Sandy Bridge peculiarities in stability testing

[ugotd8]

Hoping someone else has noticed the following two points in running Linx and or Prime95 when OCing these new chips:

1. Gone is the disparity in temps between Linx and Prime95.
I'm seeing the exact temps on 100% load no matter what I use, IBT, Linx or P95. A common quote among people regarding stability testing on the I7 45nm chips was "Linx stresses the CPU harder and that explains the sometimes 10-15C difference in temps". What explains the lack of disparity now ?

2. It's going to fail and BSOD in the first few minutes, or not at all.
Granted my longest P95 run so far is a mere 2 hours, but in every case when I didn't give it enough VCORE, I found out within the first couple of minutes.

[PatRaceTin]

Lin X "ALL" memory give more heat

prime95 "blend" with long hour run to ensure those LinX

don't forget Game / App running test too

[zalbard]

You're using old LinX, then. Or wrong settings...

[ProStreetCamaro]

I am actually getting hotter with P95 than with LinX/IBT

[st0ned]

You're using old LinX, then. Or wrong settings...

What I found out is that when running LinX 0.6.4 ( from your linx stable thread ) I'll get around 48Gflops with 100% load on all cores, CPU-Z reading 5Ghz freq. And the multi stays steady at 50x . All cores keep around 62ºC load.

But when I run a older linX version that I had on my backup disk ( 0.5.8 ) it jumps the cores straight to ~66ºC and I get around 68Gflops.

So at this moment I think that I'm getting some strange issues with linX. I've the default settings on both with [All] memory select ( 7Gb runs /gasps )

[zalbard]

Maybe it's due to the latest build... It's supposed to use AVX on SB. And you need Win7 SP1 to do so. Perhaps it's trying to do so under Win7 RTM and fails, who knows... In any case, that's the latest Linpack library (the one linked in the OP), I'd recommend sticking to that.
It's really too bad I do not have the source code of LinX, I could have a look and try to fix it.

[st0ned]

OK that might be just it. I'm running a normal 7 RTM version without anything installed at all.

0.6.4 both crash the CPU, used the same overclock on both, and both gave me a BSOD at the 7th loop, thing is 0.6.4 took a big buch of time to get there

[ugotd8]

You're using old LinX, then. Or wrong settings...

Linx 0.6.4 - Settings:
Mode: 64bit
Number of threads: 8
Priority class: normal
Data Alignment: 4
Linpack32 max problem size: 15500
Memory for OS: 5
Stop on error
External apps: none

0.6.4 - generates ~95W max temp 64 per HWMonitor:



Prime95 - generates over 100W max temp 71 per HWMonitor (and also does not push the VCore to the proper level?)

[SBB]

With regards to "failing in the first few mins" - during my benching so far I've seen fails after 10-15 mins but generally after that you're OK

[ugotd8]

With regards to "failing in the first few mins" - during my benching so far I've seen fails after 10-15 mins but generally after that you're OK

Thanks SBB. I'm starting to see that now that I get closer to stability.

The other point is still killing me... I can pass Linx and IBT all day, but Prime95 small FTTs not only generate more watts and higher temps, but it will BSOD on clocks that IBT and Linx have passed.

[FlanK3r]

lol, its strange...At AMD Thuban I have higher temps in LINX than in PRIME test..

[Leeghoofd]

Same here UgotD8, Prime95 seems to be more stressful atm. Setup passes 4.5Ghz linx with 1.275 set in bios. Prime just BSODs and needs 1.285 to be stable for an hour with Custom test 12 - 12K set...

[FlanK3r]

its strange...maybe, linx can not strassing cores with htt?

[Leeghoofd]

It's okay now, I grabbed the latest LinX version here : Linx64v10.3.0.003.zip

[RJARRRPCGP]

2. It's going to fail and BSOD in the first few minutes, or not at all.
Granted my longest P95 run so far is a mere 2 hours, but in every case when I didn't give it enough VCORE, I found out within the first couple of minutes.

Even though I haven't migrated to a later platform and still with 775, my experiences with Prime95 are the following:

You typically have a 2 or 3 hour window for Prime95 to give an error.

If you don't have enough Vcore, or at a hard wall, Prime95 usually gives an error at the 2 hour mark, if not before. And it's rare for Prime95 to fail at 6 hours.

Prime95 usually finds out within 2 hours.

[owikh84]

Even though I haven't migrated to a later platform and still with 775, my experiences with Prime95 are the following:

You typically have a 2 or 3 hour window for Prime95 to give an error.

If you don't have enough Vcore, or at a hard wall, Prime95 usually gives an error at the 2 hour mark, if not before. And it's rare for Prime95 to fail at 6 hours.

Prime95 usually finds out within 2 hours.

Not 2 hours, u can fail at 3 hours if wrong dram timing.

[justin.kerr]

On my SB at 5Ghz I get around 66 Gigamefloppens. with HT on temps very close to prime95 +/- 3 C

[FlanK3r]

maybe this is the problem....65GFloops its ok maybe for i5 2500K, but not for 2600K, because with Thuban at 4300 MHz I broked 80 GFloops. And OC i7 2600K is better than OC Thuban

[kevink82]

my prime fails after 2 hrs for small fft, linx i can run as many loops without problem so atm i consider linx is no longer a valid stability test for sandy bridge.

[Leeghoofd]

maybe this is the problem....65GFloops its ok maybe for i5 2500K, but not for 2600K, because with Thuban at 4300 MHz I broked 80 GFloops. And OC i7 2600K is better than OC Thuban

Just quickly tested the 2600K at 4.5Ghz, rams at 1866 C7

With HT on : 57.9 flops
With HT off : 66.5 flops

This is the latest build of IBT in the LinX program found at the 4.5Ghz stable thread...

In fact I could care less for the flops result, but it has to properly stress it...

[-=[kz-0]=-]

maybe this is the problem....65GFloops its ok maybe for i5 2500K, but not for 2600K, because with Thuban at 4300 MHz I broked 80 GFloops. And OC i7 2600K is better than OC Thuban

its 2 different systems you are trying to juice out..YES the X6 seems to give more Gflops than Intel's but the heck..also enabling HT allows a decrease in GFlops(depending on the number or physical cores) for the price of extra threads..

[RJARRRPCGP]

Not 2 hours, u can fail at 3 hours if wrong dram timing.

I said 2 >or< 3. I dunno exactly.

But, I know that a Prime95 failure is unlikely at 10 hours on many setups.

[gurneyhck]

Why are you guys not hitting way beyond 100Gflops??

My setup:
2500k@4.7GHz
Asrock P67 Extreme4 + 8GB Corsair Vengeance@1600MHz 9-9-9-24-1

Getting this result in Linx 10.3.0.003:



Is this because I'm also running SP1 (win7)? Very curious what the explanation could be...

BTW, CPU has clocked back in this screenshot cause LINX finished.

[Alex-Ro]

Ofc it's the SP1.And yea using AVX requires more voltage and generates more heat,making it the most stressful stability test to date for sandy bridge

[famish99]

How were you able to get a hold of SP1? I'm fairly sure Microsoft hasn't rolled it out yet. It is shocking to see how much of a performance gain is caused by switching to AVX, but I suppose it uses the execution units more efficiently than SSE.