Does any body know, if its possible to support other projekts by video card, eg. S@H and E@H?
bye
Note quite sure as i don't now anything about S@H or E@H but those video cards are not suiteable for the projects you've mentioned because of their precision... those cards are not able to provide. double precision calculations and do not confrom with IEE standards etc...
I've heard about these ideas several month ago, but what's the difference between the processing of S@H (heavy use of FFT) and the processing of F@H?
K&B
I guess the GPU's arn't IEEE conformant and though are not able to keep the necessary precision to calculate correct results.
As for S@H you can look at the codebase and most of the calculations are done in double precision which those GPUs arn't capable of or do not process them qicker that your cpu. Another thing to consider is to get the processed data back fronm the gpu... deliverving data fast to it is one thing but getting it back is another. But that changed with PCIe so that's maybe the reason while Folding is able to release the GPU app. now.
Anyhow i don't think it's possible to use GPUs without najor code changes on S@H or E@H... if it's possible at all.
But maybe Bernd could explain that further... I'm interested in that one too.
There's a dev kit (Compiler) to transition Code from CPU onto GPU's, but I don't know how far they are.
Last time I checked, they still had a healthy number of Limiting Factors and of course support only for a limited number of GPUs.
It should be possible to make good use of it for most Projects but I think it's quite alot of 'into-the-woods' intensive Code analysis which takes time (which few Project Staff have nowadays).
F@H announces support for ATI video cards, PS3 as well
)
Does anyone know, if its possible to support other projects by video card, eg. S@H and E@H?
bye
greetings KnB-Construction
RE: RE: http://folding.st
)
Note quite sure as i don't now anything about S@H or E@H but those video cards are not suiteable for the projects you've mentioned because of their precision... those cards are not able to provide. double precision calculations and do not confrom with IEE standards etc...
This idea comes up from time
)
This idea comes up from time to time. Here is the last thread on this (from June).
BM
BM
I've heard about these ideas
)
I've heard about these ideas several month ago, but what's the difference between the processing of S@H (heavy use of FFT) and the processing of F@H?
K&B
greetings KnB-Construction
RE: I've heard about these
)
I guess the GPU's arn't IEEE conformant and though are not able to keep the necessary precision to calculate correct results.
As for S@H you can look at the codebase and most of the calculations are done in double precision which those GPUs arn't capable of or do not process them qicker that your cpu. Another thing to consider is to get the processed data back fronm the gpu... deliverving data fast to it is one thing but getting it back is another. But that changed with PCIe so that's maybe the reason while Folding is able to release the GPU app. now.
Anyhow i don't think it's possible to use GPUs without najor code changes on S@H or E@H... if it's possible at all.
But maybe Bernd could explain that further... I'm interested in that one too.
There's a dev kit (Compiler)
)
There's a dev kit (Compiler) to transition Code from CPU onto GPU's, but I don't know how far they are.
Last time I checked, they still had a healthy number of Limiting Factors and of course support only for a limited number of GPUs.
It should be possible to make good use of it for most Projects but I think it's quite alot of 'into-the-woods' intensive Code analysis which takes time (which few Project Staff have nowadays).