Pål Svennevig

P, Vin has actually spent a great deal of time and effort trying to see what causes various systems/platforms to perform differently. I'm not going to get into details because I don't want to misrepresent the way this issue developed, but sufficed to say Vin really did try to get SB to address the issue, and in so doing actually gave valuable information. In the end, as far as I can recall, his research even should have put SB and Cubendo in a better light than it was, and the reason it wasn't was because SB were too stubborn (ignorant / incompetent / ego-centric; you pick a word) to acknowledge his work.

So there is a history here and it's actually not only between Vin and SB. Others have helped as well. But it goes back a while.

SB simply doesn't listen or doesn't comprehend what is important. That's what it is. After having banged ones head against a wall for long enough "the language" escalates. A company with skilled personel on every level knows how to deal with customers. SB seems clueless. So this is what you get.

By the way, I really don't think it was all that bad. It's just that it's accurate so it sounds bad.

[TAFKAT]

So, where is the scientific facts?

Just saw the official response to your question posted at C.net, and of course absolutely no detail , just unsubstantiated B.S.


Pål, there's no specific plug-in that has issues with HT. It's rather a general issue that HT almost all the time has a negative effect on the overall performance when it comes to realtime audio. Especially if you what to run on very low latencies around 64 samples, turning off HT might do the trick.

This is not the experience with i7 HT on Windows, so they need to be specific in their so called knowledge base article and their ongoing stance is specific to OSX.

Even on OSX, I would trust your opinion over theirs.

That new K.B still continues to correlate the HT on i7's with the older HT of the Netburst architecture, are these guys really so far out of touch ? There is not one other DAW manufacturer who maintains this stance on Windows , not one , all other have acknowledge and embraced the performance benefits of HT on the i7's, and why wouldn't they. Its a positive thing, something that seems lost on the geniuses at Steinberg !

Now they may have me locked out at Cubase.net, and have now locked that so called Knowledge Base against hot linking from here , but seeing they want to continue making these official statements , without providing any quantifiable evidence, I am going to take the challenge to Mr Doll publicly , across multiple open forums..

I will start with a direct thread and even sticky in the general area here to start off with, and I may even move then across to G.S.

Time we put this to bed.. !

[psvennevig]

I'll do tests on OSX (again) this evening with both Blofeld benchmarks and my own real life mixes from the Studio.
I'll keep HT on and OFF recording the differences.

Post back when I have something.

P

[TAFKAT]

Hey Pal,

Thanks for that.

BTW: I will have my new CubenDAW DSP benchmarking test sorted shortly , which I have developed specifically for the cross platform shootouts on the Hackintosh platform. Basically the same Plugin choice of Blofelds DSP II, ( Steinbergs inbuilt MBC ) but with the music content from DAWbench DSP , I think we can give Blofelds a wide birth from here on in, Man, I have heard that bloody 16 bar loop so many times over the years, its imprinted in the back of my brain.. LOL.

DAWbench DSP Universal is not as accurate cross platform because the 2 chosen plugins will not be comparable performance wise between Windows/OSX , so for this specific phase of testing, the new session will be the default from here on in.

You can quickly configure the new session yourself easy enough , just call up DAWbench DSP , and copy the channel strip from Blofelds DSP II to the loaded Sine Channels , done.

[psvennevig]

I will do just that for my tests.

P

[TAFKAT]

Thank You..

BTW:

Can you drop a vote a few notes in on the i7 Hyperthreading Poll thread in the General Area.

You are one of the few Cubendo Power users across both platforms, and it would be great to have that info included there as well..

Thanks Mate..

[TAFKAT]

Quick heads up..

I have my i7/X58 OSx86 system up and running with 10.5.7 , so I'll be doing some head to head shootouts with Cubendo over the next few days.

With EIST and C1 halt disabled at BIOS level, its going to interesting how the system scales on OSX compared to the retail Macs.

[psvennevig]

Quick heads up..

I have my i7/X58 OSx86 system up and running with 10.5.7 , so I'll be doing some head to head shootouts with Cubendo over the next few days.

With EIST and C1 halt disabled at BIOS level, its going to interesting how the system scales on OSX compared to the retail Macs.

...and interesting to see if it actually runs at full CPU speed.

Pål

[TAFKAT]

Hey Pal,

I need an OSX application to accurately report clock frequency , as the system prfiler is not recognising the CPU or clockspeed correctly.

Any ideas ?

Thanks

[TAFKAT]

Quick heads up,

Just finishing up the first round of OSX v Windows Hyperthreading shootouts.

To say the results will raise a few eyebrow is an understatement.

An i7 920 system has had the EIST/C1 Halt state disabled at BIOS level , eliminating those variables out of the equation, and configured both with OSX 10.5.7/8 and XPSP3. We also have results for a new MacPro Dual i7 Xeon E5520 running both 10.5.8 and 10.6.0 - Snow Leopard.

Stay tuned..

[Andrew J]

Oooh, you're a tease Vin!

[psvennevig]

Fark, forgot to post results here yesterday:
Here ya go.
Vin will supplement with his fancy excel graphs I guess...
-------------------------

Here are some tests I've done


Here are my test results with Nuendo 4.3 in both Leopard 10.5.8 and Snow Leo 10.6

As stated by Steinberg, HT ON is worse (at least on these tests) for low latency on Leopard.
However on Snow Leopard its different. Here HT is the clear winner.

Numbers are taken from a new Cubase and Nuendo benchmark made by Vin. Vin is the developer of the famous Dawbench DSP benchmark.
This new benchmark have made some changes from the recent Universal release. its using (once again) the native Steinberg Multi Band Compressor VST plugin that comes with C4/N4/C5.
This makes the test truly universal between OSX and Windows.
You need to download the Dawbench DSP benchmark from here:
http://media.svennevig.no/pas/benchmarks/dawbenchdsp.rar

Then you might need all my test projects (saved at different latency settings)
http://media.svennevig.no/pas/benchmarks/pal_svennevig_cubendo_test_sessions.zip

Here are my test results (technical)

Mac PRO dual quad Xeon 2.26 GHz Nehalem, 6GB RAM
dual E5520 - OSX 10.6 : Nuendo 4.3 : RME MADIe : Driver 2.67

032 : HT OFF - 38 MBC / HT ON - 42
064 : HT OFF - 87 MBC / HT ON - 120
128 : HT OFF - 107 MBC / HT ON - 152
256 : HT OFF - 126 MBC / HT ON - 160
512 : HT OFF - 145 MBC / HT ON - 168

Mac PRO dual quad Xeon 2.26 GHz Nehalem, 6GB RAM
dual E5520 - OSX 10.5.8 : Nuendo 4.3 : RME MADIe : Driver 2.67

032 : HT OFF - 12 MBC / HT ON - 10
064 : HT OFF - 26 MBC / HT ON - 22
128: HT OFF - 61 MBC / HT ON - 64
256: HT OFF - 78 MBC / HT ON - 119

As you can read from the table, HT has clear advantage on Snow Leopard. And a clear disadvantage on the first two latencies for Leopard.
Nuendo 4.3 is also a lot more jerky and skitty under Leopard than Snow Leopard.
In fact I can't believe how much it've changed for the better.

Here are some PICs to illustrate it all. (NB! Large pixels, captured on my 30" monitor)


http://media.svennevig.no/pas2/pas/n4gfx/cubendoDSP_n4_64_buffer_scale.png

http://media.svennevig.no/pas2/pas/n4gfx/cubendoDSP_n4_128_buffer_scale.png

http://media.svennevig.no/pas2/pas/n4gfx/cubendoDSP_n4_128_buffer_scaleHT%20OFF.png

http://media.svennevig.no/pas2/pas/n4gfx/cubendoDSP_n4_256_buffer_scaleHT%20OFF.png

http://media.svennevig.no/pas2/pas/n4gfx/cubendoDSP_n4_512_buffer_scale.png

All piccies from Snow Leopard. Pics from Leopard on request.
Look how good (or at least not bad) the cpu cores scale. Its not far from the ASIO meter either.
Under Leopard this scaling can be described with one word: CRAP.

Next is some "real life" project test. However this is much harder, because its so hard to measure.
How we we differ between one "real life" project and the other.
The tests will how ever be in 96khz, because the way I SEE IT, 96khz is real life today. At least in my studio.

I'll be back!!

Best regards,
Pål Svennevig

[psvennevig]

As you see, huge improvements from Leopard to Snow Leopard.
But still XPSP3 and Win7 is far, far ahead.

P

[TAFKAT]

Hey All,

First off some detail on the test session, Cubendo DSP is a new test I developed purely for cross platform shootouts with Cubedo , using Steinbergs own VST3 native plugins. This will ensure the most consistent cross platform performance for Cubendo across Windows and OSX . Its essentially the audio data from DAWbench DSP Universal and the plugins from Blofelds II DSP.

