Prime 95 and AMD 64 (90nm) underclock not compatible?

[Bean]

http://www.theinquirer.net/?article=20262
Apparently overclocking results in prime95 failures. But according to this post by JCviggen at xtremesystems.org (see below)"....never did I meet any K8 though which would not run prime95 at default speed & voltage with the stock cooler."

I wondered what happens if you undervolt/underclock? anyone?
-----------------------------------------------------------------------------
((Originally Posted by JCviggen
let me add a few things here (been ages since I last did that)

I came across A64 (FX) prime95 problems over a year ago already working for nVENTIV. At the time I was trying to specify the highest overclocked speeds "every" A64 FX would be able to do.

Prime95 kept giving problems. But i wouldnt straight out say that the systems were unstable because prime95 failed...because its a little more complicated than that.

First off I noticed that older versions of prime could complete just fine, whereas the latest version (at the time) would fail within seconds or minutes.

I had the luxury of using something like 7 or 8 different chips, and so I could compare. SOME batches did Prime better than others. The batches which were better at prime were not necessarily the best overclocking chips however. It all seemed fairly random.

But to give you an extreme example : I came across a few chips which would do absolutely EVERYTHING I threw at them @ 2800 MHz (phase change cooled of course) but would fail prime within seconds @ 2350 MHz same cooling. At the time I also talked to AMD about this issue, but they didnt really have anything meaningful to say about it. I didnt expect that either because if there was an explanation it would bound to be a highly complicated technical one.

Anyway I wasnt very confident in chips failing prime at such a low speed running much higher speeds so I did weeks of testing with any program I could find, and the result was that only prime95 kept messing up. I concluded that if an A64 fails prime95 at a given speed it doesnt necessarily mean its not completely stable. It means nothing more or nothing less than that it simply doesnt want to run prime95 at that speed. Since the problem CAN be eliminated by underclocking the CPU, it is definately an issue related to the CPU itself. It has absolutely nothing to do with FSB overclocking as I could play with the FX' multiplier.

So, do Athlon 64 chips have issues with prime95 : in my opinion that's a big YES. But somehow I'm not worried about it, I never managed to catch the chips making errors in real usable programs.

If your chip is failing Super Pi however that would be a serious concern. But prime95 alone will never dictate what is stable, for me. I also wouldnt blame Winchester too quickly, as I noticed huge differences in prime95 between different weeks and batches. One could run it at nearly every speed others would fail quickly.

But I can definately tell you that every member of the K8 family has issues with Prime. Some batches are better and some are worse but if you look hard enough you always find it. The "better" batches tended to be voltage hungry chips by the way. The ones that didnt stand voltage very well (though running cooler than the others) had a tendency to do worse at prime, but would most likely pass 3Dmark/Pi at the same speeds as the others

never did I meet any K8 though which would not run prime95 at default speed & voltage with the stock cooler.

that's my experience.))

[lm]

If prime95 fails, then there are 2 possible explanations:

a) prime95 has a bug

b) the cpu is not stable

Imo there are 2 states for a cpu: stable, or any level of unstable. Only stable is acceptable. Underclocking makes the cpu able to be stable with smaller voltage.

[Jan Kivar]

If the comp is not Prime95 stable at stock settings, something is broken. Mem timings are too tight, incompatible RAM/motherboard, unstable power etc.

Or Prime95 has a bug...

Cheers,

Jan

[josti]

Or Prime95 has a bug...

But that's very unlikly, I think.

Imo there are 2 states for a cpu: stable, or any level of unstable. Only stable is acceptable.

Very true. If prime fails, there will be other code that fails. Maybe not so easy to find.

I wondered what happens if you undervolt/underclock? anyone?

Undervolting has the same effect on stability as overclocking. In both cases you have to test what settings are stable.
If prime doesn't overclock well, you can't go very deep with your voltage at stock speed.
On the other side underclocking is no problem for stability.

[sngoda]

I recently built a new system consisting of:

Athlon 64 3000+ (90 nm, Winchester version)
Zalman 7000B cooler ( close to max [~10V] on fanmate)
Antec SLK3000B case (side panel off, fan on low)
Soltek K890Pro-939 motherboard
Geforce 6200TC videocard
Seasonic S12-380 power supply
1 Samsung SP1614C hard disk

I've been trying to undervolt the processor. To test stability, I boot w/ the ultimate boot CD (http://www.ultimatebootcd.com/) and run the Mersenne Prime Test. What I've found is that even running at the stock 1.4V, the Mersenne Prime Test fails after ~3 hr and 7 minutes. I find this extremely strange. I can't undervolt at all? And even at stock my system is unstable? Is it possible I have a bad CPU chip? Does anyone else have any experience with stock chips failing the Mersenne Prime Test?

I intend on trying two more tests: memtest86 at stock voltage/freq. and the Mersenne Prime Test w/ CPU overvolted but at stock frequency. I'll post the results after I've performed these tests, but I wanted to know if anyone had suggestions for either more tests I could perform or insight into what my problem could be.

Aside: since I'm running w/ fanmate near max and the side panel off, I believe that heating should not be playing any role in the prime test failing.

[Techno Pride]

probably not related,

but i've got a AXP-M 2500+ that fails P95 within 1 min @ 1.8GHz with either 1.475 or 1.5v. I'm too lazy to do bridge/pin mod again for lower voltages

It runs 2.0/2.1GHz fine @ 1.5v. Go figure

[Straker]

some things are just very demanding in weird ways - best recent example is probably the WoW installer. tons and tons of players had weird issues with the installer dying (can't remember if BSoD or not) repeatedly at the same area, or files not being read/written properly. turned out to be a memory issue in almost all cases, including mine; i can only speak for myself, but this was memory that had caused zero problems in 2+ years, on a machine that was never rebooted except for OS updates etc.

[sngoda]

Thanks for the replies (esp. Straker since it did turn out to be a memory problem). I just wanted to post an update. I ran again with the processor overclocked and it still failed the Prime test (this time after about 3 hr 45 min). Running memtest86 repeatedly gave errors for test #5 -- almost certainly suggesting memory was the culprit.

I have two 512 MB sticks of RAM, so I played with these. Either one alone or both at 333 MHZ speed would give no errors in memtest86, but both at 400 MHZ gave errors (note that these are Corsair Value Select DIMMs, so I doubt they are bad). Switching to the other memory bank on my Soltek board (the bank closer to the CPU) seemed to solve the problem. I can now run with both sticks of RAM at 400 MHZ with no errors in memtest86.

[mazurek]

I also had issues with 2 x 512 MB corsair value select 400 rams giving errors with Prime95, and some general system instability. Each would work alone but not together. Its on a P4 2.6. I got tired of errors, and sold them, and got some kingston ECC ram.

[Spod]

I had memtest86 errors with my Crucial Ballistix PC4000 running at DDR400, but when I took the case side off, they went away.
In the end, disabling CPU fan control (as it blows onto the RAM) and increasing case fan speed fixed it. My system isn't as quiet, but stability is more important to me. And they're Nexus fans, so they aren't especially noisy even at 12V.