OpenMM OpenCL (Core 16) available for ATI/AMD

By , the in Folding@Home Project - 1 Comment
With the release of the v7 client necessary to support this core a few days ago, it was only a matter of time: the Pande Group are rolling out Core 16 (OpenCL OpenMM) to open beta.

Firstly a warning: the beta tag is there for a reason, the core is not finished and may crash more often than usual or otherwise degrade system performance.

To receive these units you must have a v7 client set up in GPU mode with advmethods, a GPU and drivers with support for OpenCL (all 11.x series drivers contain the APP package necessary for this). Under these conditions, it is possible that you may receive the test project 11293, though the number of units available is limited for the first phase of the test and so many clients will be assigned Core 11 units instead.

This project simulates a complete amyloid beta peptide at high temperature. These simulations are very similar to those already performed using the CUDA OpenMM Core 15, and serve as scientific validation for Core 16. The project has the following characteristics (with the reference card being a GTX460 as with Core 15 projects):

  • 627 atoms (a relatively small protein)
  • Server 171.67.108.44
  • Preferred Deadline: 6 days
  • Final Deadline: 10 days
  • 1835 points

We tested the core and this project with a HD5770 (800 shader processors overclocked at 960MHz), powered by an Intel Core i7 920 overclocked to 3.5GHz running Windows XP with the 11.3 drivers package. With this setup we get 6650PPD with 100% GPU usage. This gives efficiency of about 70 PPD/W, which is almost twice what we achieved running this test with a Core 11 project. This is also comparable to the efficiency of the recently released nVidia GeForce GTX590.

On our test machine, we did not notice any degrading of performance, even when playing videos.

Unfortunately not all is well with this core. Firstly there is the high CPU and RAM usage in our test machine: Core 16 uses an entire thread at 100% and requires 375MB of system memory to operate. Hopefully these figures will improve as the core is optimised better, much as happened with the Core 11 variants in the past. The second point is more problematic: some users have reported crashes when folding in parallel with certain apps (notably Flash). These problems do however seem to be limited to cards based on the Cayman chip (HD6900). The developers have been able to reproduce the problem and are working on a fix.

Please let us know how the core is working for you...

Source: Official Forum