Changing my OS reduced maximum number of threads for parallel processing

I recently changed my OS system from Fedora to Ubuntu 24.04. On Fedora I could setup a thread-based parallel pool with 10 threads no problem (parpool('threads', 10). When I try doing the same thing in Ubuntu, I get the following error:

'A minimum pool size of 10 was requested. The maximum thread-based pool size is currently 2.'

I've tried a couple of things with no luck

• maxNumCompThreads(10)
• I checked nproc to make sure that Ubuntu recognizes all the CPU's - it does
• reinstalled the parallel computing toolbox
• Varying the settings in cluster profile manager

Is there anything else I could try?

[Damian Pietrus]

Out of curiosity, which version of MATLAB are you running? What is the output of:

feature numcores

[Liam]

Using MATLAB R2024b

>> feature numcores

MATLAB detected: 10 physical cores.

MATLAB detected: 12 logical cores.

MATLAB was assigned: 12 logical cores by the OS.

MATLAB is using: 10 logical cores.

MATLAB is not using all logical cores because hyper-threading is enabled.

ans =

10

[Damian Pietrus]

Interesting. What model of CPU are you using -- does it have a mix of "performance" and "efficiency" cores? Do you have the same issue when using the Processes profile?

[Liam]

Intel® Core™ i7-1250U Processor : 2 Performance Cores and 10 Efficiency Cores

Which I'm guessing is the issue, but it seems weird that it works fine in Fedora

I get the same issue with Processes:

>> parpool("processes", 10)

Starting parallel pool (parpool) using the 'processes' profile ... Error using parpool (line 133)

Too many workers requested. The cluster "processes" has the NumWorkers property set to a maximum of 2 workers but 10 workers were requested. Either request a number of workers less than NumWorkers, or increase the value of the NumWorkers property for the cluster (up to a maximum of 512 for the Local cluster).

Answer 1

[Damian Pietrus]

Hey Liam,

That difference between performance and efficiency cores is exactly the reason you're getting this message. MATLAB processes should only be run on the performance cores to avoid resource issues, so Ubuntu in this case is doing the right thing. Do you happen to remember the specific version of Fedora you were using? It looks like it did not distinguish any difference between the core types. I'd like to pass this behavior along to development just so they're aware.