Page 2 of 2
Re: Rejecting a4 core WUs
Posted: Tue Feb 26, 2019 6:48 pm
by bruce
JimboPalmer wrote:F@H does not allow us to choose between a4 and a7, (a much newer core) because if they did, almost everyone would hog a7 as they get more points.
Not quite true. FAHCore_a7 comes it two flavors. If your hardware supports AVX, then FAHCore_a7 will use that hardware and complete the assignment faster, earming more points. If you hardware supports only SSE2, FAHCore_a7 will load a version that's equivalent to A4 and there should be no difference between the SSE2 performance of a4 and that version of a7.
Re: Rejecting a4 core WUs
Posted: Tue Feb 26, 2019 9:09 pm
by toTOW
Actually, A7 is a bit faster than A4 even on SSE2 hardware ...
Re: Rejecting a4 core WUs
Posted: Wed Feb 27, 2019 12:27 am
by bruce
toTOW wrote:Actually, A7 is a bit faster than A4 even on SSE2 hardware ...
Could be... the A7 core is based on a later version of GROMACS than the A4 core, so they could have improved SSE2 throughput.
Re: Rejecting a4 core WUs
Posted: Wed Feb 27, 2019 1:37 am
by JimboPalmer
It certainly gets more PPD, even on my ancient CPUs (Core Duos)
Re: Rejecting a4 core WUs
Posted: Wed Feb 27, 2019 3:11 am
by Joe_H
When the A7 core was first beta tested, a project that had been done before using the A4 core was duplicated. The WU's processed by the A7 core finished about 30% faster than they had on the A4 project. My old Core 2 Duo MacBook had taken just under 24 hours to finish a WU using the A4 core, the timeout was 24 hours. With the A7 core a WU would finish in under 20 hours.