The i7 test system is my tried and true development system which has been used for all previous i7 testing, but I have also configured an OSx86 10.5.8 install. As I mentioned earlier, unlike a regular MAC, this system allows control of EIST and C1 Halt at BIOS level, which will eliminate those variables from the equation, leaving only OSX and Cubendo. All Windows testing has been on XPSP3 on this run, Win7 64 coming later.

O.K, the numbers..

------------------------------------------------------------------

i7 920 - OSx86- 10.5.8 : Cubase 5.01 : RME AIO : Driver : 2.67 :

032 : HT OFF - 10 MBC / HT ON - 15 ( Hard to get a results as it was spiking and clipping

really early on , haven't seen that behavior since pre Core 2 on windows )

064 : HT OFF - 40 MBC / HT ON - 48

128: HT OFF - 57 MBC / HT ON - 82

256 : HT OFF - 70 MBC / HT ON - 95


i7 920 - XPSP3 : Cubase 5.01 : RME AIO : Driver : 3.081 :

032 : HT OFF - 96 MBC / HT ON - 113

064 : HT OFF - 108 MBC / HT ON - 128

128: HT OFF - 114 MBC / HT ON - 140

256 : HT OFF - 124 MBC / HT ON - 153

-----------------------------------------------------------------------

http://dawbench.com/images/cubendodsp-ht-xpvosx.jpg

Some interesting points,

Its obvious from the i7 results that the performance of C5 is not even on the same page between Windows and OSX , there is something seriously amiss on OSX , even with the EIST/C1Halt state variable removed. At first I was questioning whether my OSx86 configuration was not performing to spec, until I compared my OSx86 results with Pals MAC Pro results, which then placed them in perspective.

In a nutshell, the Hackintosh in OSX 10.5.8 is in a lot of cases outperforming a real MACPro Dual Nehalem, the EIST and C1 Halt control is a major factor there. So to be clear, even with 1/2 the cores/threads, the i7 is outperforming the official hardware, which is being crippled by the bullshite Apple EFI implementation not handling the Intel thermal / clocking routines well at all.

Once we get past that , we then have OSX's MP task scheduling and threading in 10.5.8 tripping over itself, with HT On or Off , and the great white hope of Snow Leopard, although substantially better as displayed, IMO opinion simply polishing a turd.

To top it off Cubendo's ASIO/Core Audio efficiency on OSX sucks compared to Windows, so add it all up , and we have a pretty embarrassing state of affairs.

We still have some more test sessions to run thru that will cover a wider variable in regards to plugins, etc , but I really can't see any significant change to the cross platform results.

In short Steinbergs cross platform stance on Multiprocessor Performance & Hyperthreading is highly inaccurate , and needs to be clarified.

To Download the Test Session Files :

Original DAWbench DSP session with Audio Files : Here (http://195.0.211.246/pas/benchmarks/dawbenchdsp.rar)

Cubendo DSP Session File : Here (http://www.dawbench.com/downloads/cubendodsp.zip)

Both downloads are needed , as Cubendo DSP uses the audio files of DAWbench DSP , simply drop the Cubendo DSP session file into the DAWbench project folder and launch the session.

Full instructions on the methodology used to run the tests in the DAWbench DSP download file.

[Daryl]

Vin, it would be very interesting to see Logic tests on both machines, as well as DP tests. I have a feeling that Logic won't perform much better on the Hackintosh, but that DP might. Any thoughts?

It might at least help to sort out whether SB or Apple is truly at fault for the abysmal performance of Cubendo on OSX.

D

[TAFKAT]

Hey Daryl,

Already on it.. :-)

I have incremental Logic and DP tests on the way , both using their own native plugins. Its also going to be impossible to do OSX v Windows shootouts for obvious reasons, but I can do HT ON/Off comparisons , which will be interesting.

Stay Tuned

[Daryl]

Hey Daryl,

Already on it.. :-)

I have incremental Logic and DP tests on the way , both using their own native plugins. Its also going to be impossible to do OSX v Windows shootouts for obvious reasons, but I can do HT ON/Off comparisons , which will be interesting.

Stay Tuned
I would laugh if Logic ran better on your machine than on a Mac. :icon_lol:

D

[TAFKAT]

Hey Daryl,

It would not surpise me at all to be honest, controlling the EIST and C1 obviously had an influence on the Steini results.

I am already 1/2 expecting a visit from some guys in white suits , or is that black skivvies... ?? :eusa_think:

I'll have the cool aide , chilled and ready... :eusa_whistle:

[TAFKAT]

Quick heads up,

Both myself and Pal are knee deep in a whole new round of tests, throwing as many variables as we can into the mix.

We have an actual "Real World" session - cool blues track played by live musos no less , running at 96K and at 48K, we also have Cubendo DSP-96K .

Some interesting observations in regards to HT with the RW session , but very difficult to quantify it , the performance difference is not as pronounced as the incremental tests, and very difficult to measure and report on definitively. Still navigating the possible variables, one being what we believe is that a collection of different plugins will place a much higher ASIO engine loading due to arbitrating multiple PDC in real time.

The discrepancy between ASIO loading and CPU/TM is quite dramatic at 96K as well.

What is still consistent is just how far the performance variance is between Windows and OSX at lower latencies, and also the performance variable between the MACPRO and i7-OSx86 system i.e, 96K RW session would only play on the MACPro and i7 at 128 samples, but down to 032 on XP , 48K session which is far more heavily loaded , which maxes the MACPRO at 064 , and runs only with HTOFF, runs fine on the i7 OSx86 system at the same latency, with HTON/OFF , haven't tested that in Windows yet, but have no concerns it will run down to 032 , the 96K DAWbench DSP totally collapsed on OSX below 128 , I mean totally collapsed, still need to run up XP, but again, I am betting it will be fine at far lower latencies. It will be interesting if it runs at 032, seeing that is 1/2 the latency again.

Stay tuned, huge report coming , test sessions , etc, etc..

[paulwr]

wow, this is a ton of info. Thanks.

-Paul

[TAFKAT]

Some more snippets..., a few curve balls, and some further clues in what variables are at play here..

Real World 48K session on i7-XP at 064, was not even raising a sweat , remember this was peaking on the MAC Pro , even with 10.6 I might add, it also handled 032 samples without barely a blip..

Now moving to the 96K Cubendo DSP session which collapsed on OSX, with HT OFF , cruised thru from 032 to 256 , about 1/2 the number of MBC's of the 44.1 session, which is about right with the lower latency and higher DSP overhead, with HT ON however is where it gets really interesting, and I have finally found a crack in HT on XP.

Yep, I admit that I have just managed to get better performance with HT off. Some detail , at 032/064 , ASIO and TM was steady at around 25-30%, but consistent spikes in the ASIO meter, corresponding to audible clicks resulted in no results below 128, 128 was about the same results as HT OFF, again due to ASIO spikes, and 256 was just a touch better. So what we have here is a better understanding on where the balancing act with HT comes in on XP at least.

If you work at 44.1/48.0 , HT is always better on, with 96K, it may be safer to switch it off. However, there is some other things we have discovered on OSX that changes the scaling behaviour which could be something to do with core affinity, so I tried the core affinity tool in this 96K session on XP , sure enough it smoothed out the spikes at 032/064 to be able to get a result, but still below HT OFF... , however it made things worse at 128..

:willy_nilly:

More to come.., but right now I think I need a Bex and a lie down.

[colony nofi]

This is extremely interesting, useful information!
Cheers Vin & all.
My main macpro is early 2008 - not so useful for these tests...
But my second system is the newer macpro 3,1...
Need any more tests done? I can offer up those two systems...(plus other older, but hardly useful...)
Both are still running 10.5.8

But I think you have it pretty much covered already. (Though I have a suspicion and test I do would show up differences between the apogee ensemble and metric halo hardware......)

B.

[TAFKAT]

Hey Brendan,

The Metric Halo and Apogee's as a cross reference will be very useful actually.

I'll give you call this arvo if you are available, PM some details.

Cheers

[TAFKAT]

A few more snippets ,

Starting to expand the testing with a few more end users and systems, and am getting some preliminary reports of the Cubendo DSP 96K session behaving fine at lower latencies on Vista 64 SP2, with HT ON , without using the Core Affinity Tool.. :willy_nilly:

I'll get to Win7 64 shortly, and I know that Pal has moved on from 10.5.8 so the Cubendo testing on 10.5.8 and XPSP3 has pretty much wrapped up , ( for us at least, others are quite welcome to test and contribute with Leopard, XP, Vista ) , from here on in its SL v Win7 64, and before anyone asks.., yes I do have Vista64, and no I will not be bothering testing it... :wink:

I will do some Logic and DP testing on 10.5.8 with the OSx86 system tho, just to see how the dust settles there.

Stay Tuned..

