Magny Cours "Black Edition"

10e, I have to say that this rig fascinates me to the point of shopping for MC 12 cores...

Can you elaborate on what is not working properly on the benchmark utility?
 
10e, I have to say that this rig fascinates me to the point of shopping for MC 12 cores...

Can you elaborate on what is not working properly on the benchmark utility?

If I run the benchmark utility "stock" ie. using only the instructions provided and without any changes, it works fine but apparently only loads up the CPUs at 67%.

It turns out the normal "-smp" switch in the SMP client defaults to maximum 32 cores if you don't set a number after it. This was indicated in Alpha754293's thread at www.foldingforum.org when he first tried out the 48 core 6174 box he has at work.

So then I copy over a Folding@home-Win32-x86.exe binary from an SMP directory and run -configonly to change the client.cfg to -smp 48, and re-run, but then it gives strange stats.

A 2686 took either 27:40 to finish a frame, or finished 4 frames in 10:23, so something is strange.

I think in this scenario somehow it seems to run the "wrong unit" if that is possible. Ie. run a 6061 or 6701 instead of the chosen 2686 unit.

My next step is to try and just copy my client.cfg file from my regular folding client into the "data" directory and try using that.

If that works then I can print some solid numbers. I really wanna see what a 2684 does. If I get a TPF of 14:00 or lower I will be on cloud 9, as this means it really can finish any BigAdv work unit in a day or less.
 
It turns out the normal "-smp" switch in the SMP client defaults to maximum 32 cores if you don't set a number after it. This was indicated in Alpha754293's thread at www.foldingforum.org when he first tried out the 48 core 6174 box he has at work.
This *should* be fixed with the next update of the core especially considering more people are starting to fold on rigs with more than 32 cores.
 
If I run the benchmark utility "stock" ie. using only the instructions provided and without any changes, it works fine but apparently only loads up the CPUs at 67%.

It turns out the normal "-smp" switch in the SMP client defaults to maximum 32 cores if you don't set a number after it. This was indicated in Alpha754293's thread at www.foldingforum.org when he first tried out the 48 core 6174 box he has at work.

So then I copy over a Folding@home-Win32-x86.exe binary from an SMP directory and run -configonly to change the client.cfg to -smp 48, and re-run, but then it gives strange stats.

A 2686 took either 27:40 to finish a frame, or finished 4 frames in 10:23, so something is strange.

I think in this scenario somehow it seems to run the "wrong unit" if that is possible. Ie. run a 6061 or 6701 instead of the chosen 2686 unit.

My next step is to try and just copy my client.cfg file from my regular folding client into the "data" directory and try using that.

If that works then I can print some solid numbers. I really wanna see what a 2684 does. If I get a TPF of 14:00 or lower I will be on cloud 9, as this means it really can finish any BigAdv work unit in a day or less.

Start over. After you unzip it, open the client.cfg file in Notepad. Somewhere in the cryptic string you will see "extra_parms=-smp". Change that to "extra_parms=-smp 48" and you should be set. If you want, I'll send you a new client.cfg file.
 
If that works then I can print some solid numbers. I really wanna see what a 2684 does. If I get a TPF of 14:00 or lower I will be on cloud 9, as this means it really can finish any BigAdv work unit in a day or less.

That was my hope as well, but of course I fell short with 26hrs on a 2684. But of course, your average will be well under 24hrs, given the prevalence of units since the 6900 release 3.5 weeks ago:

Unit WU
6900 53 56%
2685 19 20%
2684 10 11%
2686 10 11%
2692 2 2%

Total 94

I should not have asked you to prioritise the 2686, had I thought it through I should have said 6900, 2685, 2684 were the priorities. Sorry bout that chief.

Still thrilled someone has made a 2684 killer rig. We need to stop work on the benchmark immediately and write automated software to sneakernet all our 2684s to 10e for maximum team efficiency. :p

Ahhh time for beddy byes, that glass of red went straight to my head.:eek:
 
Start over. After you unzip it, open the client.cfg file in Notepad. Somewhere in the cryptic string you will see "extra_parms=-smp". Change that to "extra_parms=-smp 48" and you should be set. If you want, I'll send you a new client.cfg file.

Sure thing. I'll try that out on my own first and let you know later today.

