CPU usage in GPU apps
log in

Advanced search

Message boards : News : CPU usage in GPU apps

Author Message
Vlad
Project administrator
Project developer
Project tester
Project scientist
Help desk expert
Send message
Joined: 26 Oct 16
Posts: 231
Credit: 96,990
RAC: 24
Message 246 - Posted: 25 Jun 2017, 23:13:15 UTC
Last modified: 26 Jun 2017, 9:03:17 UTC

The new CUDA app 1.03 does not use the CPU core anymore except for the computation of the atomic ensemble in the beginning.
The new OpenCL app 1.03 has a command line option --nowait which when specified reduces the CPU load to zero but at the same time makes the system almost unusable due to lagging. If you do not use the system while crunching, you may add this to your app_config.xml:

<app_version> <app_name>xansons_gpu</app_name> <plan_class>opencl_ati_102_windows</plan_class> <cmdline>--nowait</cmdline> </app_version>

Replace 'opencl_ati_102_windows' with the appropriate plan class: opencl_ati_102_mac, opencl_ati_102_linux, opencl_intel_gpu_102_windows, opencl_intel_gpu_102_linux, opencl_intel_gpu_102_mac or opencl_nvidia_102_linux.

Update
All OpenCL 1.03 versions are marked as beta due to potential memory leak problem.


Update 2
The 1.03 app uses exactly the same amount of memory as 1.02 does, so there is no memory leak. The memory consumption depends on the WU. Those tasks which have 'solid_material' in their names may consume up to 1.5 GB of memory depending on the initial data. Please take this into consideration when launching multiple WUs in parallel.
I'll try to reduce memory consumption in the next version.

[VENETO] boboviz
Send message
Joined: 16 Nov 16
Posts: 37
Credit: 1,140,113
RAC: 1,088
Message 253 - Posted: 26 Jun 2017, 7:22:44 UTC - in response to Message 246.

The new OpenCL app 1.03 has a command line option --nowait which when specified reduces the CPU load to zero but at the same time makes the system almost unusable due to lagging.


I continue to prefer to dedicate a cpu core to my gpu.

Message boards : News : CPU usage in GPU apps


Main page · Your account · Message boards


© 2017 Vladislav Neverov (NRC 'Kurchatov institute'), Nikolay Khrapov (Institute for Information Transmission Problems of RAS)