Custom test for maximum CPU stress

[Torpedo]

Hi!

I'm new of this forum, a salute to everyone

I want to customize the Small FFTs torture test, to maximize ONLY cpu stress.

I have two situations...

AMD 4800+ X2, socket 939 (1MB/core L2 cache)

I thought those settings, correct me if I think wrong

Min FFT size-> 8K

Max FFT size-> 1024K (so the L2 cache is entirely tested)

Run FFTs in place-> enabled (what means? )

Time to run each FFT size (in minutes)-> 15

And a Core 2 Quad Q6600, step G0 (2MB/core L2 cache)

The settings...

Min FFT size-> 8K

Max FFT size-> 2048K (so the L2 cache is entirely tested)

Run FFTs in place-> enabled

Time to run each FFT size (in minutes)-> 15

Please help me!

BauZ

[Cruelty]

For AMD X2 go with two custom blend tests (you have to adjust memory).
For C2Q: 3 x small FFT + 1 x blend test - after some time (24h) switch from blend to small FFT and run for another 3-6h.
BTW: remember to set affinity for each instance especially for C2Q configuration.

[Torpedo]

Originally Posted by Cruelty

For AMD X2 go with two custom blend tests (you have to adjust memory).
For C2Q: 3 x small FFT + 1 x blend test - after some time (24h) switch from blend to small FFT and run for another 3-6h.
BTW: remember to set affinity for each instance especially for C2Q configuration.

I want to test ONLY cpu, I use GoldMemory to test ONLY memory and Blend Test to test the entire system (cpu+ram).

My question is for customizing Small FFTs to stress entire L2 cache

With new Prime 25.5 isn't necessary to set affinity!

[Cruelty]

Originally Posted by Torpedo

I want to test ONLY cpu, I use GoldMemory to test ONLY memory and Blend Test to test the entire system (cpu+ram).

As far as AX2 is concerned, it has an integrated memory controller besides I don't trust too much those dedicated memory tests as it seems that Prime95 does a better job at discovering memory problems.

My question is for customizing Small FFTs to stress entire L2 cache

I have never done it so go ahead and experiment. Bear in mind that FFT length probably does not translate 1:1 to cache size.

With new Prime 25.5 isn't necessary to set affinity!

1. It's still BETA.
2. AFAIR it doesn't allow to choose different tests for different cores / CPUs when stress-testing.

[Torpedo]

Originally Posted by Cruelty

As far as AX2 is concerned, it has an integrated memory controller

Yes, you have reason

I forgotten that the memory controller is integrated, as the new Intel Nehalem with 8 core

Explain to me...with the Small FFTs (when I do this test, I keep the ram underclocked...at this moment for example, 260*10-> 2600Mhz, ram with divisor 100 at 130Mhz...), I don't stress however the memory controller?

At this time I want ONLY find the limit of the cpu, I want to test ram after...

The Small FFTs, in this case, doesn't sufficient?

After, when I'll clock the ram I thought to use the blend test customized with 800MB/core...What do you think?

Originally Posted by Cruelty

For C2Q: 3 x small FFT + 1 x blend test - after some time (24h) switch from blend to small FFT and run for another 3-6h.

This test is only for test the cpu limit? (Why the blend on one core?)

And for test che system (cpu + ram), what do you suggest do me?

Originally Posted by Cruelty

besides I don't trust too much those dedicated memory tests as it seems that Prime95 does a better job at discovering memory problems. I have never done it so go ahead and experiment.

For my experience, with AMD64 and AMD64 X2...24h of Torture Test Blend OK, and after random freeze!

The problem? The ram...discovered with Gold Memory...!

Originally Posted by Cruelty

Bear in mind that FFT length probably does not translate 1:1 to cache size.

Yes, surely.

But if I set 1024KB in max FFTs size, I don't have more probabilities of test the entire L2 cache?

Originally Posted by Cruelty

1. AFAIR it doesn't allow to choose different tests for different cores / CPUs when stress-testing.

Yes, true...

[Cruelty]

I've played with A64 architecture some time ago and the blend test was simply better at detecting overclocking instability on those system than the other two tests. Someone else can have a totally different experiences, however I haven't seen so far an error that was not detected by Prime95 and was found by other software (in relation to CPU + RAM): memtest, goldmemory etc.
As for 1024kB FFT lenght test, all I was saying is that I am not sure if it will fit into L2 cache entirely, even if you set it to 512kB or whatever that value might be. Prime95 was prepared to test the stability of CPU and / or RAM subsystems without focusing on L1/L2 cache.

[Torpedo]

Originally Posted by Cruelty

I've played with A64 architecture some time ago and the blend test was simply better at detecting overclocking instability on those system than the other two tests.

Therefore in summary, in order to:

Find Cpu limit-> Blend test custom memory

Test stability of system Cpu+Ram-> Blend test custom memory

Right?

What amounts of memory do you suggest to me? (I have 2GB)

I do again this question...

Originally Posted by Cruelty
For C2Q: 3 x small FFT + 1 x blend test - after some time (24h) switch from blend to small FFT and run for another 3-6h.

Originally Posted by torpedo

This test is only for test the cpu limit? (Why the blend on one core?)

And for test che system (cpu + ram), what do you suggest do me?

[Cruelty]

Originally Posted by Torpedo

Find Cpu limit-> Blend test custom memory

Right for A64 architecture. If you want to stress only CPU in C2 architecture then you should choose only small FFTs tests.

Test stability of system Cpu+Ram[/B]-> Blend test custom memory

Right for any system. However you should not run more than one instance of blend test on platform other than A64, as the memory+FSB bottleneck will cause the CPU to not operate at maximum speed. That's why I suggest running C2Q stress test with 3 x small FFT + 1 x blend test (it will stress CPU+RAM).

What amounts of memory do you suggest to me? (I have 2GB)

For A64 check the amount of free memory using task manager, divide it by 2 and round it down to the full 100MB, then choose the blend test --> custom and enter that value there. For C2 architecture you don't have to worry about it since you shouldn't be running more than one instance of blend test and Prime95 should properly detect the available memory.

[Torpedo]

Originally Posted by Cruelty

Right for A64 architecture. If you want to stress only CPU in C2 architecture then you should choose only small FFTs tests.

How much time for A64 and C2? 48h?

Originally Posted by Cruelty

Right for any system. However you should not run more than one instance of blend test on platform other than A64, as the memory+FSB bottleneck will cause the CPU to not operate at maximum speed. That's why I suggest running C2Q stress test with 3 x small FFT + 1 x blend test (it will stress CPU+RAM).

This difference beetwen Core 2 Duo/Quad and A64 in Prime95, is the consequence of memory controller integrated in AMD cpu?

2 simultanoes blend test on Core 2 Duo will sature the FSB, right?

And for Core Duo test system (cpu+ram) will be

1 x small FFTs + 1 x blend test

Right?

Originally Posted by Cruelty

For A64 check the amount of free memory using task manager, divide it by 2 and round it down to the full 100MB, then choose the blend test --> custom and enter that value there. For C2 architecture you don't have to worry about it since you shouldn't be running more than one instance of blend test and Prime95 should properly detect the available memory.

Ok, thank you.

[Cruelty]

I consider 24h test = enough. As for the rest of your questions, 2 x right

[Torpedo]

Originally Posted by Cruelty

I consider 24h test = enough. As for the rest of your questions, 2 x right

Thank you very much Cruelty