[TAFKAT]

Just a few more snippets before going on radio silence for a while ,

Win7-64 testing is completed , results have proved a touch better than XPSP3 across the board, the severe 96K / low latency issue with HT experienced on XP is not an issue with Win7 ( or Vista64 ), but results are not equal to the 44.1/48 results, so there is still some curves to navigate there. Also there is some incredibly odd spiking from TM in Win7 with any session above 44.1, with the load readout being displayed in a sawtooth pattern, not a steady constant loading as in XPSP3 ( still waiting on some screenies from Vista64 and then I'll post some shots - my Vista 64 activation was blitzed with the Win7 multi boot install, go figure , and I can't be bothered sorting it) . Despite the spiking, performance in Cubendo is fine, so it may just be a priority shift in the display of TM, but only in sessions with higher SR's.

Just to keep the seat warm until we come back with the final round.

i7 920 : 3GB DDR 3 : RME AIO :
--------------------------------------

Cubendo DSP-96K

OSx86-10.5.8. HT OFF/ON

@032 - 0 / 0

@064 - 0 / 0

@128 - 16 / 14

@256 - 23 / 31

XPSP3 : HT OFF/ON

@032 - 40 / 0 ( *AAT 35 )

@064 - 49 / 0 (*AAT 40 )

@128 - 52 / 52

@256 - 56 / 60

*Auto Affinity Tool


Win7-64 : HT OFF/ON

@032 - 39 / 32

@064 - 49 / 48

@128 - 52 / 56

@256 - 55 / 64

-----------------------------------------------------

Cubendo DSP-44.1K

Win7-64 : HT OFF/ON

@032 - 104 / 120

@064 - 108 / 128

@128 - 115 / 144

@256 - 125 / 167

-------------------------------------

DAWbench DSP-44.1K

Win7-64 : HT OFF/ON

@032 - 108 / 135

@064 - 118 / 152

@128 - 124 / 158

@256 - 127 / 161

------------------------------------------

Graphs coming later, Sorry.

The Real World 48/96K tests are harder to accurately quantify and report on , simply because the sessions that run right on the limit at 128 and 064 samples respectively on Pals MacPro Dual Nehalem system, even under 10.6, are not challenging the i7 920 system running 10.5.8 , and under Windows, run easily all the way down to 032 samples. I will be interested how they go on the OSx86-10.6 config when I get it running. I have plenty of screenies of the session running with TM and ASIO viewable, but reading meters is as useless as tits on a bull to try and quantify scalability in this and in fact, any instance.

Pal is travelling this week, and I'll be going back to my normal life and also focusing on getting Snow Leopard configured on the i7-920 ( a few hurdles to navigate there just yet ), so we'll be on radio silence for a while.

For anyone reading in who wants to get their hands dirty with any of the test sessions , drop me a PM.

:009:

[TAFKAT]

Quick heads up,

The testing has now expanded to include Logic 9.0 and D.P 6.0 on both OSX 10.5.8 and 10.6 , I'll also be testing StudioOne , SONAR 8.5 and Reaper 3.x as well as Cubendo on Win 7 , so what was a large undertaking , has now become increasingly larger.

I have changed the thread title to reflect the expanded testing.

Logic is a strange beast to say the least with its set buffered playback engine , so essentially no matter what the latency "setting" , be it 032 or 256, the results on the DAWbench scaling test remains pretty much the same, giving away that the "latency" setting from what I can make out, is input only, and only when track armed. I maybe wrong, so if anyone with some more Logic experience can shed some light, that would be great.

I have configured the new OSX Host sessions as DAWbench DSP Universal - MD5 ( WaveArts Multidynamics 5 ) , the MD5 being one of the original reference plugins . I have selected that for the OSX testing as it has an AU, VST and MAS Native version. Not exactly sure how they will compare, but in DP I can at least test the AU and MAS versions against each other, which should give us a better comparative.

Still finalising my 10.6 i7 configuration, which once settled, I'll post some Leo v Snow Leo Cubendo results..

[TAFKAT]

O.K,

Another quick heads up and a slight change of plan.

All future testing will be focusing on OSX 10.6 and Win 7 only , its just too big a task to try and test continue across 4 x O.S's i.e: XPSP3, Win7, OSX 10.5, OSX 10.6.

I will use the 10.5.x / XP / Win7 numbers I already have to do some comparable analysis for Cubendo on 10.6 with my i7 system , but all future testing with Logic, D.P , Cubendo, etc will be on the 2 newer O.S's only.

I have 10.6.1 Snow Leopard installed and settled well enough to finalise the testing on the i7 dev system over the next few weeks, and then I can compare notes with Pal ( and anyone else who would like to get their feet wet ) and complete a report.

BTW : Pal has just become a new dad in the last few days, so I suspect he will be having his hands full elsewhere for the time being..

Congratulations to Pal and Family.., enjoy the sleep when ever you can get it... :D

[psvennevig]

:sleeping:

[TAFKAT]

Hey All,

O.K I thought while Pal is catching up on some z's , and I had some downtime this morning I would at least close off on the Cubendo DSP-44K results for Win7 x64 and OSX 10.6.1 , as that will be the main interest and focus for many .

http://dawbench.com/images/cubendodsp-ht-osx.jpg
Mac OSX Leopard 10.5.8 v Mac OSX Snow Leopard 10.6.1

http://dawbench.com/images/cubendodsp-ht-w7vosx.jpg
Windows 7 x64 v Mac OSX Snow Leopard 10.6.1

My OSX 10.6.1 results were significantly better than the 10.5.8 results, but the scaling was nowhere near what Pal experienced on the MP Dual Nehalem, hinting that Apple have done some serious work in the area of Multi Processor/ Threading scheduling. All Good.

In isolation compared to the 10.5.8 results, they are quite substantial , interestingly tho, its still nowhere near the Win7 results, and the Dual Mac Pro Nehalem is still not even on par.

Both the Real World DSP 96K and 48 K sessions played back without a hitch at their designated latency settings on the i7 10.6.1 system, not even close to challenging the system actually , the 48K session played back at 032 samples easily with both HT ON/OFF.

In respect to Hyperthreading, 10.6.1 has definitely improved the overall result at these sample rates , the higher 96K sample rate is not showing the same improvement and is an area that we will be investigating further across multiple applications , on both Win7 and OSX 10.6.

Its is extremely difficult to correlate and quantify why Steinberg are maintaining the blanket statement in regards to HT performance at the lower sampling rates. At 96K they may have some leverage , but of course the devils in the detail, and none are forth coming from our friends at HQ, who seem to have gone into their usual ostrich position

I am holding off on finalising those reports until Pal gets back from Paternity leave and finalises some Mac Pro numbers for me :D

:009:

[Animus]

Nice work Vin.

[psvennevig]

I'll update some RWDSP numbers on 10.6.1 in K64 mode for Vin tonight.
K64 mode seems to even help a bit here.

Be back
:sleeping:

[efernan]

Hey Vin,

Excellent work!

Would be nice to see that Mac Pro Dual E5520 running Win7 x64.

Cheers!

[TAFKAT]

Thats already in the works as we speak... :D

Stay Tuned.

[psvennevig]

Hey Vin,

Excellent work!

Would be nice to see that Mac Pro Dual E5520 running Win7 x64.

Cheers!

It will soon. Just a bit busy with my first born...

See ya

[TAFKAT]

Will be blowing the cobwebs off this thread soonish and posting final numbers and reports including the MACPro - Win7 Dual Nehalem results.. :D

Stay Tuned.

[Animus]

Can't wait for the i9 reports! hint hint

[TAFKAT]

Hey All,

I have uploaded a new report on the DAWbench website based on the collated work we did on the cross platform comparative performance of Cubendo. This will probably be my last added focus on the Steinberg product , I'll be moving onto Protools and other native DAW's in future.

Report Here (http://www.dawbench.com/win7-v-osx-1.htm)

[Bman]

Fascintaing read Vin. Thanks. - Brett

[TAFKAT]

With the release of C5.5/N5 I'll be finalising a follow up report shortly , just waiting to get some final detail from Pal in the next week or so on his Dual Nehalem Mac Pro system.

So far its been a total non event for me in regards to MP scaling improvements in respect to the suite of tests we used, which may or may not be 100% reflective, but the promised OSX improvements have not eventuated at my end , and from the noise level at C.Net , most are not feeling the love on OSX there either..

I have had some interesting information forwarded that Cubendo may use a ASIO to Core Audio Bridge, which could explain a lot to why there is still such a massive performance variable, so I have decided to bump up my scheduled testing on StudioOne sooner to try and get an angle there, as S1 uses native CoreAudio..

Stay Tuned.

[TAFKAT]

Just a quick note before posting a final report and personally closing off on Cubendo MP performance pretty much for good.

Reading in over at Cubase.net in the MAC forum its obvious the promised % improvements have not eventuated , plus navigating commentary where they are talking about running tests at 512/1024/2048 and reading ASIO meters is just doing my head in to be honest, add to that the continuing stance from the genius representatives there recommending that Hyperthreading needs to be switched off on i7 on Win7 systems, and I have pretty much thrown my arms up.., this is a total and utter waste of time from here on in, I have way better things to do than continue slamming my head against a wall with these guys.

