CUDA enabled GPUs and Seti@home

[petrusbroder]

I got myself a present today - a Nvidia 9600GT card.
I downloaded the latest drivers and set it to crunch seti@home.

Here are some results:

GPU:
__ CPU _____ claimed ______ credits per
__-time ____ credits ________ second
_213.30 _____ 76.22 _________ 0.3573
_185.23 _____ 64.75 _________ 0.3496
__90.63 _____ 14.20 _________ 0.1567
__91.98 _____ 14.20 _________ 0.1544
Average: _______________ ___ 0.2545

CPU:
__ CPU _____ claimed ______ credits per
__-time ____ credits ________ second
2 093.78 ____ 14.51 _________ 0.00693
2 094.41 ____ 14.48 _________ 0.00691
2 081.25 ____ 14.56 _________ 0.00700
2 272.09 ____ 16.62 _________ 0.00731
2 266.94 ____ 16.62 _________ 0.00733
Average: _______________ __ 0.007096

The GPU/CPU-ratio: 35.865!

The GPU is probably at least 35 times more effective in crunching seti@home than the CPU ...

Edit: Got in the original version of this post the wrong name of the card - corrected.

 

[theAnimal]

Pretty good results for a card that doesn't exist!

 

[petrusbroder]

Originally posted by: theAnimal
Pretty good results for a card that doesn't exist!

Yes, isn't it?

Corrected the post ... thanks, theAnimal! :thumbsup:

 

[PCTC2]

What CPU is running Seti that you're comparing it to? For the TeAm Seti Race I'm going to be running dual GTX 260 Core 216 SC's, an 8800GTS 512, and an 8800GT along with an i7 and a Q6600.

 

[Philippart]

The 9600GT is a decent GPU, I bought one 2 days ago because my HD3750 failed. Great boost in ppd in F@H!!
I really like the card: it has a nice and very silent (even at full load) cooler from gainward and crunches much faster than the HD3750 and much cooler than my 8800Ultra.

 

[Rattledagger]

Originally posted by: petrusbroder
I got myself a present today - a Nvidia 9600GT card.
I downloaded the latest drivers and set it to crunch seti@home.

Here are some results:

GPU:
__ CPU _____ claimed ______ credits per
__-time ____ credits ________ second
_213.30 _____ 76.22 _________ 0.3573
_185.23 _____ 64.75 _________ 0.3496
__90.63 _____ 14.20 _________ 0.1567
__91.98 _____ 14.20 _________ 0.1544
Average: _______________ ___ 0.2545

But, the problem is, the GPU only reports the CPU-time, but only uses maybe 5% of the cpu...

If you're going to compare the GPU against a CPU, you'll need to use the RUN-time for the GPU, not the CPU-time like you've obviously doing here...

 

[petrusbroder]

Originally posted by: Rattledagger

Originally posted by: petrusbroder
I got myself a present today - a Nvidia 9600GT card.
I downloaded the latest drivers and set it to crunch seti@home.

Here are some results:

GPU:
__ CPU _____ claimed ______ credits per
__-time ____ credits ________ second
_213.30 _____ 76.22 _________ 0.3573
_185.23 _____ 64.75 _________ 0.3496
__90.63 _____ 14.20 _________ 0.1567
__91.98 _____ 14.20 _________ 0.1544
Average: _______________ ___ 0.2545

But, the problem is, the GPU only reports the CPU-time, but only uses maybe 5% of the cpu...

If you're going to compare the GPU against a CPU, you'll need to use the RUN-time for the GPU, not the CPU-time like you've obviously doing here...

Hmmm I did not know that ... when I looked now I saw that the CPU-time did not change while the time until finished declined. Well, I need to find out how to measure the GPU-time. Any ideas (except looking at the logs...) how to do that?

 

[petrusbroder]

Originally posted by: PCTC2
What CPU is running Seti that you're comparing it to? For the TeAm Seti Race I'm going to be running dual GTX 260 Core 216 SC's, an 8800GTS 512, and an 8800GT along with an i7 and a Q6600.

Oh, it was a AMD 64 X2 4200+. All the times are from one and the same computer ...

 

[Rattledagger]

Originally posted by: petrusbroder
Hmmm I did not know that ... when I looked now I saw that the CPU-time did not change while the time until finished declined. Well, I need to find out how to measure the GPU-time. Any ideas (except looking at the logs...) how to do that?

I'm not aware of any other method than to look at the logs, and see the difference between start and stop-time for same wu...

But, with no nvidia I'm not bothered keeping-up with the various CUDA-discussions, so...

 

[petrusbroder]

@Rattledagger: LoL, yeah, that is true ... but since folding takes so much more GPU-power (som apps are slower than cold molasses) I want to use the GPU for something else than gaming and seti@home is one of my favorites.
But I'll sit down with my trusted old EMACS and look through the logfile and see if I can do some subtraction in time ...

 

[Assimilator1]

Petrus
I see you already had the answer to my question here , although you hadn't said what CPU it's running on .
Trying to remember the specs of an X2 4200,..........nope I'll need to read up on it.
2.2 GHz ,512KB L2, dual core of course(I knew that bit).

