F@H GPU Client for Nvidia is coming!

Quote:

---

After first getting information on NVIDIA's acquisition of RaySpace earlier today, we have some more fantastic news for NVIDIA GPU users and fans of the Folding@Home project.



Yes, it's finally coming - the NVIDIA GPU client we have all wanted since first seeing and tasting the power of the ATI GPU Folding@Home client.

Vijay Pande, the man behind the Folding@Home project, was on hand to demonstrate the first showing of the NVIDIA GPU client.



Apparently we aren't allowed to talk about specifics on performance of the client, mostly because the numbers we saw were based on an "upcoming NVIDIA GPU". I don't think I'll be spoiling anything by saying the new GPU was incredibly fast and the upcoming GPU will be faster than any Folding client today including the PS3; you will be impressed.

Regardless, the fact that NVIDIA is finally joining the Folding@Home community is great news - especially for our own PC Perspective Folding Frogs team! Vijay also discussed the creation of the OpenMM, a platform API that will be used to allow other molecular calculations be easily distributed to various computers and will have integrated support for both NVIDIA and AMD GPUs as well as modern CPUs.

If you want more information on the Folding@Home project, a stop by our Distributed Computing forum should have all the answers you need.

---

http://www.pcper.com/comments.php?nid=5679
:up:

Sounds promising, nvidia+ati folding= :)

Without reading the entire article, think it will support the 8 series?

:shrug:If they use CUDA to do it, then yes. If not, then maybe, maybe not... depends on the marketing!

Hmm Sexy, GT200 faster then a PS3.

Quote:

---

Originally Posted by [cTx]Warboy

Hmm Sexy, GT200 faster then a PS3.

---

G80 faster than a PS3...

Yay, more points for me with the use of my 8800 gts 512!

I have kind of a question here...

Is it possible to run CPU client+GFX client?

Quote:

---

Originally Posted by [cTx]Warboy

Hmm Sexy, GT200 faster then a PS3.

---

PS3 graphics is G70.

Woot, I cant wait to fold my 8800GT. :D

Quote:

---

Originally Posted by MuffinFlavored

PS3 graphics is G70.

---

But they ain't talking about the graphics, but the console as a whole.

Quote:

---

Originally Posted by [cTx]Warboy

But they ain't talking about the graphics, but the console as a whole.

---

g80 has more theoretical processing power all by itself though...

The PS3 F@H Client uses the Cell, wich is capable of 25 GFLOPs per SPE plus 25 GLOPs from the PPE, so about 200 GFLOPs.
A 8800GTX is capable of 345 GFLOPs.

So? I'm hoping the GT200 will produce the rumored 2TFLOPs

Doing all the FAH calculations "properly" will require nVidia to implement CUDA "properly" in their drivers. Place your bets on how long it will take for nVidia to write stable CUDA drivers for G80, G92, and GT200 in Vista x64...

You wanna know what I think? I think nVidia is going to make this GT2X0 exclusive... bastards.

They wouldn't dare. How many people are still running G80 after all this time? *raises hand* The worst they'd do is make it GT200-exclusive upon release, and then when complaints start shuffling in they'll rework the drivers for G80. Or some talented people will change a 0 to a 1 in the drivers' registry keys and enable it for G80/92.

Quote:

---

Originally Posted by [cTx]Warboy

But they ain't talking about the graphics, but the console as a whole.

---

no way the gt200 is faster then the CELL, the cell has like 8 daughter cores and 1 giant one that run at 3.2Ghz and a bunch of other stuff....it cant be that fast.

Quote:

---

Originally Posted by [cTx]Warboy

So? I'm hoping the GT200 will produce the rumored 2TFLOPs

---

Stock, ~1TFlop.

for every hour you fold on a GT200 a kitten dies from co2 emission :)

I guess I don't really care whether it is GT200 exclusive or not, at least they're making a client at all. However, it would of course be nice if they offered it for all their GPU architectures, or at least the new ones. My 7800GTX is G70 based, so I'd hope for at least G70 and up! :up:

Quote:

---

Originally Posted by LordEC911

Stock, ~1TFlop.