I may also swap the drive out and try Windows 2003 x64. Windows 2008 has so much stupid-ass security that it may be causing other issues even though I "Run As Administrator" I might even try that 128GB SSD I bought a while back that's still in the box :confused:

That was my hope as well, but of course I fell short with 26hrs on a 2684. But of course, your average will be well under 24hrs, given the prevalence of units since the 6900 release 3.5 weeks ago:

Unit WU
6900 53 56%
2685 19 20%
2684 10 11%
2686 10 11%
2692 2 2%

Total 94

I should not have asked you to prioritise the 2686, had I thought it through I should have said 6900, 2685, 2684 were the priorities. Sorry bout that chief.

Still thrilled someone has made a 2684 killer rig. We need to stop work on the benchmark immediately and write automated software to sneakernet all our 2684s to 10e for maximum team efficiency. :p

Ahhh time for beddy byes, that glass of red went straight to my head.:eek:

No worries! The 6900s run in an 18 hour period so it's no biggie, and I don't care about personal points right now. If Musky's solution above works for me, or if Win2k3 works better with the benchmark I'll do at least 10 frames on all the units.

This is the fun here, not the e-peen :cool:

Once done I'll compare it to my SR-2 which funny enough is running at 3.9Ghz, while the MC box runs at 1.9Ghz, so it's kinda like a direct comparison (outside of CPU architecture of course) as one has 24 threads at 3.9Ghz and the other 48 at 1.9Ghz.

Or I'll just try 3.8 ghz with a 200mhz BCLK on the SR-2 to make it as direct as possible.
 
just disable the UAC crap in windows 2008. should solve the run as admin stuff. at least it does in windows 7 which they are basically the same OS.
 
This is the fun here, not the e-peen :cool:

I often wonder if I had just settled for quick and dirty 4.0Ghz on all and left them in a room folding what my score would be by now. :p

Nothing better than groupthink optimisation. :D
 
Excuse my trip down memory lane:-

It's funny. The last time a saw a computer with 48 processors (2x Sun e6800's in an active/ active failover pair), it took up two racks (not rack units, racks!). This is within the last decade, it's amazing how times have changed.

RIP Sun.
 
I often wonder if I had just settled for quick and dirty 4.0Ghz on all and left them in a room folding what my score would be by now. :p

Nothing better than groupthink optimisation. :D

You'd be higher, but you would have:

A) had less fun
B) had less frustration LOL

Don't even go there. If I just left everything folding under one account I'd have over 100m points by now, but where's the fun in that?

I'd rather be a two-timer or three-timer. Spread the "D" love!

Update!

The Noctua NH-U9DO A3 G34 and C32 compatible heat sinks arrived yesterday and I've installed them!

They have dropped full folding temp to 42 celsius at 100% load on CPU1 and 4 where the little shorties were installed! Yay!

Time to get a couple more methinks!

Excuse my trip down memory lane:-

It's funny. The last time a saw a computer with 48 processors (2x Sun e6800's in an active/ active failover pair), it took up two racks (not rack units, racks!). This is within the last decade, it's amazing how times have changed.

RIP Sun.

I remember these things too. Each one was basically a rack on its own. Things change, and they change pretty darn quickly.
 
10e, what were your temps before the noctuas?

On these chips, originally with the Thermaltake CL-P0314 modified with AMD Phenom stock cooler 70mm PWM fans, temps were getting up to 54 - 55 celsius. So big difference.

So I made a little tweak to the memory by enabling channel interleaving along with shutting off NUMA and my TPFs on P6900 R24, C15, G8 are down to 10:44 for 174K PPD.

Holy crap.
 
damn 55C aint bad, but 42C is damn good. to bad all 4 cant be overclocked. my poor phenom II 940 is sitting at 56C right now. to lazy to clean my heatsink. i usually wait til it hits 68C before i clean it.


sirmonkey thinks maybe you should remove the 2 non es chips and see if you can hit the 3Ghz+ barrier with those noctua heatsinks :D
 
mineral oil wont help. actually more likely to make it worse then better. problem with mineral oil is that it retains more heat then it allows to escape. so your only choice is to constantly circulate the mineral oil. at that point you might as well go full blown water cooling.
 
Back
Top