Folding Forum

Not sure if this has been posted before. Here is a talk by Vijay Pande about GPUs and Folding@Home.

http://nvidia.fullviewmedia.com/gtc2010 ... -2007.html

CUDA 2.2. has a new feature called "zero copy" memory access. It allows GT200 GPUs as well as MCP79 (?) integrated chipsets to access host memory directly over the PCI-e bus. This may in some cases be useful to speed up a computation because a some copying overhead from and to the GPU memo...

According to their Blog, they're already replacing their kludgy back-end software with some professionally developed code and they are asking the IT department to be put into a separate network because the firewall dealing with incoming traffic seemed to be overloaded handling both F@H and the rest ...

German c't magazine hat a pretty in depth article about the Large Hadron Collider computing grid (which is NOT to be confused with the LHC@home BOINC project). Unfortunately this article is in German.

It means 4.000.000.000.000.000 floating point operations per second actively contribute to folding protein molecules every single second.

Protein folding is an N-body problem. Wikipedia has more information on this type of mathematical problems. Basically you have bodies (atoms) with mass that interact through forces (e.g. electrostatic forces). Each atom interacts with all the others. So what they do is they simulate very small time ...

What's interesting is that the average GFLOPS/GPU is about the same for NVIDIA and ATI, slightly higher for ATI even. This clashes head-on with the points discrepancy between the two. What's going on? I have 5 nVidia GPUs, each of which is only contributing a few days per week. So this brings down ...

nVidia alone is about to eat PS3 for lunch! Tomorrow or the day after tomorrow guys.

I doubt it. Folding@Home is doing some very basic research, essentially trying to figure out how a protein folds and with what likelyhood the protein switch fromes one configuration to another. That's very basic research at this point, and it's done on a small selection of relatively simple and spec...

Hi, running the CPU or GPU clients on overclocked machines may lead to erroneous computations, sometimes to EUEs and loss of work units. What kind of safeguards are in place to ensure that results delivered by a client are accurate? For example BOINC uses two machines to compute the same result and ...

Would the use of double precision arithmetic actually improve the scientific return of any given F@H work unit - or is the science return currently limited more by inherent deficiencies or simplifications in the underlying physics model, rather than numeric precision? Are there plans to use double p...

Interestingly the numbers have dropped a bit post multi-GPU. Apparently people are now changing the configurations of the rigs, taking them offline temporarily.

looking into my glass sphere, I predict: By Wednesday, the GPUs will crack the 1000 TFlops mark. In 10 days, the GPUs beat the PS3s.

Once nVidia puts out WHQL certified drivers for all cards (that don't require any INI file hacks), I believe we will quickly see one Petaflop from GPUs alone... might take less than 10000 active GPUs to get there