CULA & CPU threads

[ink]

Hi there
It turns out that CULA forks out as many CPU threads as there are CPU cores available on the system. Although it's generally a good idea to use all available resources how does it know that the cores are not used by something else?
Oddly, this 'feature' goes unmentioned in the manual. The only thing I was able to find there is that "CULA uses the Intel® Math Kernel Library (MKL) internally."
The usual threading controls (OMP_NUM_THREADS, MKL_NUM_THREADS, and MKL_DOMAIN_NUM_THREADS) don't seem to work.
Is there a way to limit or control the number of threads?

thanks

[john]

Hi ink,
We are considering adding an interface function to CULA to control its number of software threads, based on your input. Can you provide a motivating code sample that shows degraded performance due to the CULA threading? These threads spend quite a bit of time idle and are not hard-bound to a given CPU core and so we wouldn't expect them to have a significant negative impact on overall program performance. A piece of sample code would help us understand the need for this; it can be provided by PM if you would prefer.

Best,
John

[hidra2000]

I've encountered the same problem myself, has there been any update on this issue?

Is there a way in R18 to limit the cpu threads per operation?