Prime 95 v.2.56 Small vs Large FFT

[Wolfcastle]

The Large FFT setting says it's for maximum heat, but I ALWAYS get higher temps with the Small FFT setting. Additionally, if I want to test the memory and northbridge, I use the memory test on the Prime 95.

Since the large FFT is no longer the test that generates more heat than small FFT, is there another reason to run it?

I also use linpack 64 bit, which generates about 7c more than Prime95 2.56 small FFT.

 

[JAG87]

This is a very good question.

Large FFT is absolutely necessary. As a matter of fact, it stresses the cpu more than small FFT. The reason why small FFT produces more heat, is because the cache is being used a lot more, not because the workload is more.

Do not rely on small FFT for stability. Small FFT is a way to check your CPU's maximum heat output, and basically test the efficiency of your cooling system. And even in that department, it is surpassed by programs like linpack, and intel TAT.

However, large FFT is still one of the best (if not the best) way to test for stability. Your cpu can be 8h small FFT stable, and then crash after 30 minutes of large FFT.

This is why blend is such a good test, because it does the best of both worlds, it heats up your cpu to its max, and it tests for stable operation.

 

[Duvie]

I run 12h of small FFT out of the gate and round it up with at least 12h of blend....

The last time I ran Large FFT I didnt' see any big power draw or temperature difference...hoenstly temp hasn't been much of an issue for me the last year or so...

 

[Idontcare]

Originally posted by: JAG87
This is a very good question.

Large FFT is absolutely necessary. As a matter of fact, it stresses the cpu more than small FFT. The reason why small FFT produces more heat, is because the cache is being used a lot more, not because the workload is more.

Do not rely on small FFT for stability. Small FFT is a way to check your CPU's maximum heat output, and basically test the efficiency of your cooling system. And even in that department, it is surpassed by programs like linpack, and intel TAT.

However, large FFT is still one of the best (if not the best) way to test for stability. Your cpu can be 8h small FFT stable, and then crash after 30 minutes of large FFT.

This is why blend is such a good test, because it does the best of both worlds, it heats up your cpu to its max, and it tests for stable operation.

For what its worth, when I setup my recent Q6600 boxes I used small FFT stability as litmus test for driving the Vcore setting and once I was small FFT stable (at my target OC of 3.3GHz) the systems were not large FFT stable unless I upped the MCH and NB votlages.

I may have completely misinterpretted the results and purpose of large FFT, but for me it seemed that instability with large FFT was more indicative of NB and ram instability than CPU instability.

I've never had an overclock be small FFT stable and large FFT instable and the solution to large FFT instability be tweaking the CPU inputs (Vcore). Large FFT instability for me has always been resolved by shoring up the memory subsystem (upping NB/MCH voltage and dimm voltage) if the system is otherwise small FFT stable. But this is just my experience.

 

[JAG87]

I have had my cpu 6h small FFT stable crash after 55 min of large FFT, and constantly crash in games.

Raised vcore by .025v and it's forever large FFT stable and never crashes in games anymore, never touched NB or DIMM voltages.

This is with water cooling, which handles small FFT heat very well, making small FFT not a stressful test by any means. This is my experience at least...

 

[graysky]

My vote is for large FFTs. I measured my power consumption with my killawatt and found that large FFT's produces slightly more power consumption link