Retail Yorkfield, load distribution issues?

[T_M]

Last night i tested Retail Yorkfield QX9650 on GA-P35-DQ6, after noticing odd behaviour on the weekend while benching under DICE.

What i had previously observed was in quad core benches my scores were lower clock for clock than other's results on hwbot (not just tweaking, a whole 150-200MHz slower). Then i found that in wPrime, one thread (3) was consistantly slower than the rest by a great deal, to the point where the end of a 1024M run should have been finished yet thread 3 was still only at 97% and still running for a good 6s+ after that.

Here's a screeny of 1024M at 3600MHz with thread 3 still running.
This screenshot is not as obvious as my DICE results which had 1,2,4 all finishing at the same time, then 3 later.


Thinking it might be a board/BIOS conflict with my X38T-DQ6, i swapped onto P35-DQ6 on air at 3600MHz.
I ran a series of checks and noticed the same behvious yet again.
Youtube videos below show good examples of what i noticed, however the compressions has made it difficult to see perfectly so i'll do my best to describe:

http://www.youtube.com/watch?v=DOeNdbM2zos
A video showing wPrime 1024M starting, then as it proceeds you can notice thread 3 slower than the rest and eventually getting overtaken in % complete by the others.
Task manager showing full 4 core loading running beside.

http://www.youtube.com/watch?v=_LGxmZ6iF9E
This is the crucial video.
SuperPI 8M started with no priority or affinity set.
Load is placed entirely 100% on core 2 (thread 3 equivalent) and remains there the entire test.
SuperPI restarted with affinity on all except core 2 (thread 3 equivalent).
Load is distributed evenly as normally experienced on other systems.

Why is all load being placed on core 2?
Why is my thread 3 lagging so much?

I discussed this observation with OPB last night on MSN, and he said he's noticed the same thing with over 20 different yorks.
He puts it down as just something we need to accept from these chips, however i dont agree.
Why didnt we see the same from ES versions of the York?

Please can some people with ES or Retail Yorks run similar checks and post results?
Im guessing that the faster times on hwbot are from ES chips that dont have this issue.

Or maybe Gigabyte's BIOS's have an issue?

[eva2000]

Weird indeed, I'll test my QX9650 as well at 3600Mhz just installed my new water cooling setup - Dtek Rad Tower box http://i4memory.com/showthread.php?t=5133

Off the top of my head i don't recall seeing anything like this on my QX9650. One thing is do you guys have the dual core hot fix patch installed WindowsXP-KB896256-v4-x86-ENU.exe ?? I have it slipstreamed as part of my Nlite custom build via RyanVM pack http://i4memory.com/showthread.php?t=4736

[MacClipper]

One thing is do you guys have the dual core hot fix patch installed WindowsXP-KB896256-v4-x86-ENU.exe

Out of curiosity, checked my SP2 installation, KB896256 is not installed but mentioned files are of version 3093 dated Feb 2007, even newer than those listed (version 3026 dated Oct 2006).

[eva2000]

not idea on version info heh

T_M, I experienced the same thing with wprime v1.51 and v1.55 with QX9650 Retail OEM L739A644 with Gigabyte X38-DQ6 F7D bios. I think this could be due to bios/board issues on iX38-DQ6 and/or wprime coding ???

But i don't experience the Super Pi 8M issue you have. With all 4 cores affinity default, 8M Super Pi the load cycles through all the cores and not just core 2. With core 1, 3, 4 affinity set, and core 2 unchecked affinity, 8M Super Pi still cycles through the 3 remaining cores leaving core 2 unchecked.

v1.51




v1.55

[eva2000]

T_M, try using systool quad 16M Pi test, it loads all 4 cores evenly so could be wprime/super pi coding issues too

Systool Quad 16M Pi test - (benchmarking -> cpu -> select 16M Pi test and check 4 threads box)

[antari]

I noticed a similar thing with my Kentsfield Q6600 on Asus P5K-E. Thread 3 in Wprime was always slower.

[BulldogPO]

Does that occur in Vista?

[Leeghoofd]

Same here one core always lags behind in prime95 a few secs with my P5K Dlx, I will try that patch Eva2000 mentioned....

[oscar3d]

I've tried that patch long time ago on a Dual CPU Xeon (2 cpux2cores) thatI use in my office, and to be honest I did not notice any changes or performance increase.

Now it's interesting that this is pointing out at the way that Wprime is distributng the workload...

Wait a minute... for what I've been reading this patch is supposed to work better with AMD CPU's which is not our case. I'm doing searches on google as we speak, and all what I can find is people with AMD x2 talking about this....