---

Stock... Who runs stock settings on XtremeSystems?... :shrug:

*folds more*
Soon we will have a cure for that if we fold more. :D

there might be a new form of cuda, cause the physx thing and this thing seems like it will only be on the new gpu's.

Quote:

---

Originally Posted by [cTx] Raptor22

You wanna know what I think? I think nVidia is going to make this GT2X0 exclusive... bastards.

---

If they have the ability to implement it in G80 (and it's effective/wortwhile PPD) and they make it GT200+ only for a sales pitch then it's a fair shame on them for using something as noble as folding to bolster their own pockets in such a manner.

On the other hand even in the worse case scenario, as above, it is still better than no client at all (depending on it's abilities and requirements).

Anyway, I'm looking forward to this client :welcome:

Try this

http://www.pcper.com/images/news/folding1.jpg and change numer 1 to 4 :) to see what happends

Quote:

---

Originally Posted by ExodusC

Stock... Who runs stock settings on XtremeSystems?... :shrug:

---

That was my point...

Quote:

---

Originally Posted by metro.cl

Try this

http://www.pcper.com/images/news/folding1.jpg and change numer 1 to 4 :) to see what happends

---

Nice catch!:up:

sweet i hope it supports the G80 cards so i can use my Ultra!!!

Quote:

---

Originally Posted by Monkeywoman

no way the gt200 is faster then the CELL, the cell has like 8 daughter cores and 1 giant one that run at 3.2Ghz and a bunch of other stuff....it cant be that fast.

---

Oooh yes, a graphics core with it's massive parallel design can indeed be faster than a cell processor. The "daughter cores" (SPE) and the controller core are sort of tuned risc processors, the last one being very near to a generic powerpc chip, nothing giant or powerful about that one. The whole thing is just 9 cores tuned to work well together, shrunk and at a higher clock speed. The thing with a Cell proc is that it is tuned to put out a high number of gflops at low power, while a gpu is tuned just tuned to max performance. And it's way more parallel then the Cell. So don't get me wrong, I kind of like the Cell design, but a GPU is just way faster than this. You're not saying that Nehalem with it's 8 cores will be faster than a GPU in specialized tasks, will you? Notice also that in the PS3 the Cell has has not always all it's eight SPE's, to increase yield IBM deactivates crappy cores on the die. But is for PS3 only, so Mainframes get only fully functional procs.

BTW folding on an oc'd GT200? Imagine the electricity bill!

Edit: That jump from 3870 to gt200 in the slide is just INSANE.

Why there are no other GPU enabled projects like WCG or QMC?

Nice catch metro!

Heres a imageshack mirror incase it goes down:
http://img86.imageshack.us/img86/4967/folding4wg0.jpg

x1900xtx was 2x more fast than a PS3.
3870 is about 384GF

4870 is 1TF ... what is insane with the GT200 with 500GF ?

Quote:

---

Originally Posted by ExodusC

Stock... Who runs stock settings on XtremeSystems?... :shrug:

---

What do you have to gain by running overclocked 24/7? Hardware is pointlessly fast as it is.

Quote:

---

Originally Posted by Monkeywoman

no way the gt200 is faster then the CELL, the cell has like 8 daughter cores and 1 giant one that run at 3.2Ghz and a bunch of other stuff....it cant be that fast.

---

No way a cheap crap console ~2006 would come even close to GT200.

The theoretical peak performance of the GT200 is above 1 TFlops.

Quote:

---

Originally Posted by Loque

for every hour you fold on a GT200 a kitten dies from co2 emission :)

---

:rofl::rofl:

Has there actually been any tangible results of all this folding thats been happening? Or is it just one big experiment?

Quote:

---

Originally Posted by The[R]eaper

sweet i hope it supports the G80 cards so i can use my Ultra!!!