From the evidence collated on the C5.1 v C5.5 (http://www.cubendo.com/showthread.php?2132-C5.5-Multiprocessor-Performance-Improvements-over-5.1.) performance thread here its pretty conclusive IMO that they have spent the last 18+ months optimising the MP low latency performance specifically for the Core2 architecture that was superseded 18 months ago. How is that in any way beneficial moving forward is beyond me , not to mention the performance variable between MAC OSX and Win7 has remained consistent, so in short they made a lot of promises and delivered absolutely nothing in that respect.

Time to move on, over and out..

[Yoshi]

Amazing thread. Can't way for the tests against another DAWs.

[TAFKAT]

Hey All,

I am in the middle of the cross platform comparative testing with Nuendo 4/5 , StudiOne 1.5 and Reaper 3.6 , using a range of 3rd party plugins - WaveArts MD5 , URS Channel Strip Pro and Elysia mPressor. This is a huge undertaking and will take me a while to work thru all of the respective test session, but I have some preliminary results @032 that are already bringing up some interesting points.

On Win7 , the 3 DAW applications across the respective plugins , are all close to par - being with a few plugs of each other, on OSX however its a totally different story. StudioOne is 60-70% better than Nuendo , and Reaper is substantially better than that, over 100% in some instances. I am also being very conservative with Reaper as well, as I am still navigating a small glitch on playback at the loop point - where I drop back about 15 plugins to clear. Playback apart from that is clean , so there is still potentially an even better result possible if I can get Justin to solve that glitch on loop. I had a similar issue with Reaper V2 on Windows, that was resolved after consultation with Justin.

Across all 3 DAW's, Windows performance is substantially better than OSX, so the comparative performance issue is not reserved to Cubendo , but Cubendo has the largest variance.

I'll post some further detail shortly

[Yoshi]

Do you have any preliminary number about the difference between Cubendo Mac with PC?

I'm installing a W7 x64 in my Mac to run my own tests.

Thanks!

[TAFKAT]

The variable is consistent with the last report across all the new 3rd party plugins.

You can do your own testing with the existing test sessions that you can download at the DAWbench website, I won't be posting the new sessions for the time being.

It will be easiest for you to use Cubendo DSP to save you installing 3rd party plugin demos.

[TAFKAT]

I have finished up all of the new series of cross platform testing on Nuendo 4/5 , Reaper 3.x and StudioOne 1.5.x using the 3rd party plugins.

I'll post some snippets over the next few days , lots to digest and plenty to say in regards to the performance of N4/N5 in light of the supposed performance improvements. I didn't use Cubase in this round of testing as the version numberings in respect to comparable seq revisions is just too confusing, so it was easier to just use Nuendo.

Stay Tuned.

[Yoshi]

Great news :-)

Thanks Vin!

[TAFKAT]

http://www.dawbench.com/images/DAWbench-DSP-Results-10-10.jpg

O.K,

This is the worksheet of all the results across the 4 applications on Windows 7 x64 and OSX 10.6.2.

This was a huge undertaking as each latency run was 28 test passes ( 14 each for Win7 / OSX) , and thats not counting going back and rerunning some test to double check the results. I will be writing a few reports based on the above data, but in short Nuendo 5 does not deliver any performance advantage worth talking about over Nuendo 4 on either OSX or Win7 on the current i7/i9 systems. In some cases it was worse depending on the plugin. Both StudioOne and Reaper also performed substantially better on Win7 over OSX , so the performance variable is not reserved to Cubendo , but what is not evident in the results is that Reaper was light years ahead of both Nuendo and StudioOne on OSX at the ultra low latencies, but had a slight glitch at the loop point, so I backed off the plugins until I cleared that glitch - I could have easily loaded another 15-20 plugins in some instances as the playback was crystal clear apart from the loop glitch. So in short Justin is on to something in regards to scaling on OSX that the other guys haven't quite worked out yet.

There is definitely a lot to absorb, but if you have a closer look at the MD5 results on OSX, you will see @256 that they are not far off Win7 on N4/N5 , also N4 does measurably better with all the 3rd party plugs than N5, which is an eye opener. The plugs with the inherent delay which need the PDC/ADC calcs - the MBC and EMP do significantly worse on OSX than say the 2 plugs without any inherent delay - MD5/CMP. Thats not immediately obvious , so its something that I need to note in the reports.

The other thing is just how much better Win7 feels when pushing the sessions to the limits , OSX just has this underlying feeling of being close to the edge all the time. Mind you it did get ugly on Win7 right at the top end of these tests - ASIO/CPU meters pegged and I mean pegged 100 % , GUI slowing to a crawl, navigation almost non existent, but the playback was still clean. Incredible actually just how well all the apps scaled on Win7. Each had their specific quirks when taken to the limits, Nuendo and Reaper were a little less ugly at those points, StudioOne was getting very- hmmmm, cranky, and you will notice that after a good start on OSX, pretty much fell in a heap even worse than Nuendo.

The 3rd party plugins were all great in regards to stability as well.

Also StudioOne on OSX had some added quirks in regards to loading / saving sessions , activations expiring, etc, which I'll detail in the reports.

One area that I haven't had a chance to test yet is the OSX 64bit kernel, which I have been informed performs a lot better at the lower latencies, and I mean a lot better with Cubendo. I have some current MacPro single Hexacore Xeon 3680- 3.33GHZ numbers that my test partner Pal sent me that are only 10-20% down on my Win7 results using Cubendo DSP- MBC, so there is something definitely in it. I need to refresh my OSx86 install to do that, so that won't be until next year I suspect as that needs new kexts,etc - probably a refreshed bootloader and 10.6.4+ . In the meantime lets see if Steinberg make that official that there is a significant improvement under the 64 bit kernel in OSX, or I may not even bother ... LOL

I am also talking to Justin to get that OSX loop glitch sorted in Reaper, its the exact same issue we had on Windows a few years back and is related to its read ahead buffering. Justin made some adjustments on Windows which resolved it, so if he can do the same on OSX, the results will be significantly better and make a mockery of the other 2 apps. Its really close now.

The Protools LE /MP RTAS testing is next, but I freaked out when I discovered you can't have both LE and MP on the same O.S install, farking prick of a thing, which means I have to do all the testing in LE with my 002 , then test one or a few of the other apps as a cross reference using the same hardware/ASIO/CoreAudio driver, then uninstall LE, install MP and use an M-Audio interface , etc. I may just stick with the official Digi hardware as MP had some quirks..

One last note - All results are with Hyperthreading ON, there are no Hyperthreading OFF results this time around, simply because after close to 2 years and countless hours of testing , with volumes of empirical data showing that in the vast majority of cases it is in fact beneficial to have Hyperthreading ON.

I am not willing to waste any more time arm wrestling the powers at be at Steinberg over their continuing stance that it is in some way preferable to have it OFF.

More to come , I have most of the results already laid out in respective graphs for the individual plugins, and 2 full reports that will be posted at DAWbench.com

[Animus]

Good work Vin! Very interesting results. looks at that mac vs pc discrepancy. Damn. Especially the 32k buffer one.

[Yoshi]

Awesome work, man.

Can you give the spec of the test machine?

Do you use a Hackintosh machine?

If you do, how does it perform? Is it stable?

Thanks!

[TAFKAT]

Awesome work, man.

Can you give the spec of the test machine?

Do you use a Hackintosh machine?

If you do, how does it perform? Is it stable?

Thanks!

Hey Yoshi,

Spec of the system is my old trusty i7 dev system I have had since November 2008 - Standard clocked i7 920 @ 2.66 / X58 / 6GB DDR3 / RME AIO

Yes it has an OSx86 configuration running OSX 10.6.2 currently - purely for testing purposes , its as stable as a rock and performance is arguably better than the real deal due to the more open nature of the hardware. I went into some detail about it in the first OSX v Windows report on the DAWbench site.

I have to be very careful talking about it here tho , so I won't discuss it further online.

Shoot me a PM if you want more detail... :-)

[efernan]

Hey Vin,

As always, excellent work. Congratulations!

I have just a dubt about the chart. What's the meaning of these "RealWorld DSP" notes at the bottom of the Nuendo benchmarks? I can't figure it out, specially related to the other latency entries.

Re OS X & 64-bit kernel, I still dubt that there will be any difference at all, but anyway, I'll try to have you some results about this subject in the near future.

Cheers!

[TAFKAT]

Hey E,

The RealWorld DSP results are the 48K and 96K sessions that Pal developed as an alternate to the DAWbench DSP incremental tests.

The details are buried in this thread somewhere... LOL

In short, they are an alternate mix of a Real World session of a blues band that Pal tracked , that are loaded to the point of break on his Dual Nehalem Mac Pro. The 48K session default playback is 064 buffer , the 96K is 128. I listed the actual buffer settings I managed to playback the sessions in the respective Nuendo versions and O.S.

