Cure@PS3 project

[Carl B]

Is Cell's less accurate handling of rounding in single-precision FLOP calculations (not up to the IEEE spec) any kind of an issue?

Well, if it were, one would imagine that Stanford not be endorsing it, let alone formally announcing it's capabilities at roughly ~100GFlops per unit.

I understand where you're coming from of course, it's just that all indicators seem to point to 'all systems go' in terms of Cell's utility in the matter.

 

[Dave Baumann]

Honestly though, I wouldn't be surprised if the all-out burn of a high-end x1900 exceeds the maximum wattage supplying capability of the PS3's PSU outright.

You can only get a maximum of 150W to the graphics card.

 

[Carl B]

You can only get a maximum of 150W to the graphics card.

No you're right - I mean I knew the x1900xtx doesn't even draw ~130 watts at burn (not that that's not extremely hot!)... I was just playing off the idea of the 500w PSU requirements.

When the PS3 launches - and when the GPU folding app launches - we can hook up some wattage meters at the walls and see which is more efficient.

 

[Guden Oden]

Is Cell's less accurate handling of rounding in single-precision FLOP calculations (not up to the IEEE spec) any kind of an issue?

Since everybody involved in this is aware of PS3 Cell's deviation on this issue I expect them to program for it, and compensate any minor differences that may occur. I'd think the IEEE deviations to more affect portability of code rather than the end results... Besides, how much of protein folding is actually float calculations?

(Easy answer: 100Mflops/sec worth... )

 

[Carl B]

Guden you keep saying MegaFlops...

 

[Guden Oden]

FFS, yea seems I do, but as long as you know what I mean!

 

[ban25]

The screenshot indicates FAH running at 4.6s/frame on p2051. Can we get a performance comparison with various PCs? Right now I am getting 3m:4s/frame on a 3.2 GHz P4, but it's a different WU.

 

[Bohdy]

/me looks forwards to the kind of folding performance we will see with a 4-core Kenstfield or K8L armed with an R600.

*drool*

 

[Geo]

The screenshot indicates FAH running at 4.6s/frame on p2051. Can we get a performance comparison with various PCs? Right now I am getting 3m:4s/frame on a 3.2 GHz P4, but it's a different WU.

The GROMACS core I have going right now is 14m:8s/frame. This on an X2 4200@2750. Tho it should be remembered that FAH is not a multi-threaded app on the PC at this time. And the timing can change radically from core to core. GROMACS are harder and worth more. This one is p2408_ribo_alanine_side.

 

[ban25]

I just installed the console version on my home PC, though it doesn't seem to give me the same performance statistics as the GUI client. It's be interesting to try to get ahold of the same type of WU as shown in the screenshot for comparison purposes.

Are the GROMACS cores the larger 5MB ones? Though I've tinkered with F@H in the past, I never really paid much attention to it as I was more involved with the d.net projects.

Fun though, now I'm getting nostalgic for the days of the Infinite Monkeys RC5-56 project!

The GROMACS core I have going right now is 14m:8s/frame. This on an X2 4200@2750. Tho it should be remembered that FAH is not a multi-threaded app on the PC at this time. And the timing can change radically from core to core. GROMACS are harder and worth more. This one is p2408_ribo_alanine_side.

 

[Gradthrawn]

Plug on IGN

IGN now has a news article up regarding the Folding project. I thought the wording for some of the parts was quite humorous. Particularly the closing comment:

Just think, the next time your girl friend gets mad because you're sitting on the couch with the PS3, you'll be able to look her straight in the eye with your best Clooney impression, point to the TV, and say "Baby, I'm saving lives. What have you done today?"

I think if IGN continues to give its userbase exposure to the project, significant support could be garnered from that alone.

 

[mckmas8808]

Hey guys what is 0.000947 Teraflops? Is that 947 megaflops? Is megaflops even a computer term?
Anyway that's the average right now that CPU's that are using windows is acheiving. See HERE
And if the average CPU isn't doing 1 gigaflop, but the PS3 will do 100 gigaflops per wouldn't that mean that this Folding Stanford group should really really reach out to us PS3 buyers?
I mean if one PS3 = 100 PCs running windows then....

