Overclock has slowed my CPU down?

Associate
Joined
12 Feb 2011
Posts
54
Okay so I've just got into using the Intel Burn Test for stability testing and for a control test I returned everything to stock in my bios and ran IBT with stress level on maximum. I plan to change settings and continue to run the exact same IBT, so I have some numbers to work off of. Anyway, the stock settings test came back perfectly stable and resulted in a consistent 36 Gflops. I then went into bios and set the ASUS auto-OC thingy to do a default 10% increase, leaving everything else on auto - I just wanted to see the results as a comparison to stock. Basically in this setting it puts the FSB up by 10%, so it went from 3GHz to 3.3GHz, and because everything else is left on 'auto', the motherboard decided to increase the volts to a stupid 1.5~! The RAM is overclocked slightly to 736 from 667 (timings stay the same), and the NB Frequency + HT Link are moved up 10% to 2200MHz - I assume the NB voltage and RAM voltages are adjusted as well? But I can't be sure as CPU-Z doesn't tell me those numbers. Either way, it doesn't matter what they've all been changed to because after running IBT on maximum stress the test finished with perfect stability - good news, right? Well not entirely, because for some reason the Gflops were consistently at 24 now?!? I thought this seemed very odd because doesn't that just mean that the CPU is massively slower now, even though it's overclocked by 10%...??

Am I interpreting this wrong, or has the OC actually made everything slower? Again, it was just a default 10% OC, where the motherboard picked all of the settings - everything runs consistently and fully stable, but apparently the CPU is 12 Gflops slower! What gives?

Can anyone shine any light on the matter for me? Or some insight perhaps? Anything really that might explain this result! Thanks!
 
Last edited:
Associate
Joined
4 Dec 2008
Posts
1,812
Fairly sure that's just an unstable OC that's managed to pass testing. Whenever my OCs have been unstable the GFlops have lowered.

Either that or the HT Link/NB Link/RAM Timings hasn't been right.
 
Associate
OP
Joined
12 Feb 2011
Posts
54
Fairly sure that's just an unstable OC that's managed to pass testing. Whenever my OCs have been unstable the GFlops have lowered.

Either that or the HT Link/NB Link/RAM Timings hasn't been right.

Righto, I'm new to this and just assumed it was stable because it passed the IBT and the Gflops, albeit very low, were at least consistent - confusing!

The RAM timings were exactly as they should be, according to the Corsair website, and they were the same as they are on stock settings, the only alteration being the increased speed (667MHz to 734MHz). Could that cause this sort of a problem, if the RAM just can't run that fast?
 
Last edited:
Associate
Joined
4 Dec 2010
Posts
2,290
Location
London
No. As spleenharvester said, it's probably a bad overclock. Try doing everything manually - the ASUS auto OC tool sometimes doesn't lock in settings and you can find your voltages jumping all over the place.
There's plenty of manual OC guides available, just take it slowly.
 
Associate
OP
Joined
12 Feb 2011
Posts
54
Okay, thanks everyone - I've gone to manual and put the Vcore at 1.4375v for this clock, things seem stable so far. It makes sense that the ASUS auto voltage was fluctuating and screwing things up.

So I've been taking it slow, like you said, and moving the FSB and Vcore up and down by an increment at a time to see what results I could get stably. One thing I've noticed however, is that when performing the IBT on maximum stress level, whereby the RAM is pushed to as near maximum usage as it can take, while keeping the OS running - in my case that's 92% - I've had the IBT fail, but then pass flawlessly on standard and high stress, whereby the RAM usage is at around 45% for standard and 70% for high. So does this mean that my system would be stable as long as the RAM usage only goes up to 70% or so, but would become unstable at nearer 100% usage?

Because no games ever take my RAM usage above 60%-ish, which would mean I could keep this overclock for everyday use, for example, but obviously it wouldn't be stable in a benchmark..? Is my understanding of this correct? Or not? Seems to make sense to me at least :)
 
Soldato
Joined
1 Jun 2010
Posts
7,053
Location
London
What is your cpu? I presume it is AMD as you inquired about IBT compatibility with AMD cpus yesterday.

Open the windows task manager, have the overclocked cpu run on maximum stress level, note down free and cached memory and tell us if cached memory goes down significantly?
 
Associate
OP
Joined
12 Feb 2011
Posts
54
What is your cpu? I presume it is AMD as you inquired about IBT compatibility with AMD cpus yesterday.

Open the windows task manager, have the overclocked cpu run on maximum stress level, note down free and cached memory and tell us if cached memory goes down significantly?

Hi, yes it's an AMD Athlon II X4 640. Memory usage in task manager when idle was:

Free 2361 MB, cached 596 MB

Then I ran IBT on maximum stress and it went to:

Free 0 MB, cached 424 MB

What does that tell you?

(I have 4 GB total)
 
Last edited:
Soldato
Joined
1 Jun 2010
Posts
7,053
Location
London
Hi, yes it's an AMD Athlon II X4 640. Memory usage in task manager when idle was:

Free 2361 MB, cached 596 MB

Then I ran IBT on maximum stress and it went to:

Free 0 MB, cached 424 MB

What does that tell you?

