• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Using a GPU to cruch seti/fold

S

STEvil

2[H]4U
Joined
Oct 17, 2000
Messages
2,819
Does anyone know of a program that would allow us to do this? I have been wanting to at least give it a try.. could double as a benchmark :D
 
holy cow i was thinking of the same thing today. with graphics card gpu's hittng 1ghz and beyond, that would be pretty cool. imagine having a tri-processor box, 2 cheap athlons folding hard and a 6800u doing the same all under one case. interesting idea.
 
Yeah, this was discussed at the Folding-community forums... Stanford is keeping a close eye on general purpose computing on GPUs, but as of today there is nothing stable and convinient enough :)
 
There are seti@home command line clients for just about everything under the sun, how hard would it be to compile one for testing on GPU's?

With multiple video cards or multiple core videocards we may get some interesting results, too ;)
 
Think about it for a min. The video card is only designed to push data out the back end. You'd have to write the data to system memory over the AGP bus then have the CPU access that memory to write to the HDD and access the internet. That sounds like a classic recipe for a system protection fault to me. I'm not saying
that you might not be able to develope the software tools to do this, just that that's a whole lot of work to go through just for folding@home. If anything, I wouldn't put any work into it until PCI Express is firmed up, there's no use designing software tools for an interface that's becoming obsolete. Now if you could actually do this, it would get you a lot more than a few extra GHz.
Remember that GPU's are optimized for 3d rendering and with a client writen to take advantage of those optimizations it would fold proteins much faster than your CPU. Another thing, once you make that link from the internet to your GPU registers, it could make for some very inventive ways to cheat at online games. It would take some "mad haxor skillz", but those people do exist and most of them love online gaming. They wouldn't need to break the game rules, or alter the game server, they could borg your boxen and slightly shift your video output. All over a direct connection to a hapless competitor. They themselves wouldn't be cheating directly, just hexing everyone else in the game. In other words, if you were not up to snuf on your security you would be flat out owned when trying to game online. Furthermore it could also extend to the X-Box. Those people(the general X-Box user) tend to be clueless when it comes to security. Microsoft is a BIG
hacker target. It would really get them(MS) mad. Hum, could be interesting to watch!
 
Calm down :rolleyes:

If anything you said were remotely true our CPU's as they are now would be at risk of people attacking them..

That said, you can just use the onboard memory of the video card while Folding/crunching and send the data back to the hdd or system memory after its done. Its quite simple and there is tons of ram anyways since it only takes about 15mb to run seti.
 
I'm not getting excited at all, just musing on posibilities. I'm afraid you fail to undestand what a "system protection fault" is. No mater where the folding program runs, it will still have to write it's data to system memory in order to access networking hardware and do periodic writes to the hard drive. These functions will also have to be controlled by the CPU. While there is limited busmastering on AGP, I doubt that any GPU has the micro code to access system resource like disk drives, network adapters and other peripherals. While PCI bus mastering is far more versital, (PCI Express should be even more so), it still requires CPU assistance to use peripherals. So what you end up with is one "program" that is running as two seperate pieces of code, one on the GPU and one on the CPU. Which are very likely to end up trying to access the same region of system memory at the same time, ie; system protection fault. The only way I can see it working is if you are a damn slick coder or you have a northbridge designed to facilitate it. Now I've never wrote anything more complex than VB code and I'm not an intergrated circuit engineer, so, obviously, I could be wrong. But I've not seen anything to make me think so. That's just how it works to my understanding.
 
Of course, if one happens to be a frog, you'll probably disagree with anything I say. Like; the sky is blue, fire is hot, water is wet and the US should bomb Paris not Bagdad.
 
:rolleyes:

So, when a video card writes to system memory because it doesnt have enough AGP texture memory or cannot store all of the geometry data its fine, but when it does it outside of a game its not?

When your pagefile is in use because you dont have enough system memory free while gaming you are fine, but whe you are gaming you are fine?

:rolleyes:

Its not like you cant use the processor to do the writes and such either, seeing as its not going to do any writes more than once an hour (I do seti@home wu's in around 2 hours and a GPU shouldnt be overly tons faster).. its not like the GPU has to even interact with the HDD or system memory at all, actually, seeing as it can all be done onboard the video card and a program running on the CPU keeps track of when the WU is done then pulls the finished data across the AGP bus to system memory then writes to the HDD.. omg i've solved the problem!

