For what it's worth, I've also been playing around with a host a bit over the last few weeks.
When the GW tasks started appearing, I thought I noticed the Fermi tasks slow down substantially, and the GW tasks took an exceptionally long time. The change from 32 to 64 bit cut the GW times by about 20-30%, and AVX had no obvious effect.
Based on the hyperthreading discussion, I reduced the number of tasks to the number of cores, and established that 2×4 is well below 6 on this machine. My initial estimate was 4.5, but I didn't recalculate it when I had more data.
- From the "Tuning" run we have all the information that we needed, so we cut it short (canceled the workunits for which no tasks were sent out yet). The results that we needed urgently were set to a shorter deadline and were collected over the last weekend. The rest of the "O1AS20-100T" run was put on hold, no tasks are currently generated or sent out.
- We are currently starting the "actual" run that was announced a few minutes ago. It will show up as two different applications that will be sent to different groups of hosts (O1AS20-100F, think of "F" as "fast hosts", and O1AS20-100I).
- We intend to manually grant credit for all successful tasks from the tuning run "O1AS20-100T", in particular the "dangling" ones without "wingman" and the tasks from app versions < 1.04 that "lost" validation to results from app versions 1.04. However, we we will postpone this cleanup until we have the other two applications running as planned.
Hmm, I don't think that deciding based on CPU name is a good idea.
What about:
Xeon E5-2xxx
Xeon E5-4xxx
Xeon E7-xxxx
Xeon D-xxxx
Intel(R) Core(TM)2 Quad CPU Q9650
Core i3-6320 has 4M cache
etc...
Also, there are sample CPUs, which don't contain a branding string and still can be very powerful: "Genuine Intel(R) CPU @ 3.10GHz" - a Xeon sample.
For what it's worth, I've
)
For what it's worth, I've also been playing around with a host a bit over the last few weeks.
When the GW tasks started appearing, I thought I noticed the Fermi tasks slow down substantially, and the GW tasks took an exceptionally long time. The change from 32 to 64 bit cut the GW times by about 20-30%, and AVX had no obvious effect.
Based on the hyperthreading discussion, I reduced the number of tasks to the number of cores, and established that 2×4 is well below 6 on this machine. My initial estimate was 4.5, but I didn't recalculate it when I had more data.
I haven't done other tests, e.g. for cache.
I see another GW app run's
)
I see another GW app run's been started "O1AS20-100F" with a handful of tasks. Name's the same except for having an F instead of a T at the end.
Update: - From the
)
Update:
- From the "Tuning" run we have all the information that we needed, so we cut it short (canceled the workunits for which no tasks were sent out yet). The results that we needed urgently were set to a shorter deadline and were collected over the last weekend. The rest of the "O1AS20-100T" run was put on hold, no tasks are currently generated or sent out.
- We are currently starting the "actual" run that was announced a few minutes ago. It will show up as two different applications that will be sent to different groups of hosts (O1AS20-100F, think of "F" as "fast hosts", and O1AS20-100I).
- We intend to manually grant credit for all successful tasks from the tuning run "O1AS20-100T", in particular the "dangling" ones without "wingman" and the tasks from app versions < 1.04 that "lost" validation to results from app versions 1.04. However, we we will postpone this cleanup until we have the other two applications running as planned.
BM
BM
Do we have to select the F
)
Do we have to select the F version or is that done automatically by the server?
Congratulations on your
)
Congratulations on your efficient test! Great news that no more time for testing is required due to your great work! :)
RE: Do we have to select
)
This is done automatically on the server. No user intervention required.
Does the F version use AVX
)
Does the F version use AVX ?
Thanks.
Bill
I'm curious, what are the
)
I'm curious, what are the criteria when deciding what is a fast host ?
-----
RE: I'm curious, what are
)
See debug.
i[57][- ]| E[38]| E5-1| X[57]| G[238]| G64
host with more than 3m cache.
Hmm, I don't think that
)
Hmm, I don't think that deciding based on CPU name is a good idea.
What about:
Xeon E5-2xxx
Xeon E5-4xxx
Xeon E7-xxxx
Xeon D-xxxx
Intel(R) Core(TM)2 Quad CPU Q9650
Core i3-6320 has 4M cache
etc...
Also, there are sample CPUs, which don't contain a branding string and still can be very powerful: "Genuine Intel(R) CPU @ 3.10GHz" - a Xeon sample.
-----