Opened 15 years ago

Closed 15 years ago

#805 closed Enhancement (fixed)

CUDA: Failure in CPU ressource allocation

Reported by: Qedinux Owned by: davea
Priority: Minor Milestone: Undetermined
Component: Client - Scheduler Policy Version: 6.4.5
Keywords: CUDA GPU Cc:

Description

Hi, I just tested CUDA capabilities of my computer.

  • O.S.: Windows XP 64 bit
  • CPU : AMD Athlon 64 Dual Core
  • GPU : nVidia GeForce? 8500 GT

First: The BOINC manager didn't detect the CUDA device when I managed this computer through Microsoft Remote Desktop.

Second: I essentially use BOINC for freerainbowtables project. Usually when running jobs from freerainbowtables project I got two tasks running, one on each core of my CPU. When I added the SETI@home project to test BOINC with CUDA, it stopped one of the two tasks and started running jobs from SETI@home. It allocated the second project to my second core. The problem is that this project doesn't use the core of my CPU but it uses my GPU. So this second core is unused.

Conclusion: Would it be possible to better allocate resources to projects. So in this case, I would like to allocate 100% of each core of CPU to freerainbowtables and 100% of my GPU to SETI@home instead of 100% of one core to freerainbowtables, the second core unused and 100% of GPU to SETI@home.

I hope my English isn't so bad.

Thanks

Change History (3)

comment:1 Changed 15 years ago by Qedinux

Sorry, I said something wrong.

BOINC manager doesn't choose on which core a job will be executed. In this case, BOINC manager allocates 50% of CPU to freerainbowtables and 50% of CPU to SETI@home which don't use. My conclusion is still running. Would it be possible to choose which project may use cores of CPU or GPU.

Qedinux

comment:2 Changed 15 years ago by Nicolas

Component: BOINC - APIClient - Scheduler Policy
Keywords: CPU allocation removed
Owner: changed from Bruce Allen to davea

comment:3 Changed 15 years ago by romw

Resolution: fixed
Status: newclosed

The first issue is a limitation in Windows in how Device Contexts are used. Nvidia doesn't believe it'll be fixed anytime soon.

Second issue should be fixed in the 6.6 recent builds.

Note: See TracTickets for help on using tickets.