X2 4400+ & Prime95 performance

swindelljd · Aug 7, 2006

I am testing the stability of my OC on my new X2 4400+ ( ADV4400CDBOX ) and noticed something odd while running Prime95. It appears that one of the cores will run the torture test faster than the other. Right now they are both @ 238x11 (2.62Ghz) w/memory @1:1 1T. If I start Prime95 w/in 1 second on each core, after about 1/2 hour one of thie cores will be at least one cycle ahead of the other. After about 2 1/2 hours it had gone ahead by almost and entire test group (around the 128K test or so). My temps are 63C which is well within tolerance (<=71C) according to the AMD specs for this model.

So I am wondering if other people have noticed the same thing. CPU-Z says they are both matched in clock speed but I am wondering if maybe they aren't quite matched or something. Any ideas what is happening?

thanks,
John

hitbyaprkedcar7 · Aug 8, 2006

Run a super pi 1m and a 32m on each core (1 at a time) and post results.

Its common for one core to be stronger than the other.. But thats a big difference between strengths.

QuietIce · Aug 8, 2006

Mine did the same thing though not as bad. Try out this program and tell us if the slow core is the hotter one ...

http://www.ocforums.com/showthread.php?t=466707

swindelljd · Aug 9, 2006

I ran two different versions of superpi and here are the results
Test....CPU 0.............CPU 1
1M......33.078s............33.781s
2M......1m 16.359s......1m 18.375s
32M.....30m 01s..........30m 48s
Max temp(active test)
............60C..........61C

The temps stayed pretty consistent with the SpeedFan numbers +/- 1-3C and the cores were +/- 2C of each other regardless of whether it was the core running the test.

One thing I did notice was that when cpu 0 was running a test, cpu 1 had a lot (10% ish) of activitiy. However when I stopped the cpu 0 test and started cpu 1 tests I didn't see that same activity on cpu 0. It's possible that XP is scheduled "normal" system processes (firewall, anti virus, monitoring software, etc) onto cpu 1 even when under load.

This might explain the difference in timings. Or perhaps cpu 1 is truly a "weaker" core.

Here is a screen shot of tonights run (cpu 0 tests on left, cpu 1 tests on right):
http://www.flickr.com/photo_zoom.gne?id=210655703&size=o

thanks,
John

hitbyaprkedcar7 · Aug 9, 2006

Interesting.. Your "Weaker" Core, runs hotter.

Weird... Well it seems you truley, have a weaker core. But there running at the same speed... Thats weird..

Call up AMD and tell them. (tell them your at stock speeds, of course. run SP on stock speeds and give them those results, maybe theyll send you a new CPU)

QuietIce · Aug 9, 2006

hitbyaprkedcar7 said:
Interesting.. Your "Weaker" Core, runs hotter.

Weird... Well it seems you truley, have a weaker core. But there running at the same speed... Thats weird..

I have the same issue - that's why I asked. Core 0 is my weak core (the first to fail under Prime95) and according to CoreTemp it runs ~10-12°C hotter than Core 1 ...

hitbyaprkedcar7 · Aug 9, 2006

Really? Dayum. You should try reseating your heatsink if its THAT much hotter... Thats a lot.

LoneWolf121188 · Aug 9, 2006

Mine's somewhat similar (I don't remember my CoreTemp values exactly though, I posted them in that thread)...I don't know if I completely trust CoreTemp. :shrug:

I also don't really trust Prime. The way my computer is set up now, Prime fails almost instantly. Put I can play FEAR and stuff no problem, so as long as its stable enough for that, its stable enough for me.

QuietIce · Aug 9, 2006

LoneWolf121188 said:
Mine's somewhat similar (I don't remember my CoreTemp values exactly though, I posted them in that thread)...I don't know if I completely trust CoreTemp. I also don't really trust Prime. The way my computer is set up now, Prime fails almost instantly. Put I can play FEAR and stuff no problem, so as long as its stable enough for that, its stable enough for me.

I've wondered about CoreTemp as well. But I've been thinking about it ever since I ran it on mine - I'm wondering if the IHS is a little off. I mean, how much extra (whatever they use as a sealant/adhesive) would it take to leave just a hair more space on one side, which would lead to higher temps on that side of the die? Or maybe it's just a matter of placement - if Core 0 is the "inner" core it would tend to run a little hotter, wouldn't it?

LoneWolf121188 · Aug 10, 2006

There is no "inner" core, the cores are side by side.

And pretty much everyone is in general agreement that the IHS isn't completely flat on the bottom, resulting in one core almost always being hotter than the other, no matter how flat the top of your IHS is.

hitbyaprkedcar7 · Aug 10, 2006

^^^

Yes i knew that, but 12 degrees celcius? Thats alot. Even for an unveven IHS. And is it the temperature difference what causes the slower times in SP?

QuietIce · Aug 10, 2006

swindelljd said:
Here is a screen shot of tonights run (cpu 0 tests on left, cpu 1 tests on right):
http://www.flickr.com/photo_zoom.gne?id=210655703&size=o

I didn't add up your numbers but I just ran a Prime95 BM and got a total of 0.25s difference between Core 0 and Core 1.

Tried again setting Prime95-0 (my slow core) to high priority (left Prime95-1 @ low) and running each test individually instead of together and got essentially the same results - interesting.

BTW - Prime95 will reset the priority level when you change tasks - you have to change the priority while the test is running.