Re the 64 Bit kernel, I hear ya , lets see where the dust settles.. :-)

[psvennevig]

Hi,

I have yet to see any difference with the Dawbench test between K32 and K64 in OSX.
With sessions that push the limit of the Cubendo 32-bit RAM usage (3,2GB) I can see (subjectively) better performance with K64.

Pål (or Pal)

[TAFKAT]

Hey Pal,

Sorry I missed your last post.

Interesting, I'll keep that in mind.

I am still going to update my OSx86 configuration, but considering what you have just noted, there is less urgency.

I have finished Parts II and III of the full reports for the DAWbench website, I just need to upload them which I'll do later today if I get a chance, and then put my helmet on as I am sure that there will be several punters on numerous forums that will be lining me up.

I get a distinct feeling I won't be on the X-Mas card list of a few devs after they read the reports... :-)

[Jcschild]

no worried Vin,
i am pretty sure you were already off their list :-)
looking forward to seeing those results

[efernan]

I get a distinct feeling I won't be on the X-Mas card list of a few devs after they read the reports... :-)

LOL

[TAFKAT]

Hey All,

New reports have been uploaded Here (http://www.dawbench.com/index.html)

Let me know what you think..

[Vinark]

Great piece of work! As expected, both the thoroughness and the results......
:emote_beerchug:
Vincent

[TAFKAT]

Great piece of work! As expected, both the thoroughness and the results......
:emote_beerchug:
Vincent

Thanks Mate.. :-)

For those wanting a quick scan without reading the reports, here are the graphs of the results for the multi app test sessions,

http://www.dawbench.com/images/dawbenchdsp-multi-md5-10-10.jpg

http://www.dawbench.com/images/dawbenchdsp-multi-csp-10-10.jpg

http://www.dawbench.com/images/dawbenchdsp-multi-emp-10-10.jpg

Links to the full reports at DAWbench :

Part I - XP/Win7 x64 : OSX 10.5.8/ OSX 10.6.2 : Cubendo (http://www.dawbench.com/win7-v-osx-1.htm)

Part II - Win7 x64/OSX 10.6.2 : Nuendo 4 v Nuendo 5 (http://www.dawbench.com/win7-v-osx-2.htm)

Part III - Win7 x64/OSX 10.6.2 : Nuendo 5 v Reaper 3.6 v StudioOne 1.5 (http://www.dawbench.com/win7-v-osx-3.htm)

[TAFKAT]

Hey All,

I have update the suite of benchmarks to include all the new sessions, anyone interested can download them now.

DAWbench DSP Universal -2011 Here
(http://www.dawbench.com/benchmarks.htm)

[OpenMind]

Did anybody tell you that you are doing great work Vin? :icon_cool:

[Animus]

yes! Great work Vin. I like it when you do our bitchwork for us.

:-)

[MattiasNYC]

Yes Vin, greatly appreciated!!!

[zephonic]

Interesting stuff. A bit unnerving for Mac users, like me, but not enough to sway me...yet.

I'll hold out for a little and see if the discontinuance of legacy support will level out the performance disparity in Cubendo Next.

If it does not, well...

[Daryl]

Interesting stuff. A bit unnerving for Mac users, like me, but not enough to sway me...yet.

I'll hold out for a little and see if the discontinuance of legacy support will level out the performance disparity in Cubendo Next.

If it does not, well...
I wouldn't worry overmuch about it. If you been happy using a Mac for the last few years, the report changes nothing. It only brings information out into the open for a change.

D

[TAFKAT]

I'll hold out for a little and see if the discontinuance of legacy support will level out the performance disparity in Cubendo Next.

Hey Z,

Discontinuance of what legacy support are you referring to ?

Also, this is not about swaying anyone , its about reporting how the chips fall , the factors involved and getting the devs to acknowledge and be honest about the variables.

As Daryl has already noted, its not like all of sudden its going to sway long term MAC users/Windows haters to switch... LOL

BTW: Protools LE / MP testing almost finished, which throws a few more curve balls into the mix due to the inconsistency between the 2 on the respective hardware, can't wait to ditch both of them and run Protools 9 on my RME reference hardware for some real head to head.

I have head to head with PTMP 8.04, Cubase 5.5 ( yes I have gone back to C5.5 for testing - Nuendo has been put out to pasture ) and StudioOne 1.6 , which will then be a good yardstick to measure PT9 against when it hits..

I'll upload the full worksheet with the results in a day or so .

Stay tuned.

[nikki-k]

VERY interested to find out what VI (or VI's) you will decide to use for a nice X-Plat shoot out.
Who will rule at a 64 buffer will be a pretty big deal IMO. Arturia, NI, Ivory, Superior2, etc are all "must use" VI's for me. PT is incredibly awful with al of those as of last use for me.

[TAFKAT]

Dedric is already using a whole stack of VI's in his current move across from Cubendo to PT and is not seeing any major shift in performance.

FWIW: For my DAWbench VI Universal benchmark, I'll be using NI Kontakt Free, when I get some time to finalise it.

I personally have a lot of your listed VI's, so I can test with a whole range if need be, it will just be difficult to develop a consistent quantifiable methodology tho.

[nikki-k]

I created a session for this purpose back a couple of years ago. I created an Instrument track in PT, and proceeded to record a dense bit of MIDI for 2 minutes. I then randomized a bit, offset, and pasted.. repeated this for a total of about 30 tracks if I remember correctly. I exported this as a .mid for use in other apps.
EDIT to add: while playing back, I always had a live midi track going since this was one of two circumstances requiring a 64 sample buffer (live audio being the other); with no non-realtime bounce, alternatives/workarounds were either a royal PITA or a flow breaker.

Obviously Logic has interesting things going on under the hood, but when comparing Reaper, Cubase and PT with Minimoog V instances, PT suffered horribly.
There have been issues with PT/RTAS since ver6. On top of that, low buffer use has been horrible. If someone is using W7 with PT/RTAS and non-sample based VI's that perform nearly identically at 64 sample buffers... I would LOVE to know the system specs. Everyone I know that has tested this has failed.

Funny thing is, even if I wanted to use PT9 now, I could not use any of my sessions. Superior 2 alone is eating 5-6G of memory.
Oh- and as for sample based stuff.. there are interesting things going on. I already write posts that are too lengthy, so I will not add to that with that info. Suffice it to say that Structure began a fun era.

[TAFKAT]

Hey Nikki,

Send me the .midi file please , I have NI Komplete here as awell as all of the Arturia VI's - all unopened mind you but thats another story.. LOL

With PT9 now using ASIO, it doesn't really matter what the performance was in the past, its a whole new ball game , re Superior 2 using 5-6 GB for your current drum kits, hmmmm, well until AVID sorted the extended memory address space then Cubendo will be your ride , but for many, I think Steini may be getting s rude surprise next update time.. !!

[kdm]

I created a session for this purpose back a couple of years ago. I created an Instrument track in PT, and proceeded to record a dense bit of MIDI for 2 minutes. I then randomized a bit, offset, and pasted.. repeated this for a total of about 30 tracks if I remember correctly.

Forgive me if I'm not following your test here nikki - you have one instance of Minimoog V, and 30 tracks of midi feeding either it, or just playing back unassigned?

Also, are you on win7 currently? I've seen a lot of Mac users reporting midi timing problems, but the few midi reports I've seen on the Windows DUC forums seem to be Win7.

I just loaded up 32 instances of Absynth 5 in PT 8, ran the same patch in the first 16, and another in the second, then triggered each of the two sets with different 16th note sequences, all quantized to test how well PT buffers and times simultaneous notes (traditionally the most difficult test for midi to pass).

I bounced two mixes and inverted one - they completely cancelled, which means that the timing is sample accurate on multiple passes. I'm starting to wonder if Minimoog V is your problem - not all VIs are necessarily sample accurate, or equally efficient in either RTAS or VST for that matter. Next variable would be WinXP (my system) vs. Win7 vs. OSX (I've seen several OSX users reporting midi timing problems).

Btw, I'm running the /3G switch on a 32bit system with 4G of ram and this project only has 1.1G free (that's about 32M per instance).

[nikki-k]

@Vin: Will need to hunt it down. I can be a bit disorganized at times, and it might have been one of the tragic victims of me thinking I copied something over before reformatting. Not really sure why ASIO would make any difference; it is my understanding that PT is still utilizing the DAE underpinning, which would suggest that ASIO/CA are simply "front-endiing" that train wreck. I think some of the distaste I have results from spending so much time investigating all that stuff. Waves does "it's thing" similarly, albeit on a far, far smaller scale. Please do not get me started on Altura and other atrocities...