(I have 4 GB total)

I think it means Windows only further freed about 172MB of physical memory from cached data after the free memory got all used up during IBT testing, while it preserved 424MB of physical memory for cached data. Theoretically cached memory should have gone down to almost zero.

But since some cached data was preserved, where do you think 424MB of memory came from as rest was taken up by windows, background processes and device drivers?;)

However maximum stress level = available ram

Available ram is combination of free ram and cached ram (storing cached data on the physical ram). So when you run IBT on maximum stress level, free memory gets used up and even cached memory is reduced (strictly speaking plenty of cached data is terminated by windows in order to free up more memory for IBT usage).
However not all cached data/instructions is terminated by windows as those may be needed by other processes, programs etc. So in this case virtual memory can be assigned by windows to compensate for the cached memory which couldn't be freed. Since virtual memory is slower, it can affect cpu processing speed and in this case you could end up with lower GFlops values which in turn means that your cpu temps will be lower.


Edit: Better to test on custom stress level and input 'Free memory' to ensure consistent GFlops values which will give indication of how fast your cpu is processing/executing those equations.

You may find that you have to run IBT several times while playing around with memory setting to ensure your GFlops is consistent. Best method is as mentioned above; using custom stress level and inputting only 'Free' memory as this means that virtual memory will play no part since you will only be using 'Free physical memory'.

I agree even this setting isn't foolproof as stated by respectable member C64 but it is the most accurate way of getting consistent GFlops.

I hope you read the links I posted in your IBT query thread.:)
 
Last edited:
Associate
OP
Joined
12 Feb 2011
Posts
54
I did indeed read those links, WingZero, very informative - thanks :)

I have no idea how to input 'free memory' into the IBT, the only options I have are: standard, high, very high, maximum and custom; and custom seems to want me to put a number in, so...?

Also, my RAM is running at 8-8-8-20 now instead of 9-9-9-24, but I have no idea if that is going to affect stability?

And one final point, I promise, my HT Link and NB Frequency have been increased to 2350MHz automatically to match the FSB increase to 235, again, I don't know if this will affect stability..? Do I need to put some more volts through the VDDNB to compensate?

I know my knowledge of all this must appear completely lacking, and yes, it is! That's why I'm trying to learn all the what's and whys!

Thanks again everyone.
 
Soldato
Joined
1 Jun 2010
Posts
7,053
Location
London
I did indeed read those links, WingZero, very informative - thanks :)

I have no idea how to input 'free memory' into the IBT, the only options I have are: standard, high, very high, maximum and custom; and custom seems to want me to put a number in, so...?

Also, my RAM is running at 8-8-8-20 now instead of 9-9-9-24, but I have no idea if that is going to affect stability?

And one final point, I promise, my HT Link and NB Frequency have been increased to 2350MHz automatically to match the FSB increase to 235, again, I don't know if this will affect stability..? Do I need to put some more volts through the VDDNB to compensate?

I know my knowledge of all this must appear completely lacking, and yes, it is! That's why I'm trying to learn all the what's and whys!

Thanks again everyone.

Choose custom and type in the 'Free memory' value from the windows task manager. Actually slightly less for more accuracy. So if it says 2450MB in Free memory, type in 2400Mb or 2350MB for more accuracy.
Also manually select 4threads in IBT since you have a quadcore.

As your RAM timings have tightened, it could affect stability though it shouldn't have big impact on GFlops values. Did you up the DRAM/DIM Voltage It is always advisable when overclocking cpu to have the ram at stock settings to take it out of the instability factor. Once you have overclocked cpu succesfully and stabilised it, only then you should begin overclocking your ram and running like memtest and prime blend test to check it's stability.


I am afraid only AMD users can accurately answer your 3rd paragraph but since those values have increased automatically, they should be a helping hand in stabilising your system.

Did you set PCIE frequency to 100MHz in bios? If you didn't then do so to keep it a constant.

Edit: Turn off any power saving options you may have as I have found from my experience that they tend to constrain cpu from using it's maximum potential when running a stress testing program.

Don't worry everyday I am learning new things:).
 
Last edited:
Associate
OP
Joined
12 Feb 2011
Posts
54
Choose custom and type in the 'Free memory' value from the windows task manager. Actually slightly less for more accuracy. So if it says 2450MB in Free memory, type in 2400Mb or 2350MB for more accuracy.
Also manually select 4threads in IBT since you have a quadcore.

As your RAM timings have tightened, it could affect stability though it shouldn't have big impact on GFlops values. Did you up the DRAM/DIM Voltage It is always advisable when overclocking cpu to have the ram at stock settings to take it out of the instability factor. Once you have overclocked cpu succesfully and stabilised it, only then you should begin overclocking your ram and running like memtest and prime blend test to check it's stability.


I am afraid only AMD users can accurately answer your 3rd paragraph but since those values have increased automatically, they should be a helping hand in stabilising your system.

Did you set PCIE frequency to 100MHz in bios? If you didn't then do so to keep it a constant.

