Einstein@Home on a PS3
16 Jan 2007 20:36:53 UTC
Topic 192313
(moderation:
I got a PlayStation 3 for the holidays and I recently installed Yellow Dog Linux 5.0 on it.
As you probably know, the PS3 has this sophisticated Cell processor which has some serious potential, even for scientific computing. The Cell has a main dual core CPU (called the PPE) which is very much like the G5 and then 8 special purpose compute engines called SPE's that can be used in parallel and controlled by the PPE.
I just built Einstein@Home on the PS3 using IBM's XLC compiler. I used the PowerPC specific code which includes AltiVec instructions. However, for now the application runs only on the PPE, i.e. it ignores the SPE's completely! So, I'm only using a fraction of the PS3's available power. Check it out:
http://einsteinathome.org/host/833948
As you can probably observe, the PS3 is performing comparably to a low-end G5 / high-end G4. That sounds poor, but again remember we're only using the PPE of the Cell processor here!
If you're interested in more scientific benchmarks on the PS3, check these out:
http://www.macresearch.org/science_on_a_g5_ps3_macpro_benchmarks
http://www.geekpatrol.ca/2006/11/playstation-3-performance
http://folding.stanford.edu/FAQ-PS3.html
Einstein@Home on a PS3
)
yeah ps3 has a real potional becuse my ps3 is on 24/7. doing prety much nothing. expect for few hours when iam gaming. i would be happy to for the ps3 to crunch. and a few of my friends would like to as well. enistein@home. should really try alot harder to get this done
RE: I got a PlayStation 3
)
Do you care to share the binary file of the compiled Einstein app?
I've got a PS3, built Gentoo on it, and would love to try to crunch some workunits on it...
Mark
I thought you're not allowed
)
I thought you're not allowed to use anythig other than the official e@home clients because of issues regarding scientific validity?
RE: I thought you're not
)
exactly that why E@home should make the clents. like folding did
Hi Well, I'm going to
)
Hi
Well, I'm going to work on trying to convince the E@H team to take the PS3's potential seriously. However, that may take some time, because they have many other things to work on, in the immediate future. Also, I'll try to spend some time to see if I can use the SPE's of the Cell too, to improve the performance the E@H application on the PS3.
Thanks.
All: Gaurav is closely
)
All: Gaurav is closely associated with the Einstein@Home developers and made the first AltiVec code we used on Einstein@Home. He has access to the source code.
Gaurav: Thanks for doing this. It would be nice if you could find out how to make use of the SPEs in general. But note that we will not use the code you are familiar with beyond S5RI, i.e. in S5R2, so work specific to this code (i.e. the "hot loop") might be wasted. After having [edit]S5R2[/edit] running I would be happy to spend some time with you porting the new code to the PS3.
BM
BM
Be aware though that the ALUs
)
Be aware though that the ALUs of the current Cell Processor are single-precision only at full speed.
At IEEE standard 80bit FPU precision, their combined performance falls slightly behind last Generation x86 CPUs and is significantly superseeded by current x86 class CPUs.
I think the Cell Compiler Devs are still working on faster workarounds for the lack of native double-precision FPU, but whatever tweaks they'll apply... It will still be only limited usable in terms of performance.
The next Cell design is supposed to fix that shortcoming.
Some peoples has tried to
)
Some peoples has tried to compile SETI on the PS3. The performance was comparable with a 1 GHz Pentium system. The problem was, that the design (paralisation) of the PS3 could not be used, and the PS3 only uses the core CPU what gave a bad performance.
An update on using the PS3
)
An update on using the PS3 for E@H. I spent some time learning the very basics of SPE programming and ran some tests on a single SPE of my PS3. Crudely, I expect that for purposes of E@H, the performance of a single SPE will be in the same ballpark as that of a high-end G5 processor. This is quite impressive, especially if you note that the PS3's Cell has 6 of these available for crunching ..
PS3 use for BOINC projects is
)
PS3 use for BOINC projects is just around the corner.
According to the german Heise Newsservice ( http://www.heise.de/newsticker/meldung/87218), the new Firmware 1.6 is (among other things) specifically addressing the use of PS3s for Folding@Home. Scientist there hope to double their overall performance through the use of volunteer PS3s. According to the quoted source, a PS3 with this firmware would be equivalent to a 100 GFlops PC, (at the expense of consuming 180 Watts). Not too bad, but the article also stresses that this performance figure suggests the use of single-precision floating point arithmetic, something that is not good enought for many BOINC science projects.
Sony Press release : http://www.scei.co.jp/folding/en/
Folding@Home on PS3: http://folding.stanford.edu/FAQ-PS3.html
Cheers
BRM