Palit GTX280 Reviewccokeman - July 28, 2008
» Discuss this article (5)
[email protected] is a distributed computing project run out of Stanford University. This project uses the spare CPU cycles (GPU Folding has been available on the Red side of the fence with ATI cards for a while), as well as GPU cycles to simulate the folding of proteins. When the proteins in our bodies misfold, things can go horribly wrong, and result in many diseases that are not yet curable. Examples include Alzheimer's, Mad Cow (BSE), CJD, ALS, Huntington's, Parkinson's disease and many Cancers and cancer-related syndromes. This project has been going on for some time now. With the performance increases in CPU and GPU computing technology seen in the last few years, the time needed to run the simulations has dramatically dropped. For more information on the [email protected] project, visit the [email protected] main page - and don't forget Team 12772 is the one you want to fold for! While monitoring the [email protected] client, I was amazed at the speed at which it completed the assignments. Initially, it was about 5 work units in less than two hours at stock clock speeds! Now that the client has matured some, the point values have increased as well as the time to complete the work units. Now about two work units in three hours is what the card can currently do while running the CPU SMP client. Running the SMP clients took about a day to process one+ unit with a quad-core CPU. There is definitely a substantial performance increase with the CUDA technology and the GTX280's 240 processor cores. Some things I found out while playing with the client - when the client is run in the viewer, the CPU usage skyrockets. As soon as it is minimized back to the system tray, the CPU usage drops dramatically. Because the resource demand to render the image back to the screen is high, the client's performance does decrease, but there's an easy fix - just minimize the client!
The processing power and processor design of the GTX200 series GPUs allows the video card to be used for things that people do not normally associate with GPUs. Using Nvidia's CUDA technology to harness this power, things like distributed computing and video transcoding can be accomplished in much less time than it would take a high-end CPU. The [email protected] client is just one of these examples. Elemental Technologies has a transcoding application called BadaBoom that harnesses the massive parallel computing potential of the GTX200 series GPU. CPU usage between the BadaBoom app and the one used for testing showed that CPU usage was fairly close, but the GPU-specific BadaBoom version did the work in less than half the time it took the CPU to complete the task.
Just to see how well this works, a sample film clip of 184MB in size was transcoded first with the CPU, and then again with the GPU - and the results were pretty astonishing. The measurement is in seconds, and best quality was selected. Hey, it really does work!