---

Hope the same,ill be disappointed if its only for next gen.

Quote:

---

Originally Posted by Luka_Aveiro

I have kind of a question here...

Is it possible to run CPU client+GFX client?

---

The recent ATI GPU client runs on Windows only and uses nearly a full CPU core feeding the GPU with data. I let one SMP client run in parallel which uses the other 3 cores of the Phenom9850.

The high CPU load of the GPU client is one problem the F@H team is working on for quite some time. But when they release a new client with nVidia support - maybe they where able to minimize cpu utilization.

Quote:

---

Originally Posted by madcho

x1900xtx was 2x more fast than a PS3.
3870 is about 384GF

4870 is 1TF ... what is insane with the GT200 with 500GF ?

---

You're comparing peak performance with sustained. No way will a 4870 fold at 1 TF sustained.

What does Folding@Home means?:shrug::shrug:

protein-folding to make a cure for cancer etc.

Are the 8x00 series all ready for this??

Quote:

---

Originally Posted by DMH

Are the 8x00 series all ready for this??

---

They support CUDA so I hope so...

Please excuse my ignorance but what actually Folding@home stands for? what it is? and how can this benefits "us" nvidia gpu users?

At least now i will actually consider a Nvidia GPU. :up:

http://img144.imageshack.us/img144/9...ohofx57wu7.jpg


GTX280 will be very fast...

Is the legend for that red column "Radeon HD3870"?

Quote:

---

Originally Posted by Blacky

Please excuse my ignorance but what actually Folding@home stands for? what it is? and how can this benefits "us" nvidia gpu users?

---

http://folding.stanford.edu/

Like in all distributed computing projects the end "user" gets nothing. You basically donate computing power to the project.

Quote:

---

Originally Posted by Blacky

Please excuse my ignorance but what actually Folding@home stands for? what it is? and how can this benefits "us" nvidia gpu users?

---

No benefit because Nvidia users don't get cancer. At least I think that was their latest sales pitch...

If it's anything like what we studied in one of my math classes it's a program that simulates all kinds of ways DNA strands or other protein strands get themselves in knots. These knots or "misfolds" can cause all kinds of different diseases and this program will help us to understand it. Instead of having to use extremely expensive supercomputers they just get millions of people to donate their computing power. Every boost in speed like this is a nice boost in progress.

Quote:

---

Originally Posted by Blacky

Please excuse my ignorance but what actually Folding@home stands for? what it is? and how can this benefits "us" nvidia gpu users?

---

Folding @ Home is a distributed computing project dedicated to curing cancer and other diseases such as Parkinson's and Alzheimer's. For people that already "fold" this is great because now we can use the GPU power to get more points. :up:

Info: http://folding.stanford.edu/English/Science

How fast the client is doesn't matter too much, really. What's more important is:

a) The WUs themselves

b) CPU utilization

Current work units for the SMP client...well...suck. They take forever, are unstable, and are worth very few points. Also, the deadlines are shorter than they used to be. What I'm worried about is that the work units for the CUDA client will be large, slow, not worth much, crash constantly, and by the time you get done crashing and actually complete the unit, the deadline will be up. So after a day of percussive maintenance trying to get it to work, you will get exactly 0 points and Stanford can't use any of the WU you did calculate. That's what I'm most worried about.

Also, when the Radeon GPU2 client was first released, it would use up to a whole core on the CPU just to sort out the graphics card's own calculations...which meant anyone folding SMP on a dual or dual SMP on a quad was losing PPD due to a whole core gone. Stanford has since fixed that, but I hope that issue doesn't plague the CUDA client.

Quote:

---

Originally Posted by DerekFSE

Considering that http://www.pcper.com/images/news/folding1.jpg says that "70 million CUDA GPU's in the world....", I assume that means that all CUDA supporting GPU's will be able to use the client.

---

They also includes all the GPUs in Nvidia's inventory, AIB's inventory, etc.