Maximizing CPU tasks efficiency?
23 Nov 2014 1:44:58 UTC
Topic 197818
(moderation:
There is a SETI process that uses SSE2 (OpenCL) as a CPU task. It utilizes a cmnd.txt file to customize the parameters on how the CPU resources are allocated to more efficiently execute the tasks.
-use_sleep -unroll 10 -ffa_block 12288 -ffa_block_fetch 6144 -tune 1 64 8 1 -tune 2 128 8 1 -ocIFFT_plan 256 16 256 -hp
Does Einstein have anything similar to improve the running time for its CPU tasks, such as FGRP4-SSE2. If so, what are the parameters?
I don't buy computers, I build them!
Maximizing CPU tasks efficiency?
)
No.
Scanning for our furry friends since Jan 2002
All of my FGRP4-Beta tasks
)
All of my FGRP4-Beta tasks (v1.05) have been validated against an AEI machine running Linux and using SSE2.
e.g., Gamma-ray pulsar search #4 v1.04 (FGRP4-SSE2)
http://einsteinathome.org/workunit/206170424
It seems to help the efficiency a lot. Will that be coming to the Windows world?
FGRP4 v1.04 is SSE2 for all
)
FGRP4 v1.04 is SSE2 for all platforms according to the Applications page and according to Bernd's announcement of the 1.05 version it's supposed to have some more optimizations so should definitely have SSE2 even though it's not part of the filename or plan class.
OK, it looks like Linux is
)
OK, it looks like Linux is just more efficient than Windows. I am running them on 4 virtual cores of an i7-4771, which is otherwise lightly loaded. That answers another of my questions right there.
EDIT: Interestingly, on the S6Bucket Follow-up #1 v1.06 (SSE2) when my PC is paired against that same AEI machine, the times are almost identical. Maybe that is just because the science is different, or maybe they have worked to make Windows more efficient.
Or more likely, some the
)
Or more likely, some the compilers optimizer for Windows does better with one app and the one for Linux better with the other.