PS3 Kicks F@H over 500 TFLOPS!

[mr.tie]

http://gizmodo.com/gadgets/home-entertainment/breaking-ps3-folding-ps3-triples-folding-at-homes-computing-power-to-over-500-tflopspflops-in-spitting-range-246664.php

Good god, Well it looks like the big shiny retard is useful for something, kinda make me feel a bit bad I sold mine for a profit....nah :toast:

Anyway cheers to you guys out therr crunching away on these babies, you're all doing one hell of bang up job.

[TopherTony]

Wow 13000 ps3 have double the computing power of 160000 pcs? that sounds like a bargain to buy a ps3 for folding instead of purchasing a new pc

[found404]

I just saw the stats page, currently 14971 PS3s = 367TFlops, pretty incredible.

[angra]

can I curse the lack of a linux client again? :)

I am planning to add a bunch of PS3 to my cluster, but they have to run linux. yarg.

[angra]

OK so I installed this on my lone (for the moment PS3).

Couple of questions:

Does anyone know yet how many points project 3127 gets?

And, does anyone know how to configure it to run automatically when the PS3 is idle?

[sc00p]

OK so I installed this on my lone (for the moment PS3).

Does anyone know yet how many points project 3127 gets?http://fah-web.stanford.edu/psummary.html ;) ...
...so it's 287 points...

[angra]

yikes.

for being 30x more output than a x86, it sure is getting the same # of ppd (about) as each of my dual opteron nodes. geeze. They need to adjust the points.

[sc00p]

yikes.

for being 30x more output than a x86, it sure is getting the same # of ppd (about) as each of my dual opteron nodes. geeze. They need to adjust the points.yes it seems to be a little wrinkled considering the bandwidth.

[Bloody_Sorcerer]

Despite the PS3's client apparent asswhoopedry, GPUs are still kicking out more TFLops per client; theres just fewer running.

[NickS]

PS3 could be the F@H crunchers dream. Small compared to a PC, minimalistic, a lot cheaper than a PC of even 1/2 the F@H power.. etc.

w00t :)

[angra]

I guess I hadn't run the numbers but I'm not surprised to hear that. The most recent generation of ATI GPUs have more raw FLOPS capability available in the fragment processors than the cell does. I kind of figured that the more controllable execution model on the cell would make up the difference...

[Kunaak]

so whos gonna get the first PS3 farm going??

[Vapor]

so whos gonna get the first PS3 farm going??Shouldn't be hard...availability is pretty good.

Would do a ton of production too :D

[Kunaak]

and the 20 gig versions are real cheap on ebay these days too.

[ColonelCain]

I just wonder how much raw power the R600 will have...:slobber:

[jimwah]

The thing is, I don't see the ppd being too great on these, or being raised in the near future, unfortunately. Sure Pande wants PS3's folding, but at the cost of alienating folders without PS3's? The debate raged on about this with the GPU client, and the SMP client. Also, all of these different clients work on valid science, some work is best suited to a PC folding the vanilla client, is that really worth less than someone working 30x faster on a PS3 unit?

It gets complicated & some people will be unhappy whatever the ppd. I need to see some hard figures on PS3 ppd, but at the moment I'd still buy a cheap SMP folder for folding 24/7.

[sc00p]

I agree it's likely time that this thread has run its course. I think the fundamental misunderstanding is that FAH doesn't do just one type of calculation, but does many, many types, and not all types are sped up on all platofrms. What makes the GPU clients and PS3 clients so fast is not universal. They do certain types of calculations (implicit solvent) very quickly (20x to 40x faster), but can't do other types of calculations at all (eg explicit solvent, where the SMP client is kicking ass).

We have been pushing multiple technologies in order to take advantage of the best of what's available, using each technology where it's most useful. If we only had GPU's, FAH would fail, as we couldn't do important explicit solvent simulations (eg important for our drug design efforts and folding efforts). If we only had PC clients, we would lose out on what we could do with GPU's.

FLOPS can look impressive, but there's lots more to the story than just how many FLOPS a client can do. In terms of scientific results, I think our papers, results, and awards give some sense of what we've been able to do and the impact we've had so far, but I'm always most excited about where we're going, not just what we've done so far. The future is looking particularly exciting!
OK...
...as I can't undestand the project as a whole...
...I'm just gaining (-ed) back my confidence (if that ever leaved ;)) about how projects/wus should be credited. I believe Stanford does think through how to guide "the customers".

Cpus are still very important tools to simulate more complex "things". Cpus are able utilize 7 (I think) different types of cores... each different types of different calculations.

If "customers" (=folders) were to stare too much about FLOPS and credited generously... I think Stanford could be afraid... afraid such farms like of PS3 could happen. That might not be most suitable for the science (F@H generally).
End of the line... if someone has already a PS3-> put on F@H. If want to buy a gaming console-> buy PS3&put on F@H (<-that is if one is Dedicated @ F@H enough) and believe great games are on theirs way.

And had read somewhere iirc... that effects crediting: PS3 are not to expected to become/be dedicated folding machines (<- crazy idea ;)) ... Stanford thinks they are bought for gaming and that's another reason for not yielding such a PPD someone could draw from FLOPS figures.

So must be real PITA making decisions ~credit for each wus. (like p2126 for example cause those wus were important to fold also :)).

Just my thoughts...

...yet it would be great if a Cell-fanboy comes out of closet with this (http://www-03.ibm.com/press/us/en/pressrelease/20210.wss) :D

of I had to go now... had more write, but... later

[angra]

Well, Dr. Pande is definitely right that flops aren't everything. They need to value the points in the way that maximizes the science output for the project of the whole, so it's all good. It's just disappointing a little when Stanford brags to the world about all the flops they are getting, but doesn't proportionally "reward" the donors.

[angra]

Shouldn't be hard...availability is pretty good.

Would do a ton of production too :D

I have plans in the works...unfortunately they will be running linux so no folding...

[Scimitar]

In my opinion, if they lowered all of the SMP work units to half of their current points, it would go a long way towards fixing things. Then the PS3 at 900 to 1,200 ppd and an X1950XTX at 700+ points per day don't look so bad.

[sc00p]

2ch@PS3 (http://folding.extremeoverclocking.com/team_summary.php?s=&t=54733) -team is flying :slobber:

[embeejay]

well they have over a 1000 teammembers, so no wonder :P

[embeejay]

I have plans in the works...unfortunately they will be running linux so no folding...

Why no folding on linux? ;)

[[XC] moddolicous]

I don't believe that the linux (yellowdog, right) uses the cell cpus to their full power. Is it true that there are only 7 cores running out of the 8 because one has to be used to do other tasks on the PS3?

[eXceededgoku]

one is disabled to increase yields and another one is used to do the tasks, essentially there is 1 CPU with only 6 SPEs left to do the work.

[angra]

F@H doesn't, and apparently won't have a Linux PS3 client.

[embeejay]

F@H doesn't, and apparently won't have a Linux PS3 client.

Ahh the linux client is x86 only, i missed that part :/

[[XC] moddolicous]

Ahh the linux client is x86 only, i missed that part :/
No technically. The SMP client is 64-bit (which raises the question Why isn't there a single core 64-bit client?).

[embeejay]

moddolicous;2095017']No technically. The SMP client is 64-bit (which raises the question Why isn't there a single core 64-bit client?).

It is still x86, even tho it is for the 64bit-extended version of x86 ;)