Folding to Cure on the Edge with GPUs (Page 1 of 4 )
Performance is the most critical factor in every area of computing. This is true for the "
Folding@Home" (aka FAH or F@H) project too. Speed is crucial; the faster proteins get folded, the more chances are left for research results that will lead toward finding cures for diseases, including cancer. Also F@H has become for most a competition to achieve higher ranks, whether as an individual or as a team. This motivates users to participate and contribute to the project.
Folding@Home is a distributed computing project that is quite popular. Distributed computing networks keep track of the contributions of their users: total number of WUs (working units) folded, total points, PPD (points per day), etc. This article's main purpose is to give in-depth information about "GPU Folding" since it has a lot of benefits. Also I will focus on techniques and tips to maximize your folding performance.
If you're not familiar with F@H yet then I recommend checking out the following list of articles: "Distributed Computing" by KaoMAN, "Building a Folding Farm" and "Customizing Your Folding Farm" by Dngrsone, and the "Folding@Home Distributed Computing" official website run by Stanford University. Also if the above information is not enough you might check out few of our DevHardware Forums' featured threads: "The Ultimate Folding Thread," "Folding FAQ," and "Folding Tweaks." Ultimately, you might want to find out how to join our DevFolding team; when you're ready, you'll want to read this.
GPU Folding in a Nutshell
Folding@Home on Graphical Processing Units has been possible since October 2006. Stanford's programmers successfully ported the protein folding client onto GPUs. This process took a lot of experimenting and hard work. Needless to say the results were fantastic. Statistics and studies showed that apparently around 20-40x more performance gain can be acquired folding with GPUs as compared to Intel Pentium 4 NetBurst-based processors.
This is possible because of their architecture. Graphical Processing Units are able to perform insane amounts of FLOPs (Floating Point Operations), but these FLOP calculations must be well-suited for GPUs (3D calculations) otherwise miscalculations and errors might occur, which aren't good at all for distributed computing.
The only negative is that currently only ATI video cards are supported, more precisely the x1600 (rv530), x1800 (r520) and x1900 Series (r580). We're enthusiastically waiting for the significant processing power of the r600s. You might be frustrated because you own an NVIDIA based video card but the reality is that currently folding on NV cards is impossible. Further research and experiments are being done on NV GPUs but apparently they can't offer the same level of efficiency of "folding processing power" as the ATI GPUs.
As a side note, keep in mind that at least 25% of your total system resources must be left available to the GPU client. That means you can't fold 100% on your CPU while folding on your GPU too. Folding simultaneously on both is possible by leaving the right amount of system resources free for the GPU client. It makes sense because your GPU must be "fed" with data and instructions by your CPU; if that's already running at its maximum performance level then bottlenecking occurs.
I'd advise folding on about 70% of your CPU's total power, leaving about 30% available for the GPU client—experiment and customize the percentages, see what works best for your system. With a dual core system I would leave one of the cores free.
Let's get into the depths of GPU folding—installation and folding.
Next: Installation and Folding >>
More Video Cards Articles
More By Barzan "Tony" Antal
| Recommended by Dev Hardware |
|---|
|