LoneWolf121188 said:
There is no "inner" core, the cores are side by side. And pretty much everyone is in general agreement that the IHS isn't completely flat on the bottom, resulting in one core almost always being hotter than the other, no matter how flat the top of your IHS is.

Surely the die isn't 100% symmetrical? But that's just a WAG - I've had a very hard time finding any information on physical placement of components on the die ...

davidk21770 · Aug 10, 2006

Since both cores are running from the same clock, they will be exactly as fast. The only thing that is different is what windows is doing with each core. You are running many things on the processors and they will take time away from prime/super-pi. If windows always uses the same core for it's tasks every time (or most of the time), that core will be "slower" because it is doing more than running prime.

The other thing that could make one core seem slower is memory contention. Are you sure that you are not running out of memory or have a memory bottleneck (slow memory or memory on a slow divider)?

AlabamaCajun · Aug 10, 2006

Davidk I think you hit on the point. One core which I'm sure to be defaulted to core zero will have a slight advantage due to the shared bus and memmory. Both cores are clocked at the same rate but one has to lead.

As for the higher temp on one core, [Speculative]I venture to guess core one is slightly warmer for kicking in wait cycles waiting on the queue to load. This is different than sleep cycles which would make it run cooler. [/Speculative]

QuietIce · Aug 10, 2006

I decided to take that line of reasoning to heart and it seems to be partially confirmed. I set affinity of all the username programs to CPU 1 and CoreTemp reports a drop in Core 0 of ~2°C - Core 1 temp didn't change. As supporting evidence the thermal probe I attached near the water block/CPU interface shows a 1.3-1.6°C drop (scale of 1/3°C).

Anybody know of a way to set affinity on the system programs ...?

deathman20 · Aug 10, 2006

AlabamaCajun said:
Davidk I think you hit on the point. One core which I'm sure to be defaulted to core zero will have a slight advantage due to the shared bus and memmory. Both cores are clocked at the same rate but one has to lead.

As for the higher temp on one core, [Speculative]I venture to guess core one is slightly warmer for kicking in wait cycles waiting on the queue to load. This is different than sleep cycles which would make it run cooler. [/Speculative]

David is right on that totally. While running 2 Primes one will absolutly run faster since one will chew up the core and the other misc processos in the back ground (like clock, processing desktop, explorer chews up some). Also you can get slower results if you are the limit of your OC, which will start degrading preformance, not on just on but both cores if you watch the results sometimes you'll see something that doesn't add up for the speed increase.

As for higher temps. Reseating is a biggy to get those temps down or at least try too. Are you running stock heatsink? Either way those temps are quiet warm most stay under 50C with 55C sort of what we call a limit, and if you are really giving it 1.5V ouch you really want some extra cooling on that core. Its not good for the CPU long term, nor is it probably for the rest of the components in the area of the CPU or PWM's I bet.

hitbyaprkedcar7 · Aug 11, 2006

davidk21770 said:
Since both cores are running from the same clock, they will be exactly as fast. The only thing that is different is what windows is doing with each core. You are running many things on the processors and they will take time away from prime/super-pi. If windows always uses the same core for it's tasks every time (or most of the time), that core will be "slower" because it is doing more than running prime.

The other thing that could make one core seem slower is memory contention. Are you sure that you are not running out of memory or have a memory bottleneck (slow memory or memory on a slow divider)?

Makes sense... I didnt think windows processes would kill a SP time that much. I disable every non important process because all they do is annoy me.

deathman20 · Aug 11, 2006

hitbyaprkedcar7 said:
Makes sense... I didnt think windows processes would kill a SP time that much. I disable every non important process because all they do is annoy me.

The more thats running the more it bogs it down. To bad we can't see that fraction of a percent its using. But we can guess that the CPU is probably around 1-2% load all the time.

AlabamaCajun · Aug 11, 2006

deathman20 said:
David is right on that totally. While running 2 Primes one will absolutly run faster since one will chew up the core and the other misc processos in the back ground (like clock, processing desktop, explorer chews up some). Also you can get slower results if you are the limit of your OC, which will start degrading preformance, not on just on but both cores if you watch the results sometimes you'll see something that doesn't add up for the speed increase.

Sorry if this looked like I was correcting him but that was not the case. I should have stated, if we could test the cores without the win backgrounders interfereing would we see the difference followed by the technical explanation.

Somone else try this and see if windows gets out of the way or if the thread link is BS, Let us know.
http://www.xtremesystems.org/forums/showthread.php?t=81429

swindelljd · Aug 11, 2006

AlabamaCajun said:
...
Somone else try this and see if windows gets out of the way or if the thread link is BS, Let us know.
http://www.xtremesystems.org/forums/showthread.php?t=81429

I'm willing to give it a try. Too bad I am travelling next week so I wont be able to test this for 10 days or so. I'll report back in with results once i have them.

If you look close at the screen shot I posted you can see a lot of cpu 1 activity while cpu 0 is running a stress test but not the opposite. Which implies that XP is scheduling system stuff onto cpu 1.

Thanks everyone, this thread has been a big help in validating (and better understanding) what I was seeing.

John

X2 4400+ & Prime95 performance

Registered

Member

Disabled

Registered

Member

Disabled

Member

Member

Disabled

Member

Member

Disabled

Member

Member

Disabled

High Speed Premium Senior

Member

High Speed Premium Senior

Member

Registered

Similar threads