OC'd CPU not giving me better performance - this normal?

[RiffWraith]

I have an ASUS P5Q Pro with an Intel 3.0 GHZ Core 2 Duo E-8400 chip.

I have done some overclocking, tho I did not go crazy, so I am pretty sure what I have done is not extreme.

I created a test project with a bunch of VSTis playing simul, getting to the point where there are a small amount of dropouts. They happen consistantly in the same spots, and to the same degree each and every time I play the transport.

Thing is this: There is no difference whatsoever in performance, and the amount of dropouts when the OC is on, as opposed to when I turn it off. Is this normal? Shouldn't I get at least a little better performance when OC'd?

[paulwr]

are you choking at the cpu in each case? Sure of that? I'd run without the overclock if that is the case.

-Paul

[Animus]

I have an ASUS P5Q Pro with an Intel 3.0 GHZ Core 2 Duo E-8400 chip.

I have done some overclocking, tho I did not go crazy, so I am pretty sure what I have done is not extreme.

I created a test project with a bunch of VSTis playing simul, getting to the point where there are a small amount of dropouts. They happen consistantly in the same spots, and to the same degree each and every time I play the transport.

Thing is this: There is no difference whatsoever in performance, and the amount of dropouts when the OC is on, as opposed to when I turn it off. Is this normal? Shouldn't I get at least a little better performance when OC'd?


That's not normal. Be sure you are not overheating. I get a noticeable speed bump when overclocking with no ill-effects.

[Michael Tibes]

That's not normal. Be sure you are not overheating. I get a noticeable speed bump when overclocking with no ill-effects.

So do I. If you're not overheating I wonder if you've hit another bottleneck than the cpu performance. Or the overclocking is too marginal?

Michael

[The Guru]

I have done some overclocking, tho I did not go crazy, so I am pretty sure what I have done is not extreme.

You don't give technical data, so how can you ask for technical advice? :eusa_eh:

Be precise, squirrelboy. :icon_rolleyes:

[RiffWraith]

are you choking at the cpu in each case? Sure of that? I'd run without the overclock if that is the case.

-Paul

Yes - I am choking the CPU. Point is, if I am barely choking it without the OC, shouldn't it not choke with the OC?


That's not normal. Be sure you are not overheating.

Not overheating -

Temperatures
Motherboard 30 °C (86 °F)
CPU 35 °C (95 °F)
CPU #1 / Core #1 45 °C (113 °F)
CPU #1 / Core #2 45 °C (113 °F)
GPU Diode 65 °C (149 °F)

Cooling Fans
CPU 2722 RPM
Chassis #1 1288 RPM


So do I. If you're not overheating I wonder if you've hit another bottleneck than the cpu performance. Or the overclocking is too marginal?

Michael

I might have in fact hit another bottleneck - but then the Q is where? Or maybe the OCing is in fact so marginal, that it wouldn't matter...


You don't give technical data, so how can you ask for technical advice? :eusa_eh:


Set FSB Freq to 400, RAM Freq to 1066 (my RAM is 1066 MHz) and the CPU voltage to 1.30. That's about it. This, according to CPU-Z, brought my:

Core clock from 2000 MHz to 2400 MHz
Bus speed from 333.3 MHz to 400 MHz
Rated FSB from 1333 MHz to 1600 MHz
DRAM Freq from 400 MHz to 533.3 MHz

In addition, the BIOS is now reporting that my CPU speed went from 3.00 GHz to 3.62 GHz.

I was figuring that even tho I didn't go nuts, those little changes would make at least a little difference...maybe not?

[LEX]

Well, I noticed with certain MOBO's and amount of RAM back in the AMD days that OC varied.

On my ASUS board, I got a slight performance boost but stablity was shakey. Probably because of the RAM speed and the mulitplier. Unpredictable.

ON the Gigabyte board, I got a bit better performance and the stability was better and more consistent but at times would cause some problems.

I have since, stopped OC'ing. The boost wasn't worth it in the long run.

LEX

[The Guru]

What's the ASIO meter reading? The choke could be at the interface.

Another thing is the dual core. For what you do you should have opted for a quad, if not one of the i7's that last upgrade. Considering your investments in other areas, it seems a no-brainer.

[Vinark]

Shouldn't the core clock also read 3.6ghz? Maybe you have some power saving on in either windows or the bios.
Here it is very linear, if I overclock 25% I get 25% better performance.

[Michael Tibes]

I have taken my I7 920 from 2.7 to 3.2 if I remember right, and this has given me a noticeable performance boost in the top range. It has given me the extra power I neeeded on pretty large mixing sessions. It seems like a similiar percentage like what you did, so it should so something in your case. Did you run benchmarks (like sandra) with and without the overclocking in order to verify the result? It should make a clear difference.

Michael

[RiffWraith]

What's the ASIO meter reading? The choke could be at the interface.


The ASIO meter is WAY up there, touching the red at times. The choke could be at the interface - little more detail on that, please?


Shouldn't the core clock also read 3.6ghz?

I was thinking that too - I don't know. I am no OC expert; I am just reporting what I am seeing.

There is no power saving anything on in either windows or the bios.

Thanks.

[The Guru]

The bottlenack could be occuring at the conversion stage therefore making CPU improvement undetectable.

Does that board have Speed Step?

[TAFKAT]

Riff,

If the ASIO meter is peaked, overclocking the CPU is not going to be of any benefit as the headrooms is already tapped. I have seen instances where the ASIO is peaked and there is still 40% CPU overhead available.

What you are reporting the bottleneck may well be at the interface end.

Whats the audio interface and at what latency are you usually running the system ?

RE benches, run up DAWbench DSP , and do some quick comparisons at various latency settings both standard and overclocked, it will tell you pretty quickly whether you are getting some scalability from the overclock.

