Random Musings About the Value of CPUs vs CUDA

Message boards : Number crunching : Random Musings About the Value of CPUs vs CUDA
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 8 · 9 · 10 · 11

AuthorMessage
Profile Francois Piednoel
Avatar

Send message
Joined: 14 Jun 00
Posts: 898
Credit: 5,969,361
RAC: 0
United States
Message 854063 - Posted: 16 Jan 2009, 4:16:10 UTC - in response to Message 853976.  

Keep trying ...
Show the best you can get ...
ID: 854063 · Report as offensive
Profile Voyager
Volunteer tester
Avatar

Send message
Joined: 2 Nov 99
Posts: 602
Credit: 3,264,813
RAC: 0
United States
Message 854099 - Posted: 16 Jan 2009, 6:36:28 UTC

That's all well and good, but you must have the APunits to crunch along with cuda.I haven't gotten any APs so with my cuda alone doing Mbs I figure a wall rac of ~2764.Thats just using one core to feed with one idle.It would be very hard to get enought APs to get high rac.If I put it on my quad I have 3 cores doing nothing..
ID: 854099 · Report as offensive
Profile mr.kjellen
Volunteer tester
Avatar

Send message
Joined: 4 Jan 01
Posts: 195
Credit: 71,324,196
RAC: 0
Sweden
Message 854104 - Posted: 16 Jan 2009, 7:01:33 UTC - in response to Message 853962.  


Top 20 hosts

Current host #14:

Intel(R) Core(TM)2 Quad CPU Q9450 @ 2.66GHz + 1 x GTX 280:

Avg. credit 8,594.86

Congrats to mr.kjellen! :-)

Ah, my mod installed on one of top 20 hosts *DANCE*
Leaving with this thought :D :D :D :D


Yep...

In addition, since Mr. Kjellen has been around and posting recently, it would be interesting to find out if he recalls what his 9450 was pulling before he went to CUDA Boost. ;-)

Alinator


At the time before I put CUDA on it wasn't on 24/7...more like running daytime...then RAC was at around 4000. It hit a high at around 6K when running flat out 24/7.

Too bad the AP Validator is acting up. I was hoping the RAC woul level out soon. Judging from a look at boinc stats, north of 10K.
/Anton
ID: 854104 · Report as offensive
Alinator
Volunteer tester

Send message
Joined: 19 Apr 05
Posts: 4178
Credit: 4,647,982
RAC: 0
United States
Message 854249 - Posted: 16 Jan 2009, 15:08:35 UTC - in response to Message 854104.  
Last modified: 16 Jan 2009, 15:18:17 UTC


Top 20 hosts

Current host #14:

Intel(R) Core(TM)2 Quad CPU Q9450 @ 2.66GHz + 1 x GTX 280:

Avg. credit 8,594.86

Congrats to mr.kjellen! :-)

Ah, my mod installed on one of top 20 hosts *DANCE*
Leaving with this thought :D :D :D :D


Yep...

In addition, since Mr. Kjellen has been around and posting recently, it would be interesting to find out if he recalls what his 9450 was pulling before he went to CUDA Boost. ;-)

Alinator


At the time before I put CUDA on it wasn't on 24/7...more like running daytime...then RAC was at around 4000. It hit a high at around 6K when running flat out 24/7.

Too bad the AP Validator is acting up. I was hoping the RAC woul level out soon. Judging from a look at boinc stats, north of 10K.
/Anton


Thanks for that intel, Anton. That's not too shabby a gain for bringing the GPU online. ;-)

One more question here, which should help put a lot of this to bed once and for all.

We can see accurate info for the CPU tasks (both AP and MB) and thus can calculate the CPCS(HR) for the host ourselves for them.

However, we can't do this for CUDA currently. So if you would be so kind, what would your 'best guess' estimate for the average Total Task Execution Time for CUDA. For this purpose, we will define this as the total Wall Time BOINC had the CUDA task executing for (don't forget it might get swapped out, thus start in one time slice and finish in another). We'll call it CPGS (for GPU, obviously).

Granted this puts the CPU at a slight disadvantage, since we are assuming CUDA always gets all the GPU all the time when it's running. Obviously, it can't, since the card might still get called upon once in while to actually perform it primary function. However, for hard core, dedicated crunchers, this is a reasonable assumption for this exercise.

Now contrary to what some would like us to believe, we can now evaluate the true measure of absolute sheer performance, the average per task CPCS. I was scanning over Mark's Frozen Nehi. Let's say, for the sake of argument, it's pulling in the low 60's for Credit per Hour per core (CPCS= ~.0175) overall. So there is a baseline to shoot for when comparing to the current CPU state of the art.

The simple truth is that RAC is nice to use for bragging purposes, and rough performance 'health' indicator for a host, but is close to worthless as a true performance metric.

@ Richard: I don't know if you where calculating this, but if you were, what did you get for an 'effective' CPGS on your test run?

Alinator
ID: 854249 · Report as offensive
Profile mr.kjellen
Volunteer tester
Avatar

Send message
Joined: 4 Jan 01
Posts: 195
Credit: 71,324,196
RAC: 0
Sweden
Message 854597 - Posted: 17 Jan 2009, 10:26:35 UTC

There is lots of manual work involved in investigations like tihs, and it would help a >LOT< if we had a value in the results describing GPU time...then it would be a trivial matter for Fred to bring out his DATAVAC excel file and examine the output.

I can say this however. The times I counted there were ~320-330 files reported during a 24 hour period.

I looked at my cache and got these timings for you for som different ARs

AR 3,3

01/17/09 09:49:06|SETI@home|Starting task 05no08ad.18070.11934.9.8.236_0
01/17/09 09:53:44|SETI@home|Computation for task 05no08ad.18070.11934.9.8.236_0 finished

AR 0.79

01/16/09 21:24:38|SETI@home|Starting 08no08aa.14123.1300.4.8.217_2
01/16/09 21:32:43|SETI@home|Computation for task 08no08aa.14123.1300.4.8.217_2 finished

AR 0,467

01/17/09 01:14:20|SETI@home|Starting task 14no08af.21130.9479.12.8.165_25
01/17/09 01:24:45|SETI@home|Computation for task 14no08af.21130.9479.12.8.165_2 finished


AR 0,378

01/17/09 06:20:12|SETI@home|Starting task 02no08ac.31941.23385.10.8.38_0
01/17/09 06:32:20|SETI@home|Computation for task 02no08ac.31941.23385.10.8.38_0 finished

AR 0,12

01/17/09 01:24:45|SETI@home|Starting 08no08ab.14807.78154.8.8.96_2
01/17/09 01:42:27|SETI@home|Computation for task 08no08ab.14807.78154.8.8.96_2 finished

Hope that helps...gotta run.
/Anton
ID: 854597 · Report as offensive
Previous · 1 . . . 8 · 9 · 10 · 11

Message boards : Number crunching : Random Musings About the Value of CPUs vs CUDA


 
©2022 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.