Speaking about AVX, I'd like to advise to pay more attention to CPU temperatures. Some CPUs have a sort of erratum, which can cause them to exceed the designed TDP when running certain AVX workloads.
This is also the reason why later CPU generations run AVX code at lower frequencies (usually by a few 100 MHz) than the original maximum non-turbo clock.
I just downloaded an "01". It was promoted to high priority while another job was placed in a "wait" state. Are these jobs being elevated to a higher priority? I do not have a concern but just curious.
It's tough to make a quantitative comparison between SETI and Einstein, but the AVX optimized apps for SETI seemed to produce significantly better results than the plain v7 app, but kind of scattershot compared with the SSE4.1 and SSSE3 optimized apps.
But I did not know about the AVX throttling issue one of the other posters mentioned, so maybe that's why I didn't see the difference among optimized versions.
In this earlier message, I mentioned that (for a little while at least) there will be two beta test 'CPU only' runs being conducted. This means that people enabling beta test work in order to get the new GW tasks will also be getting FGRPB1 whilst it remains in beta test.
I'd been looking at this just from a CPU tasks perspective so it didn't initially cross my mind that there is also a GPU app in beta test - the BRP6 cuda55 app. With all the excitement and interest generated by the amazing news of today's press conference, I suspect there may be lots of people wanting to get a piece of the action here, at least until the gloss wears off :-). Perhaps there will be a lot of experienced people who participate at other projects who decide to shift their CPU crunching here for a while to join in the fun.
My concern is that such people wanting CPU work for the new GW run only, may get a surprise to find both FGRPB1 and BRP6-cuda55 tasks (if they have an NVIDIA GPU) which they weren't expecting. If they had deliberately selected only O1AS20-100T, they might be st least mystified and possibly upset by getting all three.
I just intend to point any people questioning this to the earlier explanation about beta tests, via the above link. If the Devs have any other suggestions, it would be good to hear them.
For my own fleet, I regard them as 'production' machines so I'm taking the opportunity, presented by yesterday's false start, to stock up with 'production' tasks, so that I can set NNT when the O1AS20 run recommences. I'm setting up an extra machine or three (no GPU) to participate in the O1AS20 beta test. It won't bother me if they get a mix of FGRPB1 and O1AS20. Because of the short deadlines and expected periods of no work to send, I anticipate being able to 'top up' the production machines when necessary without impairing their production status.
The first phase of the new GW
)
The first phase of the new GW search is now underway.
There are ~800 tasks showing on the server status page.
I wonder how long they will last :-).
Cheers,
Gary.
Speaking about AVX, I'd like
)
Speaking about AVX, I'd like to advise to pay more attention to CPU temperatures. Some CPUs have a sort of erratum, which can cause them to exceed the designed TDP when running certain AVX workloads.
This is also the reason why later CPU generations run AVX code at lower frequencies (usually by a few 100 MHz) than the original maximum non-turbo clock.
-----
I just downloaded an "01".
)
I just downloaded an "01". It was promoted to high priority while another job was placed in a "wait" state. Are these jobs being elevated to a higher priority? I do not have a concern but just curious.
Ron, I think they have a
)
Ron,
I think they have a 'Shorter' completion deadline.
Bill
RE: Ron, I think they have
)
Ah. Makes sense.
It's tough to make a
)
It's tough to make a quantitative comparison between SETI and Einstein, but the AVX optimized apps for SETI seemed to produce significantly better results than the plain v7 app, but kind of scattershot compared with the SSE4.1 and SSSE3 optimized apps.
But I did not know about the AVX throttling issue one of the other posters mentioned, so maybe that's why I didn't see the difference among optimized versions.
Will the GPU version come
)
Will the GPU version come after the beta phase or?
RE: Will the GPU version
)
That sounds good to me (and all the other OC'd GPU's we have here)
But I wouldn't mind some of these CPU versions to jump onto some of my HD's and SSD
RE: Will the GPU version
)
Please read this particular message.
Cheers,
Gary.
In this earlier message, I
)
In this earlier message, I mentioned that (for a little while at least) there will be two beta test 'CPU only' runs being conducted. This means that people enabling beta test work in order to get the new GW tasks will also be getting FGRPB1 whilst it remains in beta test.
I'd been looking at this just from a CPU tasks perspective so it didn't initially cross my mind that there is also a GPU app in beta test - the BRP6 cuda55 app. With all the excitement and interest generated by the amazing news of today's press conference, I suspect there may be lots of people wanting to get a piece of the action here, at least until the gloss wears off :-). Perhaps there will be a lot of experienced people who participate at other projects who decide to shift their CPU crunching here for a while to join in the fun.
My concern is that such people wanting CPU work for the new GW run only, may get a surprise to find both FGRPB1 and BRP6-cuda55 tasks (if they have an NVIDIA GPU) which they weren't expecting. If they had deliberately selected only O1AS20-100T, they might be st least mystified and possibly upset by getting all three.
I just intend to point any people questioning this to the earlier explanation about beta tests, via the above link. If the Devs have any other suggestions, it would be good to hear them.
For my own fleet, I regard them as 'production' machines so I'm taking the opportunity, presented by yesterday's false start, to stock up with 'production' tasks, so that I can set NNT when the O1AS20 run recommences. I'm setting up an extra machine or three (no GPU) to participate in the O1AS20 beta test. It won't bother me if they get a mix of FGRPB1 and O1AS20. Because of the short deadlines and expected periods of no work to send, I anticipate being able to 'top up' the production machines when necessary without impairing their production status.
Cheers,
Gary.