Linx/Linpack test fails for stock system/parts

[mitush]

Linx/Linpack consistently reports errors on my self-built system without any OC (i7-860, Asus p7p55d evo, 4GB Corsair xms3, ATI Radeon 5870, Corsair TX850W PSU, Win 7 Home Premium). When the test runs, CPU-Z shows CPU voltage changing from 0.9v to 1.15v. CPU Temp seems to be ok. See the attached screenshot. System is completely stable otherwise at least for the web browsing, but I have not tried any other cpu intensive apps, like gaming. Also, I've disabled SpeedStep in BIOS. Not sure if it is related, but the system seem to get into those couple-seconds brain-freeze moments (for the lack of a better word) when SpeedStep was enabled.

Any other tests could help me pinpoint the problem? Any other suggestions?

 

[mitush]

I tried the 32-bit version of Linpack test on the same box and it looks ok:

Intel(R) LINPACK data

Current date/time: Thu Jan 07 12:58:04 2010

CPU frequency: 2.809 GHz
Number of CPUs: 8
Number of threads: 8

Parameters are set to:

Number of tests : 1
Number of equations to solve (problem size) : 10000
Leading dimension of array : 10000
Number of trials to run : 1
Data alignment value (in Kbytes) : 4

Maximum memory requested that can be used = 800204096, at the size = 10000

============= Timing linear equation system solver =================

Size LDA Align. Time(s) GFlops Residual Residual(norm)
10000 10000 4 34.430 19.3689 9.299606e-011 3.279135e-002

Performance Summary (GFlops)

Size LDA Align. Average Maximal
10000 10000 4 19.3689 19.3689

End of tests

Thu 01/07/2010
12:58 PM

GFlops are not great though and I do need the 64-bit to work too.

 

[Boulard83]

First off all, try to run at least 2 memtest86 pass. This can prove that your RAM can take it.

( this is a bootable prog that you can burn to a CD )

 

[MattNo5ss]

Sorry, you seem to be having trouble, but the OCF community should be able to help you determine what's going on.

You might want to post your BIOS settings for reference too.

Here's a couple things to try:
- Test your RAM with MemTest, to see if your RAM is bad.
- Try increasing vCore a little.

 

[redduc900]

Like Matt suggested, bump the Vcore to around 1.175 - 1.200V (reading in CPU-Z w/ LinX / Linpack running).

 

[mitush]

still no luck

Increased CPU voltage to 1.2v via BIOS. Temereatures went up. Got Linpack to intermittently work (see screenshot), Linx sometimes failed on the first test and sometimes finished 3 tests ok (screenshot). Apparently Linx has an older version of linpack exe included (older than 10.2.3). Why it reports 35GFlops vs Linpack's 20GFlops is not clear. Also, Windows 7 had an auto update in the middle of my testing.

Also, memtest86+ 4.0 shows no errors after two passes (1 hour of testing).

Last few Linx/Linpack tries were again all errors.

Any thoughts?

 

[mitush]

any more ideas?

The issue is still unresolved. Is there any other modern cpu checking utility which is easy to use and has a small footprint?

 

[Bobnova]

For giggles, try Prime95.
Also try Memtest86+

 

[vgta88]

try prime 95 with small fft which will fit in your cache (to rule out cpu without touching memory)

I think it's a chipset voltage issue. chipset voltages could affect memory. or maybe not enough volts to your memory.

 

[mitush]

prime 95 test

Tried prime95 torture test from http://www.mersenneforum.org/showthread.php?t=12155

"Small FFT" (few times) and "In-place large FFT" worked OK, while it reported error for "Blend" test right away (see the screenshot).

Where is that stress.txt file it says I should read?

 

[Bobnova]

Sounds memoryish, try Memtest86+

 

[mitush]

DRAM voltage

vgta88 / bobnova,

indeed memory looks to be a likely culprit... I've increased the DRAM voltage in BIOS to 1.8v (from auto, about 1.5v-1.6v) and now both Lynx and Prime95 report no errors so far. I ran Lynx with 20k problem size (3GB) and tried all three mods in Prime95.

By the way, bobnova, I did run memtest86, 2 passes, no errors (see one of the posts above).

Thanks a lot for your help guys!

 

[Bobnova]

So you did, i missed that, sorry.

1.8v is a lot for DDR3 ram, is it rated at that voltage?

 

[mitush]

memorey voltage

You are right, this memory is rated 1.65v. I am a noob, so what do I know . I changed the voltage to 1.65 and doing testing again. One Lynx and one Prome95 "blend" tests were ok.

 

[Bobnova]

Excellent, hopefully that was the only issue.

 

[MIAHALLEN]

Make sure if you want to run your memory higher than 1.65V for 24/7 ops, you increase your VTT to within 0.5V of the mem voltage...otherwise, it could damage the CPU

 

[mitush]

progress

After fixing DRAM voltage, played around with overclocking via BCLK frequency only. About 40+ Lynx tests were all good, except once (after this I increased the "IMC voltage" to 1.2v. MIAHALLEN, is that the VTT voltage that you mentioned?)

At BCLK default setting (133MHz) I get about 35GFlops in Lynx, CPU core temp maxes at 60 degrees C. 166MHz yields 43-44Gflops, but the temperature is too high at 98C. 150MHz looks to be the best option with 42.5Glops and 80C. Also tried 160MHz, but both Gflops and temperature is lower than 150MHz, must be some sync reason.

Finally, disabled CPU Hyperthreading in BIOS, and now at 150MHz I am getting 44Gflops (Lynx with 18k or 20k problem sizes) and the CPU stays relatively cool at 70C.

Anything else (simple) I can try to meaningfully increase the Gflops and stay below 80 degrees C (with the existing hardware)?

 

[MIAHALLEN]

...."IMC voltage" to 1.2v. MIAHALLEN, is that the VTT voltage that you mentioned?)

Yes