:rolleyes:

Calm down.
 
making use of a GPU as a co-processor is announced in longhorn and next version of direct x
 
Would you really want to fold on what would be a very slow and expensive peice of hardware? Why fold on your $500 gpu when you could get the same results with a passive cooled Via cpu? I wouldn't want to run my expensive gpu 24/7 for such little output. The risk/reward is off balance here, it just wouldn't be worth it.
 
Originally posted by pageian
Would you really want to fold on what would be a very slow and expensive peice of hardware? Why fold on your $500 gpu when you could get the same results with a passive cooled Via cpu? I wouldn't want to run my expensive gpu 24/7 for such little output. The risk/reward is off balance here, it just wouldn't be worth it.
Click to expand...
What makes you think that a $500 GPU will fold slow? GPUs are designed specifically to push geometry data blindingly fast. Folding, by some weird coincidence, is all about geometry data too. So the potential is definitely there, but the appropriate tools aren't... For the time being.
 
I don't think there would be a problem with any system protection faults. The video card almost constantly sends data back to the CPU. 1 of the major benefits of PCIE is its fullduplex so more data can move between GPU and the CPU.

The problem with coprocessing will be the GPU itself. Although the GPU has more transistors and is fast with graphics, shading, ect thats what its designed to do. Even if your were to write the code to fold on it the results wouldn't be great.
 
What makes you think that a $500 GPU will fold slow? GPUs are designed specifically to push geometry data blindingly fast. Folding, by some weird coincidence, is all about geometry data too. So the potential is definitely there, but the appropriate tools aren't... For the time being.
Click to expand...
 
Just a small correction ... I may be mistaken but I suspect that
one of the older members at the [H], called PhyberOptik, coined
the original term 'Sir Soldering Iron' ... and he deserves some
'massiv respek' since he has sacrificed more of his hard earned
green to 'Sir Soldering Iron' than I care to recall! :D
 
Just think of a GPU as a retardely fast RISC processor that will cut down geometry calculations more efficently than any piece of hardware your gonna find in a desktop. There are little to no limitations with computers and software, it just takes time and dedication. I know little to nothing about any sort of programing but i wouldn't be surprised if someone does this, or at least for something other than gaming...hell i'm surprised Nvidia and ATI aren't pushing for this. Imagine what would happen to 500 dollar video card sales? probably fucking skyrocket because it wouldn't be a the niche market of just graphics.
 
Geometry? I thought the folding program was intiger based, not floating-point based. That is what i was thinking when we found out that AMD processors tend to fold faster than Intel processers.
 
Originally posted by Lord Hyperion
Just a small correction ... I may be mistaken but I suspect that
one of the older members at the [H], called PhyberOptik, coined
the original term 'Sir Soldering Iron' ... and he deserves some
'massiv respek' since he has sacrificed more of his hard earned
green to 'Sir Soldering Iron' than I care to recall! :D
Click to expand...

I have offered up another sacrifice to Sir Soldering iron - Count Radeon gave his life in the pursuit of victory for the Kingdom. The Black Tower and it's monarch, King Barton, is without a tactical advisor, currently.
 
If I'm not mistaken, and correct me if I am, but didn't Nvidia just release software that allows Quadro cards to process huge chunks of data blindingly fast and basically whores it out as a second processor?
 
Originally posted by Smoky
What makes you think that a $500 GPU will fold slow? GPUs are designed specifically to push geometry data blindingly fast. Folding, by some weird coincidence, is all about geometry data too. So the potential is definitely there, but the appropriate tools aren't... For the time being.
Click to expand...

Let me clarify. AMD Athlon XP 2400, $75. Latest and greatest gpu, $500. The XP 2400 would smoke a gpu in folding and I wouldn't be comfortable having my $500 gpu running full out 24/7 no matter how well the cooling solution was designed. If the Athlons fan failed you're out $75, if the gpu fan fails you're out $500. That's the risk I'm refering to.

I'm not a programmer so geometry data doesn't mean anything to me, are you refering to the Folding clients graphics? If so then consider that the text client, which I believe is the most used and most efficient does not utilize graphics. I agree with you that gpu's have folding potential but considering that you could build two or three descent folding machines for the price of the gpu it just doesn't seem worth it, even if you're going to own the gpu for gaming anyway.

An analogy would be - Would you risk driving your tricked out ricer in a Nascar event? The ricer wouldn't come close to winning but it would definately take a beating in the process.
 
1: We are not talking about graphics when we talk geometry. Geometry is floating point and fixed function calculation.

2: If your card is going to smoke during crunching/folding, its gonna do it while gaming.
 
You must log in or register to reply here.
Back
Top