All of my crunchers run both GPUGrid and Rosetta@home. The configuration has always been to let Rosetta have 100% of the available CPU and this has never been an issue in the past. This situation changed last week.

On Windows 7, Rosetta appears to be more intrusive recently. GPUGrid work units were not completing as quickly as they did in the past on my GTX 580s. A quick investigation revealed the GPUs were starving for access to the CPUs. Limiting BOINC to 25% of the CPUs appears to have cleared the issue. The limit had to be changed on both Windows 7 cruncher.

This issue was not observed with Windows XP.

Was anything changed on GPUGrid to reduce the work unit thread priority? A similar question will be posted to the Rosetta forums.

---

Thx - Paul

Note: Please don't use driver version 295 or 296! Recommended versions are 266 - 285.

Hi Paul,

I just check the thread priority on an MJHARVEY type of WU on my Win7x64 box and it is running at 'Below Normal' which is the same as it has always been.

This particular WU type requires a full core all to itself but it reports in BOINC that it needs .552 of a CPU so BOINC does not count this as a full CPU usage and if I set CPU usage to 100% it would run 8 CPU tasks and 1 GPU task at the same time.

In the 8 + 1 scenario, eventhoguh the GPU task is running at a higher priority it will still have to contend with CPU resources being reallocated everytime the GPU needs it (switching context, flushing mem caches etc).

Hope this as useful,
Steve

At present there are a few CPU tasks about that seem to eat up a lot of resources, interfering with GPUGrid task performances. This has been seen running different GPUGrid task types.

On Linux some people are having to set the niceness manually, otherwise tasks take about twice as long, and this is after freeing a CPU core. It's always advantageous to have some headroom in the CPU/thread department, especially if the system sees normal usage.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help