@kdm: I created 30 Instrument tracks in PT, Minimoog V on each, each MIDI sequence slightly different, no quantization initially (I did use other VI's, such as NI stuff as well.. cannot comment on some aspects :( ). Back with PT6 and 7 on a single core single CPU 3G P4, 2 instances after a save and reload would crap out PT at 64 or 128 sample buffers. With a 1st gen 2.66G Mac Pro, I got 17 instantiated initially, but upon save and reload, it had to be reduced to 7, and that was on both OSX and XP. Logic allowed 27, and Reaper allowed 21 at a 128 buffer (pretty sure it was not a 64 buffer, would have to review old correspondences with Digi). I will give your run a shot as soon as I have some time (currently hyperfocused on some gaming environments I am designing.. nothing modern, lol)

[TAFKAT]

Hey All,

I was just rounding off a whole stack of cross platform testing with PTLE/MP 8.04 when AVID unleashed the new PT9 with instant download availability, which was perfect timing to dot the i's and cross the t's.

My main interest, as would be for many others not locked into the Digi/AVID hardware circle , is how well PT9 performed using open hardware. My reference hardware for all my non AVIDesign testing is RME HDSPe , so being able to use that hardware natively is the game changer.

O.K The results..

First off with the M-Audio Interface which was the hardware platform I was doing the M-Powered testing on.

http://www.dawbench.com/images/dawbench-1-10-2.jpg

Results and overall feel for PT9 and PT8MP on Win7 were identical, and I do mean identical, which makes me very suspicious what the actual driver protocol is being used , overall PT9 behaved exactly the same as M-Powered, except now we had ADC. On OSX, PT9 was measurably better than PTMP 8.x

As you can see from the above results, PT swept the floor on OSX and the comparative cross platform performance is a lot better than the other DAW's, except from that one weird anomaly with the MD5 @ 064 on OSX .. ??

The RME results..

http://www.dawbench.com/images/dawbenchdsp-11-10.jpg

Well just when I thought I was heading for a slam dunk..., check out the Win7 results on PT9 @ 128 and above on the Channel Strip Pro and Elysia mPressor , I swear I thought I had entered the Twilight Zone... LOL

The 032/064 results had me floored, easily way better than any of the other competing DAW's across the board , but @128 Buffer those 2 test sessions went decidedly askew, the 032/064 sessions would not load @ 128, it was ugly to be honest, meters pegged in the red, GUI crawling, I was like WTF. So I dropped back to 032 samples and reopened the saved sessions, and sure enough they both played fine, they were right on the edge as expected, but they played back perfectly. Moved to 064 samples, opened the 032 sessions , no problem, opened the 064 sessions, no problem. At that point I was seriously scratching my head what was going on @128.

Not exactly sure what to make of it yet, and I doubt I'll be getting any response out of AVID while they are in their honeymoon bliss during AES, but I'll stay on it and see if I can get some response over the following week(s)

Something definitely not right when you can get better performance @ 032 than you can @ 256.

Which leads to me to this,

Who BETA tested this with 3rd party hardware, and how did they miss this ?

The above behaviour does open up some bizarre scenarios where someone working at 064 samples for example, gets close to the edge and wants to move up the buffer settings to get some more headroom, will in fact have the negative effect, or even better yet, someone working at 128 buffer, will have far better performance by dropping to 032.. ?? !!

Something screwy there.

Early days yet , definitely a promising start, but some ASIO bugs to iron out.

Before anyone asks why I haven't posted this at DUC , easy , I would rather not have to deal with admin from the parent company controlling the information flow.., and it seems that this information has to be approved before it is allowed @ Gearslutz, I kid you not.

If they block this info @ G.S, it will be pretty clear who is pulling the strings there... LOL

[kdm]

Vin - did you exit ProTools between buffer setting changes? In my experience, it seems something is not reset if you simply close one session and open another. Resetting the buffer latency within a session can cure it, but at times, quitting PT and starting fresh was the only/best solution.

[TAFKAT]

Vin - did you exit ProTools between buffer setting changes? In my experience, it seems something is not reset if you simply close one session and open another. Resetting the buffer latency within a session can cure it, but at times, quitting PT and starting fresh was the only/best solution.

Hey Dedric,

Depends on the hardware, with the AVIDesign / M-Audio hardware I didn't usually exit out at buffer changes, I closed the session , let the resources settle, and then opened the next session.

With RME ASIO in PT9 you have no choice but to exit when changing buffer settings , it has to be set within the RME CP, it can't be done with PT running- so while troubleshooting the above I was certainly exiting right out and then launching fresh before testing these sessions, and it didn't make any difference.

[LEX]

So it looks like RTAS is finally up to par if not more with VST. That's excellent. Even more confident now about moving, not that I wasn't.

Installing PT 9 now with CPTK2. This is probably going to be the first time ever, that a software has been released and I've bought, installed and started using on mission critical work.

LEX

[True North]

I have been running PT9 on my RME HDSP Madi card and it isn't playing nicely with my UAD 2. I am getting weird pops and clicks, I haven't had time to try using it with normal RTAS plugs. I just bought the crossgrade rigtht away just so I could could GAWK at the sight of PT working through my RME Card....truly a strange, yet glorious, sight I didn't think I would ever see.

UAD released a statement that was a little confusing. Not sure if they were finally creating RTAS plugs or if they are going to work closely with Avid to make sure the UAD plays nice with their DAW......finally....

Very interesting benchmarks you have there Vin. It's completely BIZZARRO that it gets worse above 128 samples. I will try to run some tests on my rig to see what happens. Sorry to ask, but where can I find the test files to download.

[LEX]

I have been running PT9 on my RME HDSP Madi card and it isn't playing nicely with my UAD 2. I am getting weird pops and clicks, I haven't had time to try using it with normal RTAS plugs. I just bought the crossgrade rigtht away just so I could could GAWK at the sight of PT working through my RME Card....truly a strange, yet glorious, sight I didn't think I would ever see.

UAD released a statement that was a little confusing. Not sure if they were finally creating RTAS plugs or if they are going to work closely with Avid to make sure the UAD plays nice with their DAW......finally....

Very interesting benchmarks you have there Vin. It's completely BIZZARRO that it gets worse above 128 samples. I will try to run some tests on my rig to see what happens. Sorry to ask, but where can I find the test files to download.

Windows I assume. Might be a conflict. Are you FX Expansion VST into PT?

UAD is making RTAS version plugins for PT that run on their DSP cards.

LEX

[TAFKAT]

I have been running PT9 on my RME HDSP Madi card and it isn't playing nicely with my UAD 2. I am getting weird pops and clicks, I haven't had time to try using it with normal RTAS plugs. I just bought the crossgrade rigtht away just so I could could GAWK at the sight of PT working through my RME Card....truly a strange, yet glorious, sight I didn't think I would ever see.

UAD released a statement that was a little confusing. Not sure if they were finally creating RTAS plugs or if they are going to work closely with Avid to make sure the UAD plays nice with their DAW......finally....

Very interesting benchmarks you have there Vin. It's completely BIZZARRO that it gets worse above 128 samples. I will try to run some tests on my rig to see what happens. Sorry to ask, but where can I find the test files to download.

Hey T,

You can download all the new test sessions on my DAWbench web page Here (http://www.dawbench.com/benchmarks.htm)

You'll have to grab the 3rd party plugins from the respective devs websites.

Re the UAD / RTAS announcement, its all a little Diagonalese , but I believe its that they are making RTAS versions of the plugins to run on their DSP, instead of using the wrapper.

Either way, I am not convinced its going to play nice.. , that would be a final nail for TDM.

[True North]

Windows I assume. Might be a conflict. Are you FX Expansion VST into PT?

Yes, it's Windows 7, 64 bit. I had to use FX Expansion to get the UAD's working, I am not using them on any of my other VST plugs.


UAD is making RTAS version plugins for PT that run on their DSP cards.

That is what I thought the statement said. There was some debate over this point over at GS and I didn't have the time to fully read bruhaha that ensued. That is great news to see they will be preparing RTAS plugs.......I hope they don't come at an added price, but I have suspicion that they will.

[LEX]

Yes, it's Windows 7, 64 bit. I had to use FX Expansion to get the UAD's working, I am not using them on any of my other VST plugs.



That is what I thought the statement said. There was some debate over this point over at GS and I didn't have the time to fully read bruhaha that ensued. That is great news to see they will be preparing RTAS plugs.......I hope they don't come at an added price, but I have suspicion that they will.

UAD tried the added price thing with UAD2, and went back on it. Returned the money and said they wouldn't do it again.

Owners of UAD will get RTAS versions when they update the software, just like they always do.

LEX

[True North]

Owners of UAD will get RTAS versions when they update the software, just like they always do.

From your lips to gods ears :)

I have small fortune 'invested' in UAD plugs, so I really hope so

[LEX]

From your lips to gods ears :)

I have small fortune 'invested' in UAD plugs, so I really hope so

They ticked alot of people off with that "upgrade" BS before. They know people are moving to PT. If they did that, they'd lose people.

