Sign in with
Sign up | Sign in
Your question

Strange Behavior in Prime95? (PhII X3 720)

Tags:
Last response: in CPUs
Share
July 19, 2009 5:44:28 PM

I'm not sure if this is strange, but this is my first CPU that has more than 1 or 2 cores, and I've not experienced this behavior before.

Whenever I stress test in Prime95 (v25.9 x64 on Vista HP 64-bit), Core-0/Worker 1 gets through tests far faster than the other two, repeatedly. Example, I ran an in-place large FFT test for ~2 hours and Worker 1 completed mroe than 90 tests, Worker 2 about 80 tests, and Worker 3 only 75. Is this normal? Every once in a while, Worker 3 will get through more tests than Worker 2, but rarely.

Coming from an X2 5000+ BE machine, I've not seen such a wide difference while stressing a CPU. It's cores were always very close in final test count after an extended time test.

Motherboard is a Gigabyte MA790X-UD4P, and I've seen the same results w/ BIOS versions F4 & F5. Could it be Vista causing this due to background apps being pushed upon the 2nd and 3rd cores? If so, would it be better to runs these tests in Safe Mode? Would that even work? I've never tried it, as I usually test OC using software, such as AOD or the motherboard's AOD-like program.
a c 127 à CPUs
July 20, 2009 4:14:53 AM

I wounder if it could be related to the power savings that AMD has. I know its supposed to have independant power planes so I wounder if possibly one CPU is running slower than the others?
July 21, 2009 2:57:36 AM

Man, I have no idea what's doing it, but it keeps happening. Meanwhile, my brand-new Phenom II X2 550 machine hasn't shown a difference of more than 1 test in similar, ~2-hour runs. I just don't get it...

Should I be contacting AMD about this? Did I get a faulty, "cripple-core" or what?

I called Gigabyte about it and they said it could be Vista or Prime95 causing it... They were no help at all with this matter.
Related resources
July 21, 2009 10:22:48 PM

Self-bump... Called AMD, they say it could be Prime95, Vista, the motherboard, drivers, etc...

Seems everyone wants to put it on something else until an item totally freaking fails.
July 22, 2009 7:01:52 PM

Not that anyone cares, but...

I booted into Safe Mode and ran Prime95 25.9 and 25.8 for one hour each. All 3 cores ran the exact same amount of tests.

It seems such results would confirm that background operations within Vista-64 are being relegated properly to the 2nd and 3rd cores.

So, that said, I guess this is "Normal."
a b à CPUs
July 22, 2009 7:35:16 PM

there is a reason why we bash vista, and its not because it runs super efficient lol
August 12, 2009 11:49:54 PM

Update:

Turns out part of the problem lies in Gigabyte's over-clocking software - EasyTune 6. I'll explain...

I decided to try EasyTune6 since the MA790X-UD4P has built-in AMD OD tuning to see how well it would work for software OC'ing. I was amazed at how highly my CPU would OC with little to no voltage adjustment. I went from the stock 14X multi to 16.5X with no change in voltage, stable. I bumped it up 0.025v and hit 17.5X stable. I bumped it up another 0.050v and was 18.5X stable. That's 18.5X stable at 1.40v. That's pretty damn good, as most reports I've seen use voltages closer or equal to 1.50v for 18-19X stability. I was really happy, but it was those Prime95 stability tests that would constantly show Core 0 tearing through the tests way faster than Cores 1 and 2.

On a whim, I decided to use AMD's OverDrive to monitor the CPU while I Prime95 stressed after an EasyTune6 overclock. I had been using CPUID's HWMonitor. Son of a gun! There it was right in front of me...

EasyTune6 was only OC'ing Core 0! That's it! Other 2 cores were at the stock clock speed. I was baffled. So, I decided it was time to tear apart the other AMD machines lying around here and do the same thing. Those CPUs were a Phenom II X2 550, X2 7850 Kuma, and my old X2 5000+ BE. Wouldn't you know that EasyTune6 wouldn't OC the 2nd core in any of 'em?

I currently have a ticket in to Gigabyte about the issue, but we'll see what that gets me. Probably nothing but an apology...
!