Advanced search

Message boards : Multicore CPUs : Just joined. Numbers look good? Newbie questions...

Author Message
Profile Free
Avatar
Send message
Joined: 22 Aug 09
Posts: 14
Credit: 187,584
RAC: 0
Level

Scientific publications
wat
Message 12050 - Posted: 22 Aug 2009 | 23:34:51 UTC
Last modified: 23 Aug 2009 | 0:17:28 UTC

Hi,

I'm a long time SETI cruncher (same userid over there) that was bored the other night, so I decided to install Linux/Boinc on a PS3 for the heck of it. Originally was crunching SETI on the PS3, but the results weren't very impressive (~5.89 credits/hour). That was with the most optimized SETI (Multibeam) application for the PS3. There is also a SETI (Astropulse) application for the PS3, but it is v5.00 and outdated as v5.05 is now required for SETI AP.

Anyway, some of the folks over there said that the numbers look much better for GPUGrid on the PS3, so here I am.

Downloaded and installed the Boinc client off of this website, which I'm assuming is a custom Boinc just for this project and PS3.

I have Boinc/GPUGrid installed under a separate user than Boinc/SETI on my PS3, so the two don't conflict with one another. I have also stopped Boinc/SETI, so that only Boinc/GPUGrid is running. I'm running it manually for the time being (./boinc &, logout) until everything looks good, then I will set it up as a system daemon. I also have Linux running slim on the PS3. Runlevel is text. I do everything on it from my laptop (SSH).

Just wanted to make sure my numbers look on par for what to expect with GPUGrid on a PS3:

(Correctly identified on the computer account page:)
-----------------------
CPU type: PS3 Cell Broadband Engine [Model 0 ]
Number of processors: 2
Operating System: Linux 2.6.28-0.ydl61.1



(This I was wondering about, seemed low.)
(Edit add: Forgot that, unlike SETI, it would be really easy to find other PS3's to compare with on here! Numbers seem on par.)
---------------------------
Measured floating point speed: 395.04 million ops/sec
Measured integer speed: 1338.22 million ops/sec



(From BoincView.)
--------------------------------
FPU Speed: 395.04 MFlops
Total FPU Speed: 790.1 MFlops
Integer Speed: 1.338 GIOps
Total Integer Speed: 2.676 GIOps
Credit Per Hour: 7.22
(Assume this number will go up in time?)


Also, a newbie question:

- Should I continue using the GPUGrid application that Boinc automatically downloaded after I registered and attached to the GPUGrid project, or are there optimized GPUGrid applications out there like there are with SETI?

- Also, read that the GPUGrid runs only one WU at a time on the PS3, but uses the "whole cell processor and SPE's". SETI was crunching two WU's on the PS3 simultaneously. I know one of the "cores" in the PS3 is actually "virtual", but SETI was getting ~390 MFlops on both "cores" (two WU's) while GPUGrid is showing 395.04 MFlops for the "entire processor" (one WU). Not sure how these numbers add up. Seems really low (?).

Thanks!

Profile Free
Avatar
Send message
Joined: 22 Aug 09
Posts: 14
Credit: 187,584
RAC: 0
Level

Scientific publications
wat
Message 12094 - Posted: 25 Aug 2009 | 3:34:02 UTC
Last modified: 25 Aug 2009 | 3:34:23 UTC

Guess I was worried about things that didn't need to be. GPUGrid looks like it is running nicely on the PS3. RAC is climbing nicely.

I'm no longer worried or trying to figure out why some of the stats look wrong (i.e., "time left") when viewing the PS3 via BoincView or remotely via BoincManager on a computer. Just writing it off as GPUGrid running on a "weird" (PS3/Cell) architecture.

Can't complain about over 5,000 credits per WU!

:)
____________

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1893
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 12095 - Posted: 25 Aug 2009 | 7:34:28 UTC - in response to Message 12094.

The stats are wrong. If you are interested you should read the Cell processor architecture.

gdf

Jari Pyyluoma
Send message
Joined: 2 Aug 08
Posts: 12
Credit: 1,147,651,344
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 12111 - Posted: 25 Aug 2009 | 18:46:13 UTC - in response to Message 12094.

Yes, luckliy it is not the power pc core that is used for the project. It is really trimmed down.

Hope you'll enjoy this project. :-)

Watch out for unwanted upgrades to the xmb from sony. http://www.gpugrid.net/forum_thread.php?id=1297

Profile Free
Avatar
Send message
Joined: 22 Aug 09
Posts: 14
Credit: 187,584
RAC: 0
Level

Scientific publications
wat
Message 12171 - Posted: 28 Aug 2009 | 2:21:36 UTC - in response to Message 12111.
Last modified: 28 Aug 2009 | 2:23:17 UTC

Yes, luckliy it is not the power pc core that is used for the project. It is really trimmed down.

Hope you'll enjoy this project. :-)

Watch out for unwanted upgrades to the xmb from sony. http://www.gpugrid.net/forum_thread.php?id=1297



Thanks!

Reading about the biology science behind GPUGrid and watching the videos now. Cool stuff. Been thinking about switching entirely to one of the biology projects after hitting the 10 year mark in SETI (October 2009) for a while now. I mean, SETI is cool and all, but we have the rest of human existence to find those little green men -- which wouldn't be long without bio sciences. :-)

Post to thread

Message boards : Multicore CPUs : Just joined. Numbers look good? Newbie questions...