LEX

[Animus]

Very interesting benches Vin.

Yeah, Gearslutz have become very "corporate" of the last year or so. sluts indeed.

[nikki-k]

posted in the other thread, but will say again... interesting, but I expected interesting. VI's are what I am curious about, and since I have the capability now, I will try several myself. Best feeling PT in years IMO.

[TAFKAT]

Quick heads up,

I ran some test on an RME Fireface 800 today to see if the issue I experienced on the RME HDSPe card was reproducible on the Firewire units as well. The lowest acceptable buffer for the Fireface 800 on Win7 is 064 , as PT9 will not accept the 048 buffer setting. I ran up and fully loaded one of the test sessions that was displaying the issue previously, using the URS Channel Strip Pro , and sure enough , the 064 session would not load @128 or 256. The variable isn't as big as the HDSPe, 133 CSP's at 064, 128 CSP's @ 128 and again 128 CSP's @256 - it did not scale at all from 128 to 256 this time.

I am in contact with AVID and I'll keep you guys up to date if/when anything eventuates.

[TAFKAT]

Well my patience has run out at G.S,

Its obvious that the moderators have not approved the info for PT9 in its separate thread in the PT9 sub forum, despite the absolute garbage that they have allowed to be posted in that sub forum, defies logic.

Anyhow,

Info has now been posted in my other running thread Here (http://www.gearslutz.com/board/music-computers/545208-cross-platform-daw-multiprocessor-x-scaling-performance.html)

I haven't decided if/when I'll post this at DUC , the whole vibe over there reminds me of C.Net , and I'm not sure if I really want to enter into a debate there.

[kdm]

Well my patience has run out at G.S,

Its obvious that the moderators have not approved the info for PT9 in its separate thread in the PT9 sub forum, despite the absolute garbage that they have allowed to be posted in that sub forum, defies logic.

Anyhow,

Info has now been posted in my other running thread Here (http://www.gearslutz.com/board/music-computers/545208-cross-platform-daw-multiprocessor-x-scaling-performance.html)

I haven't decided if/when I'll post this at DUC , the whole vibe over there reminds me of C.Net , and I'm not sure if I really want to enter into a debate there.

Hey Vin - don't blame you on GS, or reluctance on DUC, but I'll be happy to link to it on DUC and other forums if you would prefer/approve, and would rather not debate it. Imho, it really makes RTAS look quite a bit better than most people once believed, so I doubt there will be much opposition.

[TAFKAT]

Hey Dedric

I don't mind you linking it at DUC and or anywhere else, please do, I was refering to not wanting to enter a debate at the Lions Den DUC for the time being, as they will simply see me as an outsider as I don't participate there at all, and I'm not sure of the reaction I'll get.

Yeh, I'm with you, this is mostly good news, I am not sure what the problem was @ G.S , but it will be interesting if they now try and ping it from my other thread. I have a feeling it will be too late by the time they realise its posted, the genie will be out of its bottle... LOL

[TAFKAT]

FWIW:

I have stepped into the Lions Den.. :-)

I posted my cross platform performance data over at DUC earlier this morning ( my time ) , and so far not even a peep from the membership there.

Lets see what if anything eventuates there.

@ Gearslutz the info is not getting any responses either, maybe I'm just making too much of it.. ?

[kdm]

Cool Vin. I linked to that thread over on VI.

[TAFKAT]

Cool

Thanks Mate :-)

I just noticed another thread asking about PT9 performance, and it already had numerous responses , while mine is being ignored.

I am getting a distinct feeling I am maybe swimming in the Fruitilian Sea.. LOL

[Daryl]

Cool

Thanks Mate :-)

I just noticed another thread asking about PT9 performance, and it already had numerous responses , while mine is being ignored.

I am getting a distinct feeling I am maybe swimming in the Fruitilian Sea.. LOL
The Fruitilian Sea. That's a great title for a children's book. Hmmmmmmmmmm......

D

[TAFKAT]

Oh , that reminds me..,

The Fruitilian Sea ©

:icon_lol:

[OpenMind]

Now I'm getting curious... I know what GS is, but what does DUC mean? :icon_lol:

[Daryl]

Oh , that reminds me..,

The Fruitilian Sea ©

:icon_lol:

Drat, too slow.

D

[kdm]

Cool

Thanks Mate :-)

I just noticed another thread asking about PT9 performance, and it already had numerous responses , while mine is being ignored.

I am getting a distinct feeling I am maybe swimming in the Fruitilian Sea.. LOL

It does seem to be a bit of a scrolling ticker over there - hard to track threads without a link or subscription, lol.

Replied to your report to bump it a bit, and keep the 128/256 anomaly in view as well.

[TAFKAT]

Now I'm getting curious... I know what GS is, but what does DUC mean? :icon_lol:

DUC- Digidesign User Conference (http://duc.avid.com/index.php)

@ Dedric,

Thanks Mate.

I expect some of the usual suspects from the Windows LE/MP camp who have a focus on scaling performance will chip in if/when they spot it.

[OpenMind]

Thanks! 8)

[nikki-k]

link(s) to thread? Cannot find it on the DUC.


FWIW:

I have stepped into the Lions Den.. :-)

I posted my cross platform performance data over at DUC earlier this morning ( my time ) , and so far not even a peep from the membership there.

Lets see what if anything eventuates there.

@ Gearslutz the info is not getting any responses either, maybe I'm just making too much of it.. ?

[TAFKAT]

Hey Nikki,

Thread is Here (http://duc.avid.com/showthread.php?t=286424)

Its seems that its an absolute non event for the usual suspects over at DUC., the only response I got from a member there was , have I tested at 512 and 1024, like WT.., I haven't set a buffer to 512 for 5 years, let alone 1024.., what am I missing ?

[TAFKAT]

Just a quick heads up,

I have uploaded the latest cross platform DAW performance reports, this time focusing on Protools / RTAS / DAE and of course ASIO/CoreAudio / RME with Protools 9.

Parts IV / V can be found Here (http://www.dawbench.com/win7-v-osx-4.htm)

Updated charts using the RME HDSPe now including PT9

http://www.dawbench.com/images/dawbenchdsp-multi-md5-11-10.jpg

http://www.dawbench.com/images/dawbenchdsp-multi-csp-11-r.jpg

http://www.dawbench.com/images/dawbenchdsp-multi-emp-11-r.jpg

[Animus]

Interesting. Protools actually performs less than Nuendo the lower the latencies.

[TAFKAT]

Interesting. Protools actually performs less than Nuendo the lower the latencies.

Do you mean better ?

Yes that was the wierd U turn @ 128 samples with the RME HDSP ASIO driver on Windows 7 I reported last week.

Very strange.

I'll keep you guys upto date if I hear anything from AVID about a fix.

[Animus]

Do you mean better ?

Yes that was the wierd U turn @ 128 samples with the RME HDSP ASIO driver on Windows 7 I reported last week.

Very strange.

I'll keep you guys upto date if I hear anything from AVID about a fix.

I meant Nuendo seems better at lower latencies.

[TAFKAT]

Where are you seeing that ?

Apart from the mPressor on Win7 , PT9 kicked Cubendo's arse at lower latencies !

On OSX PT9 performed better across the board except for that weird dip with the MD5 @ 064.

Oh BTW:

Time to poke the bear and post a thread with a more provocative title.

Protools 9 : ASIO : RME HDSPe : Windows 7 : Better Performance @ 032 than 256 : (http://www.gearslutz.com/board/music-computers/550742-protools-9-asio-rme-hdspe-win7-better-performance-032-than-256-a.html#post6009711)

Man, if that gets ignored, then I'll just throw my arms up... LOL !!

I have already wiped the idea of participating at DUC.. !!

[Tumppi Järnefelt]

Seems to be pretty quiet over there...

[TAFKAT]

Yeh.. , time to move on, seems no one really cares.

http://www.neweurasia.net/wp-content/uploads/2010/02/elephant_davidblackwell.jpg

Maybe I did make too much out of it.. :eusa_think:

[LEX]

I think the PT users don't care if its better or not. They aren't moving away from PT anyway.

LEX

[TAFKAT]

I think the PT users don't care if its better or not. They aren't moving away from PT anyway.

... and obviously more than happy to keep using their hardware dongles, despite it now being open in that department... LOL

I think one of the reasons this hasn't gained many legs with the PT users at G.S , or evn with the usual suspects over at DUC, some who also post regularly over at G.S, is that it won't effect them directly, as they still have their Mbox's, 002/003's, etc, still running DAE !

Either way, I did my bit bringing it to public attention , with the lack of interest I doubt AVID will give a shite about fixing it in any great hurry, seems the cool aide has kicked in over ADC .. LOL

I'll laugh my head off if this rears its head at a later date once more and more new punters hop on the PT9 ASIO train, by then I'll be...

