prime95 using a P4 2.8C

Hey all, I got my P4 2.8C oc'ed at 3,3gigs and I wanna make a stability test which prime95 is famous of doing by torturing testing it... so what of the 3 options in the torture test is best to choose? Small FFTS? In-place large FFTS? or Blend?

And something weird is when I open Windows Task Manager while running Prime95 it says only 50% CPU usage.. shouldnt it be 100%? I heard someone saying that you should run 2 instances of it.. and someone saying that it should be enough with 1 instance of it...

I'm clueless!

Peter

[nomadicus]

Originally Posted by Unregistered

And something weird is when I open Windows Task Manager while running Prime95 it says only 50% CPU usage..

You have hyper threading (HT) turned on. This means that Windows sees two virtual CPUs when in reality it is just one physical CPU. The task manager displays it as 50% of the combined two virtual CPUs which really means 100% of the physical CPU (hey, it's confusing, but that's virtual for ya).
If you have Windows XP, you can leave HT on but for each % on the second virtual processor means that prime95 runs that much slower (the reported 20-30% boost with hyper threading enabled does not apply to prime95 due to its optimized assembly code).
Personally, I have Windows XP and I have HT turned off so the stats make more sense to me (just my preference).
Windows NT is reported to have an inefficient scheduler which means you should have HT turned off to maintain best performance if you are running NT.

Thanks for your quick reply ;)
Because I was a little bit worried I'm not stress testing my computer efficiently? But now when you're saying that 50% is actually 100% calms down my nerves ;)

Actually I ran 2 instances with the affinity option on before just to be sure my system is being stress tested at 100% but that made my system go instable.. but maybe that isn't so weird since it was then running at 200% since 50% equals 100% cpu load right?

Do you recommend the Blend option in the torture test btw? To see what my system is capable of?

Sincerely,

Peter

[nomadicus]

Originally Posted by Unregistered

Thanks for your quick reply ;)

You are welcome.

Actually I ran 2 instances with the affinity option on

Someone can correct me, but I believe affinity for two instances of prime95 each on their own virtual processor doesn't make any significant difference thruput wise. They run half as fast, but do twice the work . . .

just to be sure my system is being stress tested at 100% but that made my system go instable..

How do you mean unstable? the machine ran slowly or had a hardware error? If the former, you may be paging/swapping heavily and need to reduce the amount of memory prime95 can use. If the later, most likely it is a hardware problem that needs attention.

but maybe that isn't so weird since it was then running at 200% since 50% equals 100% cpu load right?

HT means two virtual CPUs and
50% means 100% of the physical CPU and
50% + 50% means 100% of the physical CPU but
each 50% runs roughly half as fast and
the thruput is very dependent on the optimization, or,
rather, the lack of optimizations, of the code, therefore,
prime95 doesn't really take advantage of HT one way or the other when it comes to LL. In fact, others have reported that 2 LL's run less efficiently (a couple percent?) on HT and Windows XP.

I know it is rather confusing but HT is meant to take advantage of unused pipeline cycles. Prime95 is coded to fill the P4's pipeline and keep it full as possible; therefore, all of my prime95/P4/LL machines have HT turned off or else I would go .

Originally Posted by Unregistered

Do you recommend the Blend option in the torture test btw? To see what my system is capable of?

I'll let someone else answer that. I haven't used the new torture test very much.

Alright things are getting a bit clearer here ;)

But as for stability options running one instance of Prime is as good as running two if I've understood you correctly?

[TauCeti]

Originally Posted by nomadicus

If you have Windows XP, you can leave HT on but for each % on the second virtual processor means that prime95 runs that much slower (the reported 20-30% boost with hyper threading enabled does not apply to prime95 due to its optimized assembly code).

That is true if you exclusively use the system with prime95. If you also use the system for 'normal' work or simultanously participate in other DC-projects with that box, using HT gains up to 60% 'free' performance (for both projects combined) with Win XP.

How much performance you gain depends on the kind of workload the programms have. One more extreme example (close to that 60% gain) is mixing the not-bandwidth-limited prime95 TF workload above 2^64 (Integer SSE2) with the memory-bandwidth-limited NFSNET client (no SSE2).

It is also interesting that the power consumption of the CPU in the aforementioned setup does not increase.

Tau 'HT' Ceti :)

[outlnder]

Use the "Blend" setting. Makes the CPU and Ram work harder.