Parallel Computing with hyper-threading

I run some of my programs via the parallel computing toolbox using a reasonably powerful machine (20 Cores) to accomplish long running jobs (20 Min - 24 hours). It has recently been brought to my attention that since Matlab doesn't actually make use of the hyper-threaded cores, that the processing may see performance improvement by turning this off in the BIOS. I turned off the hyper-threading and then ran a speed test utilizing all 20 cores on the machine. For some reason, even though I was utilizing a higher percentage of overall processing power available on my machine, the jobs actually ran consistently slower when the hyper-threading was turned off.

Can someone help me to understand the benefits or drawbacks of turning hyper-threading on/off. Should I see performance improvement from turning this functionality off? If so, why do my results differ from my expectation?

Thanks in advance for your help, Brian

[Thomas Ibbotson]

As long as you're not actually trying to use of the virtual hyper-threaded cores, I wouldn't expect turning on and off hyper-threading in BIOS to make a big difference. However, I haven't done the test myself.

What made you decide you were using a higher percentage of overall processing power? I ask this because if you have 20 physical cores, and they are hyper-threaded to 40 virtual cores, but you only use 20, then there is not much difference between this and turning hyper-threading off and still using 20 cores.

If you want more information, perhaps the discussion in this thread about hyperthreading and PCT might help.