Hi,
If you are using Intel CPUs (especially Xeons with AVX512 support),
compiling RELION with Intel Compiler and linking with MKLFFT makes
motion correction 30-40 % faster than GCC + FFTW
(at least when I tested it in 2020).
Intel Compiler is freely available as a part of Intel One API HPC Tookit.
A CPU workstation with single EPYC7742 or dual Xeon Platinum 8280
can process 600-700 movies/hr (K3, non super-resolution).
The reason it does not support GPUs is that when I wrote the code
at LMB, CPU nodes were far less occupied than GPU nodes.
CPU nodes are also much cheaper to use in HPC centers at Japanese universities.
Now that SPA is no longer my main focus, it is unlikely
I implement GPU support in near future.
Best regards,
Takanori Nakane
On 2022/07/25 22:06, Ferdinand wrote:
> Dear CCPEM,
> do any plans exist to speed-up Relion's MotionCorrection in the future? At which point in the code could there be an optimization step / shortcut to increase the speed (throughput/minute?). The ever-increasing data collection speed of >800 movies/hour would really profit from such an optimization.
> Thanks
> Ferdinand
>
> ########################################################################
>
> To unsubscribe from the CCPEM list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCPEM&A=1
>
> This message was issued to members of www.jiscmail.ac.uk/CCPEM, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
########################################################################
To unsubscribe from the CCPEM list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCPEM&A=1
This message was issued to members of www.jiscmail.ac.uk/CCPEM, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
|