Prime95

[FP]

I changed the memory usage in Prime95 from 8M to 512M and ran it. It failed after 7 hours. Other forums members mentioned that allowing Prime95 to use a lot of memory can lead to problems on A64's.

How much memory do you allow Prime95 to use when running the Large FFT test?

 

[CheesePoofs]

To test the CPU, you should run the Small FFTs test, not the big one.

 

[John]

With 1GB of ram I set it to 256MB & 512MB for 2GB. I always use large fft's (max heat) to stress test for 12-24 hrs, and if it passes I do a 32M superPI.

 

[Duvie]

Originally posted by: John
With 1GB of ram I set it to 256MB & 512MB for 2GB. I always use large fft's (max heat) to stress test for 12-24 hrs, and if it passes I do a 32M superPI.

S&M 1.5 will do a much better job then that....

IMO it is better (especially with these finicky A64 controllers) to test small FFT and use ltittle memory in testing to isolate CPU...There are far better apps to isolate and test the memory far better then Large FFT (which dioesn't do a good job of it), or superpi which is far too short to stress the memory....

 

[sharkeeper]

The new P95 is awesome.

 

[Duvie]

Originally posted by: sharkeeper
The new P95 is awesome.

Enlighten us!!! What is the new P95??

Edit: All I see is version 23.8 and that is been out for quite awhile...Is that what you meant???

I still think the version 22's were the most stable.....look at the newer versions...the newer and newer the versions go the more and more errors, and known issues the program has had with no fault to the system...

I use 23.8, but the version 22 is where it should have been left...simple!!!! NOt convoluted with multiple test (where we constant threads and confusion about what errors in some test and not in others mean), memory settings that can mess things ups, etc....

 

[VirtualLarry]

Originally posted by: Duvie
IMO it is better (especially with these finicky A64 controllers) to test small FFT and use ltittle memory in testing to isolate CPU...There are far better apps to isolate and test the memory far better then Large FFT (which dioesn't do a good job of it), or superpi which is far too short to stress the memory....

I don't get it. If the A64's memory controllers are "finicky", wouldn't it be best to test the most-stressful scenario, rather than the least? So for A64's, that would be the "blend" tests, with a custom RAM amount specified, such that P95 doesn't go into swap, I think.
I was just playing with P95 on my Athlon tonight, after moving some RAM back into my system, and tweaking the timings slightly. I set my RAS precharge, RAS-to-CAS delay, and something else, all to "AUTO", and then checked with AIDA32, and it did what I suspected - the BIOS simply used the most conservative settings possible for each of those, which is what I was planning on doing manually anyways.

No new version of P95, but they did find a new prime number, just last month. That was cool news.

 

[Duvie]

The blend test is not a very good cpu test cause it doesn't run the temps up high enough and stress the cach out as much as the others...I also have seen no errors in hours of prime95 blend but errors in memtest and/or superpi much sooner....I dont trust it even does a good job on the test that test ram (somewhat that is)....

Isolate is stil the best IMO...I have seen errors or issues arise on all other methods of testing then the one I listed....Then jump from P4 tto AMD and that has to change as well...