Any idea what the extra power draw running the 9600 @ ~100% uses?

 

[petrusbroder]

Not really, but a 400W PSU (Antec) is fully sufficient - no probs at all. The meter shows approx 300W (at full tilt) from the wall, OTOH, I have three disks in a RAID-configuration and two extra fans ...
The CPU is not OC'ed, and is a 95W Power rating. Does this help?

 

[ericlp]

Hmmmm, if you got a cheap AMD motherboard with two PCI slots you could buy a couple of cheap 9600 cards --- also 9800 cards are getting cheap these days and bigger rebates. Any idea what the speed difference between the two?

I am thinking for under 300 bucks you could get a really cheap low end AMD chip / Motherboard / two GPU cards and a Power Supply to have a really fast dedicated cruncher.

 

[Assimilator1]

I think you mean PCI-E

I still think of ATIs old cards when people say 9600 & 9800 , of course I know you mean the much newer Nvidia cards, stupid Nvidia naming .

Petrus
Just take a power reading with SETI running on the 9600 & then without .
On my Q6600 rig, the HD 4830 adds about 45-50w to total power draw when running F@H GPU (230v mains).
Oh btw total power draw is 275w running DPAD + F@H GPU (285w with F@H SMP), how come my rig uses less power than yours? lol. I do only have 2 HDDs but you would think my quad o/ced to 3.34 GHz would make up for that! .
You're on 220v mains aren't you?

 

[keeleysam]

I'm running SETI@home on my GTX 260 Core 216. It's working pretty well, though I'm glad I don't pay the power bill.

http://setiathome.berkeley.edu...ail.php?hostid=4548663

 

[petrusbroder]

Nice, very nice!

 

[Assimilator1]

Hey Petrus, did you see my questions?

 

[petrusbroder]

Sorry, Mark. I did see your questions and it took some hands on research to answer them properly:

1. Nvidia 9600GT uses approximately 79 W more when crunching seti@home compared to when it is idle (i.e. all the processes shut down). [4 measurements]
2. My system uses 294 - 298 W from the wall when crunching seti@home at full tilt: that includes the 3 HDD, the CPU HSF (the fan there takes some 12W) and two other fans (10W each) [4 measurements]
3. The CPU (AMD 64 X2 4200+) is a 95W part, similar to the Q6600.
4. Part of the difference is probably the PSU. It is an Antec 400W part, but it is approx 5 years old and has definitively not an efficiency of 85%, I would rathe guess an efficiency of 70 - 75%. The difference you can see between our comps is probably explained by the difference in the age and build of the PSU.

I hope this helps ...

 

[Assimilator1]

Thx for the info Petrus, if you were busy for a little while you should of just said, no probs anyway mate .

My PSU is also an Antec but a TPII 430, I think that's the generation after yours?, so yes presumeabley more effiecent. I think mine is 3-4yrs old.

The TDP for my Q6600 is 95w but that's for default speed & vcore, mine's running 940 MHz o/ced & vcore is set to 1.40625v (1.344v loaded) vs 1.2875v setting. So mine will be consuming much more power! .

Looks like your vid card plays a major part too in the power difference.
I've got 3x120mm fans, 2 case & 1 for the HSF.
So are you on 220v then?

1. Nvidia 9600GT uses approximately 79 W more when crunching seti@home compared to when it is idle (i.e. all the processes shut down). [4 measurements]

All processes?, did you mean your system uses 79w more when your GPU is crunching?

Awesome crunching power on your GPU anyway .

 

[petrusbroder]

Yeah - I shut down all BOINC and Seti processes. and then it uses 79W less compared to when it is crunching.
And yes, I am on 230V

Edit: The above statement is partly wrong: I shut off only the Seti-processes.
The CPU was running BOINC and Ibsercivis.

 

[Assimilator1]

You've mis-understood my question I think, I just wanted to know the difference between GPU crunching & GPU idle, if I've understood you correctly you've shut off CPU & GPU crunching so the 79w power drop is for both CPU & GPU going idle. Is that right?

 

[dajeepster]

has anyone figured out of the modified apps work for the gpu?...

willl cuda run well with the modified apps for the cpu running? or should i delete the modified aps when i'm running cuda?... any one got any ideas on this... i'm not getting the answers I want from my computers and want to know if anyone else has given this any thought.

 

[petrusbroder]

You can run the modified apps on the CPU and the standard CUDA-apps on the GPU, so just let them be and crunch on!
I do that on one comp and it works very well.

 

[petrusbroder]

Originally posted by: Assimilator1
You've mis-understood my question I think, I just wanted to know the difference between GPU crunching & GPU idle, if I've understood you correctly you've shut off CPU & GPU crunching so the 79w power drop is for both CPU & GPU going idle. Is that right?

You are not right. I was running BOINC Ibercivis on the CPU. Sorry for the misunderstanding in a previous post!

 

[Assimilator1]

Ah ok no probs :laugh:

So it seems the 9600GT is much thirstier than my HD 4830.