Edit: Turn off any power saving options you may have as I have found from my experience that they tend to constrain cpu from using it's maximum potential when running a stress testing program.

Don't worry everyday I am learning new things:).

I would love to leave the RAM at stock clocks, but I don't know how to! As soon as I up the CPU FSB the RAM changes due to being on auto, and if I manually change the RAM to its stock values, then it shoots well over them when OCing the CPU. So I've tried lowering the stock MHz of the RAM to try and compensate for this, but then the RAM is slightly underclocked on the MHz side of things, but keeping the stock timings. Is RAM affected by having a lower clock speed, but its stock timings? Also, will a lower clock speed have any detrimental effect on it, aside from the obviously lower performance?

But thanks for the notes, I'll go into bios now and put the PCIE at 100MHz.
 
Soldato
Joined
1 Jun 2010
Posts
7,053
Location
London
I would love to leave the RAM at stock clocks, but I don't know how to! As soon as I up the CPU FSB the RAM changes due to being on auto, and if I manually change the RAM to its stock values, then it shoots well over them when OCing the CPU. So I've tried lowering the stock MHz of the RAM to try and compensate for this, but then the RAM is slightly underclocked on the MHz side of things, but keeping the stock timings. Is RAM affected by having a lower clock speed, but its stock timings? Also, will a lower clock speed have any detrimental effect on it, aside from the obviously lower performance?

But thanks for the notes, I'll go into bios now and put the PCIE at 100MHz.


I think your Ram speed is linked to your FSB speed just like in the case of my core 2 quad q6600. So when you increase FSB, Ram speed increases automatically. Do you have memory dividers in bios? They go by different names such as system memory mulitplier etc but they are used as a ratio between FSB and RAM speed hence they can help in keeping your RAM at stock speed.

However ideally your RAM speed should be equal to FSB or higher when running IBT as it involves a lot of RAM. This means the data from Ram will always be available to cpu for processing and that it doesn't have to wait for ram to pass on the data which could result in cpu slowing down hence affecting GFlops values. This would happen when Ram speed is lower than FSB speed.
So in this case you are fine with slightly overclocked RAM.

As for RAM speed vs timings, opinions are divided on this issue but usually tighter timings are preferred over ram speed. Have a read through this article:

http://www.tomshardware.com/reviews/tight-timings-high-clock-frequencies,1236.html

Edit: Imo you are best to keep Ram speed equal to or higher than FSB speed but with stock timings for now.
 
Last edited:
Soldato
Joined
1 Jun 2010
Posts
7,053
Location
London
FSB increase to 235

The RAM is overclocked slightly to 736 from 667 (timings stay the same)

Is your FSB currently set to 235MHz

Is your ram DDR3 667Mhz whose actual speed is 667/2 = 333MHz.

In this case even at stock speed your ram is much faster than FSB. So even if you underclock it slightly; as long as it is equal to or above 235Mhz, you will be fine:).
 
Last edited:
Associate
OP
Joined
12 Feb 2011
Posts
54
Is your FSB currently set to 235MHz

Is your ram DDR3 667Mhz whose actual speed is 667/2 = 333MHz.

In this case even at stock speed your ram is much faster than FSB. So even if you underclock it slightly; as long as it is equal to or above 235Mhz, you will be fine:).

Ahh okay, that makes sense! Thanks :)

I'll go and have a look for those divider things in bios now too.
 
Associate
Joined
10 Apr 2011
Posts
10
Okay so I've just got into using the Intel Burn Test for stability testing and for a control test I returned everything to stock in my bios and ran IBT with stress level on maximum. I plan to change settings and continue to run the exact same IBT, so I have some numbers to work off of. Anyway, the stock settings test came back perfectly stable and resulted in a consistent 36 Gflops. I then went into bios and set the ASUS auto-OC thingy to do a default 10% increase, leaving everything else on auto - I just wanted to see the results as a comparison to stock. Basically in this setting it puts the FSB up by 10%, so it went from 3GHz to 3.3GHz, and because everything else is left on 'auto', the motherboard decided to increase the volts to a stupid 1.5~! The RAM is overclocked slightly to 736 from 667 (timings stay the same), and the NB Frequency + HT Link are moved up 10% to 2200MHz - I assume the NB voltage and RAM voltages are adjusted as well? But I can't be sure as CPU-Z doesn't tell me those numbers. Either way, it doesn't matter what they've all been changed to because after running IBT on maximum stress the test finished with perfect stability - good news, right? Well not entirely, because for some reason the Gflops were consistently at 24 now?!? I thought this seemed very odd because doesn't that just mean that the CPU is massively slower now, even though it's overclocked by 10%...??

Am I interpreting this wrong, or has the OC actually made everything slower? Again, it was just a default 10% OC, where the motherboard picked all of the settings - everything runs consistently and fully stable, but apparently the CPU is 12 Gflops slower! What gives?

Can anyone shine any light on the matter for me? Or some insight perhaps? Anything really that might explain this result! Thanks!

Hi, saw your other thread and asked but it's a few pages back now, was just wondering how you're finding the casecom case?
 
Back
Top Bottom