:sleeping:

[Vinark]

2 things to keep in mind Vin. Indeed not everybody switched to RME right this week, but they might in the coming years and then they (and avid and maybe RME?) will care. And second not everybody is a nerd like me (and you). OMG the laws of physics have been broken...instead of going up 6 plugins it´s going down 6 when I go from 64 to 128. OMG what is happening here....... Pigs are flying, oh that was another company. I do care and am curious as hell (and we will probably not find out again). Maybe if performance would have been bad all over or low latency would have sucked maybe then....it just isn´t bad enough I guess
Thanks again and again and you get it
Vincent

[TAFKAT]

I hear ya that it isn't that big a variable on the surface, however, to me it hints at something deeper in the buffering that could rear its head in other areas if not addressed.

It will be funny if/when it starts hitting the wild in some RW scenarios, and then the shit hits the fan. Case in point will be someone on a laptop where the overhead can easily be tapped more so than say an i7.

Thing is AVID/RME may already be on it and may quietly sort it while its still under the radar.

Lets see where the dust settles.

[kdm]

I hear ya that it isn't that big a variable on the surface, however, to me it hints at something deeper in the buffering that could rear its head in other areas if not addressed.

I think it already has, but there just aren't enough of us out there making noise yet. As I've mentioned elsewhere, at 128 and lower, VEPro streams are out of sync with the buffer (or so it sounds to me). At higher latencies, it kind of works, but too many clicks/pops (and too high latency) be useful. A Logic user reported something similar with VEPro from PT8, but never heard any more about it from him.

Seems it could be that PT has a "sweet-spot" data size for streaming through buffers and 128/256 aren't it (or rather, certain block sizes might cause a log jam of sorts).

On performance in general, I think your tests made the biggest point of all - RTAS is just as good as VST (in some ways, I prefer it). On the deviations, I agree with Lex and Vinark - to some degree most people just are tweakheads like the rest of us. Contrast that to the frame/sync bug with N5 and PJPeg and that I've mentioned it 3x on Nuendo.com with only one reply. Apparently, like the lack of response on your performance tests, having accurate sync isn't a big concern when you use Nuendo for post. Maybe it's too geeky. lol

[LEX]

Apparently, like the lack of response on your performance tests, having accurate sync isn't a big concern when you use Nuendo for post. Maybe it's too geeky. lol

Kind of says something about the true number of people using Nuendo for Post.

LEX

[TAFKAT]

Re RTAS,

Well I have been reading in on some threads with PT9 / MAC and Virtual Instruments, and it its not exactly glowing, so I need to get a heads up on that with the DAWbench VI / Kontakt test that I have kept putting off, but I will now knuckle down and start configuring while I have a small window of downtime this week.

PT9 does have a sweet spot, and from what I can make out ist 064 bufefrs for some reason, also its very particular about buffer sizes, anything that is not a dirivative of 32 will not be accepted - i.e 048 samples. I have also read a post from a guy who has a Prism Orpheus (http://www.gearslutz.com/board/so-much-gear-so-little-time/549840-problems-setting-buffer-times-prism-orpheus-when-using-pro-tools-9-a.html) - which has probably the worst ASIO drivers I have ever tested - that has buffer sizes set by derivatives of 125 only, so PT9 is simply not accepting it at all.

Issues with PT9 / Apogee on OSX Here (http://www.gearslutz.com/board/music-computers/548362-pro-tools-9-apogee-compatibility.html)

This is just proving that PT9 BETA did not have the breadth or the width in regards to 3rd party hardware, I am betting the majority of those on the BETA list tested with AVID hardware, simply bacuse that is what they had been using up until they unlocked the software from the hardware. How many long term users of Protools would have high end ASIO/Core Audio hardware.

I think some of the PT9 honeymoon glow is starting to fade... LOL

[TAFKAT]

Kind of says something about the true number of people using Nuendo for Post.

LEX

LEX,

Focus please, this thread is not about Deadendo/Post... LOL !

[kdm]

Re RTAS,

Well I have been reading in on some threads with PT9 / MAC and Virtual Instruments, and it its not exactly glowing, so I need to get a heads up on that with the DAWbench VI / Kontakt test that I have kept putting off, but I will now knuckle down and start configuring while I have a small window of downtime this week.


Looking forward to these Vin. I'm assuming OSX vs. Win7? I'm finding some differences on XP between PT8 and Nuendo, but not in the same areas. PT8 flagged a "memory too low" error when setting up multiple Aux busses for up an Ocean Way Drums kit after having a couple of heavier Prominy guitars loaded in another Kontakt 4 instance (a good bit of memory in general - hard to load more than this with Nuendo). The session wouldn't save - no matter as it was a test and I knew I was pushing the limits of the /3G switch with both high memory guitars and a large drumkit.

Nuendo usually just starts corrupting/crashing when pushing it too hard. PT8 simply wouldn't allow a save, but played fine. Dropped back to the presave version without the auxes and it was fine. (Not a detailed baseline test though - I need to duplicate exact instrument loads between the two, but since I'll be moving to Win7/PT9 soon, it isn't high on my list).

Omnisphere loads also seem similar to Nuendo.

I think there may be some issues on OSX in general. All of the midi timing problems with PT I've seen are OSX. All of the instrument problems have been OSX. Maybe there aren't that many on Win to compare, so your test should help there some, but midi timing here is very good, perhaps slightly better than with Nuendo (at least with accuracy recording live input).

[TAFKAT]

Hey Dedric,

Just started to configure the new VI test with the NI Kontakt Free and have already hit a major brick wall in regards to available sounds.. :-(

Better to chat about this over at DAWbench Forum tho, chat to you there.. :-)

[TAFKAT]

O.K,

Managed to get some time to sort thru some of the rubble this morning.

I have a BETA version of this test ready for those who are game, Cubendo only so far - you will need a full copy of Kontakt 4 , no way around it unfortunately , the Kontakt Free Player is a total and utter waste of time as it will time bomb demo libraries I found , where the full version doesn't. I have shifted all of the instruments to the Kontakt standard sounds , mostly the VSL orchestral in this session and some synths /piano/bass , etc.

I will warn you guys this is brutal, I am able to belly up my i7 system quite easily with this test , so you'll need to some horsepower under the bonnet if you want to play

Let me know if you want to have a BETA run , PM your email and I'll email it across . File is about 4.5 MB

[Sound Drifter]

Hey Vin,

PT just arrived here. I have an RME card and a lynx FW card in my Aurora's. Would like me to run your benches with my rig and send you the results using the different sound cards?

I also have an mbox too if you want a bench with that as well.

[TAFKAT]

Yes Please , that would be great.

Not too concerned with the Mbox as that is running under DAE, but I'll be really interested in your RME and Lynx results under ASIO.

You can download the current DSP ( Plugin) based tests at the DAWbench website Here (http://www.dawbench.com/)

The Virtual Instrument test is well on its way, as long as you have Kontakt 4.

Thanks Mate.. :-)

[Sound Drifter]

Ok. I do have Kontakt 4. So when it is ready, let me know and I can PM you my email, or you can PM me a link to download and do it.

I am shooting next week to install PT and do the benches once I get through this weekend's work. So look for early to mid next week for my results.

[Sound Drifter]

One quick question since I have read of oddities with Win7 and FW.

I should have windows use a legacy driver for the FW correct?

Not really sure about this since I only use the Lynx FW to connect to my Macbook when doing location recording. Is there a procedure to do this easily if needed?

[TAFKAT]

Hey S,

I always shift it to legacy on the systems, I seem to get better low latency performance on most of the interfaces I have tested so far.

In short, it won't hurt, and it may just improve things.

Re The DAWbench VI BETA, PM me now, I'll forward the download link and instructions via email...

[Sound Drifter]

Just installed and configured PT last night. All except the Lynx Firewire driver, so I can start with the RME HDSP aes-32. Downloding the cubendo and PT benches right now. Hopefully load them up tonight after I finish mixing for the day.

Check your PM. Sent you my email.

[TAFKAT]

Cool,

Will be interesting how the tests pan out at your end.

For anyone following along and waiting for some snippets on the VI testing.., hmmm, lets just say the afterglow after the PT9 launch and good performance with RTAS DSP plugins is about to take a screaming turn south... :eusa_whistle:

Stay Tuned - or login at DAWbench to read in on the fun... :-)

[TAFKAT]

Sneak Peak..


http://www.dawbench.com/images/dawbenchvi-b-11-10-2.jpg

Numbers represent the added notes of polyphony above the session template which is already playing between 50-70. There are 2 versions of the test :

CV - ConvoVerb on VSL string parts of the arrangement

NCV - No ConvoVerb on VSL string parts of the arrangement .

The intial testing with the M-Audio was to eliminate the ASIO/CoreAudio variable out of the equation , the RME results will follow in a few days.

I think AVID are about to get a nasty wake up call.. :eusa_whistle:

[Sam]

Ouch!

[leggy]

Wow! :)

[dementedchord]

that hurts... pass the dilaudids...