Also click below for nice CPU folding stats
http://fah-web.stanford.edu/cgi-bin/main.py?qtype=cpustats

 

[Carl B]

IGN now has a news article up regarding the Folding project. I thought the wording for some of the parts was quite humorous. Particularly the closing comment:



I think if IGN continues to give its userbase exposure to the project, significant support could be garnered from that alone.

IGN's a bunch of tools...

BUT, the more who know about this the better.

 

[mckmas8808]

IGN's a bunch of tools...

BUT, the more who know about this the better.

Hey what about my comment above yours?

 

[Carl B]

Hey what about my comment above yours?

Your comment is right on target. If the average PC yield at present is just under 1 GFlop, and Stanford is asserting a ~100GFlop yield on Cell/PS3, then basically upon PS3's launch, it will be the equivalent of roughly 100 PCs in terms of average work accomplished folding.

 

[mckmas8808]

Your comment is right on target. If the average PC yield at present is just under 1 GFlop, and Stanford is asserting a ~100GFlop yield on Cell/PS3, then basically upon PS3's launch, it will be the equivalent of roughly 100 PCs in terms of average work accomplished folding.

Okay I knew my math was right. So basically 2,000 PS3s will do the same amount of work that every single computer that is signed up now to date? They have close to 200,000 PCs signed up.

It's amazing that only 2,000 PS3s will double their data crunching. Also read this....

Some more cool facts.

About 200,000 volunteer computer users around the world crunch Pande's protein data. Leasing the equivalent amount of computing horsepower, he notes, would cost $1.75 billion a year at commercial rates. The flood of data generated by the volunteer computer users has resulted in 40 scientific papers.
"It allows us to do things that wouldn't be possible otherwise," he says.

Link
So about 200,000 PCs running at an average of 1 gigaflop per second saves Stanford $1.75 Billion dollars a year? I said "WOW"!
Now imagine 10,000 PS3's running at 100 gflops. That would save them around $10 Billion dollars a year if my math is correct. Now imagine only 2,000 PS3s saving them an additional $1.75 billion dollars.

 

[Carl B]

Link
So about 200,000 PCs running at an average of 1 gigaflop per second saves Stanford $1.75 Billion dollars a year? I said "WOW"!
Now imagine 10,000 PS3's running at 100 gflops. That would save them around $10 Billion dollars a year if my math is correct. Now imagine only 2,000 PS3s saving them an additional $1.75 billion dollars.

Well see, you're looking at it backwards though...

It's not that anyone 'saves' money per se - it's that this research simply wouldn't get done otherwise... or get done at a glacial pace. THAT's why the folding@home project is so impressive/vital/interesting. It's not that it's just some money-saver for xyz research org or university, it's that the funding for the conventional method simply doesn't exist.

Which is why Stanford claims that problems that would have been viewed as impossible to solve prior to F@H have already been solved due to it's collective efforts.

 

[Gradthrawn]

IGN's a bunch of tools...

BUT, the more who know about this the better.

Well put. Tools being the key word. We have to make sure we put the tools available to us to good use.

Your comment is right on target. If the average PC yield at present is just under 1 GFlop, and Stanford is asserting a ~100GFlop yield on Cell/PS3, then basically upon PS3's launch, it will be the equivalent of roughly 100 PCs in terms of average work accomplished folding.

Aha! The plot thickens! Your thinly veiled attempt to rally support for a "good cause" is actually a ruse to get PS3 users to boost your teams computational power and recorded contribution exponentially! You're just using us! We're another one of your “tools!â€￾

^_^

 

[Carl B]

LOL, no I just don't like IGN - they benefit from the fanatacism ******ism generates hit-wise, and as such they do nothing to stem it's tide; quite to the contrary in fact.

But that's just my take on the matter...

As for the response to Mckmas (and I do know you're just joking around btw), no honestly I thought PCs did more on average in terms of their output. But average is the key word, and there are a lot of 'ancient' chips slaving away there.

 

[Dave Baumann]

Guys, you are vastly underselling current processing. A 3.0HGz P4 is rated at 12GFLOPS of MADD capcability in the pipeline and F@H is also optimised for SSE extensions. AMD64's and Conroe's are going to be higher than this still. Multiply your PC CPU factors by a scale of at least 10x (and you'd still get cained by a PC with X1900 anyway! )