multithreading doesn't work with AMD CPU when estimating ICLV model
Posted: 29 Sep 2021, 09:20
Dear Apollo team,
Thank you for producing this wonderful library!
Please forgive my poor writing, I am not from a English speaking country.
I am encountering a strange problem during the estimation of ICLV models.
I have 2 computers, one laptop with Inter i7-1065G7, and a desktop with AMD Ryzen 5600x. Both of them are using R4.1.1 on windows 10, with Apollo version 0.2.5. I found that Apollo on my AMD platform estimating the ICLV model far slower than my Inter Laptop, which is theoretically unlikely.
To ensure the problem doesn't come from my coding, I run example 24 on Apollo homepage, only changing the nCores in apollo_control to enable multi-threading. During the estimation, I can tell from the windows Task manager that the Intel CPU utilization is always at 100% for all 8 cores, while the AMD CPU utilization is only around 2%-10% for all 12 cores. None of the AMD cores get fully used.
By the way I also do the same test based on Apollo example 16, and the multi-threading works perfectly on both CPU platforms... this completely confused me.
Is this a bug? Or did I do something wrong in setting my environment for the AMD platform?
Any feedback would be appreciated.
Best Regards,
Gao
Thank you for producing this wonderful library!
Please forgive my poor writing, I am not from a English speaking country.
I am encountering a strange problem during the estimation of ICLV models.
I have 2 computers, one laptop with Inter i7-1065G7, and a desktop with AMD Ryzen 5600x. Both of them are using R4.1.1 on windows 10, with Apollo version 0.2.5. I found that Apollo on my AMD platform estimating the ICLV model far slower than my Inter Laptop, which is theoretically unlikely.
To ensure the problem doesn't come from my coding, I run example 24 on Apollo homepage, only changing the nCores in apollo_control to enable multi-threading. During the estimation, I can tell from the windows Task manager that the Intel CPU utilization is always at 100% for all 8 cores, while the AMD CPU utilization is only around 2%-10% for all 12 cores. None of the AMD cores get fully used.
By the way I also do the same test based on Apollo example 16, and the multi-threading works perfectly on both CPU platforms... this completely confused me.
Is this a bug? Or did I do something wrong in setting my environment for the AMD platform?
Any feedback would be appreciated.
Best Regards,
Gao