Wouldn't it be wPrime itself having problems there... which version are u using... going to test it on my yorkie.... If it performs the same, I'd vote that this is a software issue and not a hardware... and if it is... we can make Intel serve us breakfast and bend over backwards....

[T_M]

please do help by testing yourself on york
i was running 1.51, tonight ill check 1.55.

Also can you run the superPI test, because i'd liek to see if you get the same weird 100% loading on core 2.

[strange|ife]

pretty sure that patch is for amd systems

have you guys tried the microcode hotfix?

i would see if it pops up in vista aswell, or run a linux live cd, and download one of the stress programs to check in that os aswell, it could be related too droop, i dunno if your board is modded though probably not, could also just be a duff chip..

[T_M]

ive installed all critical XP updates

[camouflage]

not idea on version info heh

T_M, I experienced the same thing with wprime v1.51 and v1.55 with QX9650 Retail OEM L739A644 with Gigabyte X38-DQ6 F7D bios. I think this could be due to bios/board issues on iX38-DQ6 and/or wprime coding ???

But i don't experience the Super Pi 8M issue you have. With all 4 cores affinity default, 8M Super Pi the load cycles through all the cores and not just core 2. With core 1, 3, 4 affinity set, and core 2 unchecked affinity, 8M Super Pi still cycles through the 3 remaining cores leaving core 2 unchecked.

v1.51




v1.55

Were do I get coretemp 095.6 ???

[T_M]

Did you really need to quote his whole post?

[Leeghoofd]

We all love George but let's get back on topic....

You already had the critical updates installed T_M and still enountering the 3rd core lag ? No need for me to install it then

[Johnny Bravo]

Hmmmm got the same issue as do several other people I know.

Different boards so I'm not thinking its a motherboard issue
different cpus so I'm not really willing to say its the cpu issue - unless it affects the quad family :S
Thinking most likely it's a software issue, but is it OS or program?

[hipro5]

Same here with ES cpu AND the Retail one....
Did someone of you tested in Windows Vista?.....Maybe it's an OS issue....

[sonofander]

qx6700 + abit ip35 pro... Same here!

[eva2000]

We all love George but let's get back on topic....

hehe.. yes lets

Hmmmm got the same issue as do several other people I know.

Different boards so I'm not thinking its a motherboard issue
different cpus so I'm not really willing to say its the cpu issue - unless it affects the quad family :S
Thinking most likely it's a software issue, but is it OS or program?

i think wprime issue is coding issue tried on several boards same...

Same here with ES cpu AND the Retail one....
Did someone of you tested in Windows Vista?.....Maybe it's an OS issue....

really both for wprime issue or low 3dmark06 cpu score as well ?

qx6700 + abit ip35 pro... Same here!

which issue wprime or 3dmark06 or super pi ??

3dmark06 cpu score looks okay to me for mine

512MB Inno3D OC Edition stock @742/1836/2000
3dmark06 = 16,033



maybe 3dmark06 cpu issue is only under extreme sub zero cooling ??? and/or very high vcore 1.8+ causing cpu damage ??

[hipro5]

really both for wprime issue or low 3dmark06 cpu score as well ?

Not on 2k6........only on wPrime.....

[eva2000]

i wouldn't worry about wprime.. probably coding issues.. might try setting wprime to 2 threads, set affinity of wprime to cpu 0/1 and 2/3 for tests and see what happens heh

[safan80]

Weird indeed, I'll test my QX9650 as well at 3600Mhz just installed my new water cooling setup - Dtek Rad Tower box http://i4memory.com/showthread.php?t=5133

Off the top of my head i don't recall seeing anything like this on my QX9650. One thing is do you guys have the dual core hot fix patch installed WindowsXP-KB896256-v4-x86-ENU.exe ?? I have it slipstreamed as part of my Nlite custom build via RyanVM pack http://i4memory.com/showthread.php?t=4736

dang you you took my cardboard box idea

[AriciU]

This also happens on my dual core. I just ran wprime 1024m few hours ago and Core1 was lagging behind core2 by around 11%

[eva2000]

Yeah more feedback i see, seems it's wprime related not cpu

dang you you took my cardboard box idea

haha.. great minds think alike I guess

[STEvil]

I'm not using Yorkies but i've noticed that my Xeon 5340's @ 3ghz are doing this as well.

Also in 3dmark 2003, 2005, 2006 there is stuttering happening. Got 6500-ish on 3d06 cpu score iirc..

edit - Running XP-64.