I agree with the other guys here who are overclocking, you should be getting a pretty linear increase.

[RiffWraith]

Thanks for the suggestion, Vin.

I run 2 M-Audio 24/96 cards with the 64-bit drivers in Vista Ult. 64, both set to 768 samples.

I did the test, and have very curious results with OC off:

512 samples - audio started to break up at row 2 ch 24, and got really bad by row 2 ch 30

768 samples - audio started to break up at row 2 ch 27, and got really bad by row 2 ch 30

1024 samples - audio started to break up at row 2 ch 23, and got really bad by row 2 ch 27

Is it me, or is that odd? I expected 1024 to give the best perf. throughout - but it didn't really. Why was 512 better than 768 - sort of? Anyway, I got noticeably better perf with the OC on:

512 samples - audio started to break up at row 2 ch 37, and got really bad by row 3 ch 3

768 samples - audio started to break up at row 2 ch 39, and got really bad by row 3 ch 5

1024 samples - audio started to break up at row 3 ch 3, and got really bad by row 3 ch 8

I followed the .txt; rebooting after each latency adjust, and running each test 2x.

The one thing I did not get was this: "At that (breakup) point solo the Monitor Channel, raise the fader double check for any breakup."

Did that, and didn't hear anything - even when the audio got really bad. Was I supposed to hear anything?

Cheers.

[The Guru]

So, OC is indeed working and something is definately screwy with your results judging by any of Vin's tests I've seen. There should be marked improvement at each latency.

Are there different versions of the M-Audio drivers that you could try, either newer or older?

[Vinark]

I still think the 2.4 ghz is wrong. check the bios for both C1E support and speedstep in the cpu configuration menu. Both should be disabled for now.

[RiffWraith]

So, OC is indeed working and something is definately screwy with your results judging by any of Vin's tests I've seen. There should be marked improvement at each latency.

Are there different versions of the M-Audio drivers that you could try, either newer or older?


Yeah - screwey, I agree. I thought the results would be lateral with the OC off as with it on. But anyway....


Turned off C1E support and speedstep in the cpu configuration menu - good suggestion, thanks. Now my core clock reads 3000MHz and 3600MHz OC off/on. So I redid Vin's test and got the same results with the OC on as before, but now, the results are more lateral with the OC off:

512 samples - audio started to break up at row 2 ch 24, and got really bad by row 2 ch 30

768 samples - audio started to break up at row 2 ch 27, and got really bad by row 2 ch 33

1024 samples - audio started to break up at row 2 ch 29, and got really bad by row 2 ch 36

OC on (same as before):

512 samples - audio started to break up at row 2 ch 37, and got really bad by row 3 ch 3

768 samples - audio started to break up at row 2 ch 39, and got really bad by row 3 ch 5

1024 samples - audio started to break up at row 3 ch 3, and got really bad by row 3 ch 8


So, there is a definite difference with OC off, as opposed to on - at least this way. But here's the thing: I ran my test project again, and I am still getting the exact same performance with OC off, as with it on. So, performance is better with OC on in Vin's test proj, but it is the same in a real-world test project triggering VSTs. Any ideas?

Thanks again :)

[The Guru]

As stated, that means the bottleneck isn't at the CPU, but most likely at the DA conversion.

Also, as stated, try to roll back the interface driver if you have the latest. Could be M-Audio's 64 bit drivers need work. It could even be MB drivers or chipset drivers or such that are the culprit.

The ASIO meter's still pegging?

[RiffWraith]

As stated, that means the bottleneck isn't at the CPU, but most likely at the DA conversion.



Oh, does it mean that? Oh.

I do not have a newer, nor an older driver version. This is it. The older version was a beta version, which didn't work properly anyway.


Yes, the ASIO meter is still pegging - this is one thing I forgot to mention...oops...thowwy....:smash:....In Vin's test .cpr, the TM CPU usage went up with the ASIO meter; as the ASIO meter was pegged, so was the TM CPU meter. But in my test project, when the ASIO meter is pegged, I still have about 40% of CPU left. That say anything?

[Vinark]

Hey Riff
what I do regarding C1E and speedstep is I disabled C1E but keep speedstep on. Speedstep can be switched on from the power profiles. That way in low load sitations the cpu stays much cooler and if the auto switching ever gives problems it's easy to switch of from the os.

Regarding the problem project, could be so many things, a badly written plug, a short period where voices overlap, HD chocking if it is sample playing etc. You could post the project if you like. Does much higher latency help?

[RiffWraith]

Regarding the problem project, could be so many things, a badly written plug, a short period where voices overlap, HD chocking if it is sample playing etc. You could post the project if you like. Does much higher latency help?

Well, there are many voices that overlap, that's for sure. Bad plug? Possible, but the only VSTi or plug is Kontakt. I could post the project, sure - but unless you have my custom multis loaded, you won't be triggering exactly what I am triggering. Higher latency does help a bit, yes.

Cheers.

[Vinark]

If it's all Kontakt it could be the Hard disk chocking from streaming samples, that would explain the OC not having an effect. Don't know how to test for that or maybe set DFD differently.

[Jcschild]

if you are talking about sample sets (virtual inst)
then memory (BANDWIDTH) is always the bottle neck
while OCing should also increase the memory bandwidth if not set correctly it wont.

if you run Vins RXC test you should see more effects OCed vs not.

samples testing is harder due to memory contraints both in lack of large addess aware and lack of memory bandwidth period.

on my website is a samples test the core 2 duo OCed to 4GHz @ 1900 FSB was a big jump in samples ability but i was also using DDR2 1066 ram running @ 950

anyone doing large samples sets should be on a core i7.

Scott
ADK