I seem to recall that assigning points was a bit of a random process.
At the beginning of the project around 16 years ago they used the default system where BOINC assigned the points based on a complicated formula, and nobody was happy with that. Next they went to a fixed-point awards system, and that seemed to be somewhat randomly assigned. For instance, there is a big discrepancy between the project points awarded for the Gamma-ray pulsar binary search #1 on GPUs (3,465 points) versus the new Gravitational Wave search O3 All-Sky #1 (1,000 points).
My working assumption has been that the points must be proportional to the number of floating point operations. If it aims to measure the value of the science, then I can't explain why gpu tasks are worth so much more than cpu tasks across different projects. Are cpu-only projects inherently less valuable? That doesn't make sense.
I seem to recall that assigning points was a bit of a random process.
At the beginning of the project around 16 years ago they used the default system where BOINC assigned the points based on a complicated formula, and nobody was happy with that. Next they went to a fixed-point awards system, and that seemed to be somewhat randomly assigned. For instance, there is a big discrepancy between the project points awarded for the Gamma-ray pulsar binary search #1 on GPUs (3,465 points) versus the new Gravitational Wave search O3 All-Sky #1 (1,000 points).
My working assumption has been that the points must be proportional to the number of floating point operations. If it aims to measure the value of the science, then I can't explain why gpu tasks are worth so much more than cpu tasks across different projects. Are cpu-only projects inherently less valuable? That doesn't make sense.
One thing some projects do, I do not know about Einstein in particular, is to give more credits for something they want more people to work on so they get that part of the research done quicker and for the 'credit whores' that works great.
One thing some projects do, I do not know about Einstein in particular, is to give more credits for something they want more people to work on so they get that part of the research done quicker and for the 'credit whores' that works great.
I would rather work on what the researchers think is important. They seem to be sending mixed messages here, because the gamma ray pulsar app gives more credit, but I get mostly gravitation wave tasks when I don't disable it.
One thing some projects do, I do not know about Einstein in particular, is to give more credits for something they want more people to work on so they get that part of the research done quicker and for the 'credit whores' that works great.
I would rather work on what the researchers think is important. They seem to be sending mixed messages here, because the gamma ray pulsar app gives more credit, but I get mostly gravitation wave tasks when I don't disable it.
that's more a consequence of how BOINC decides what work to fetch than the project intentionally prioritizing gravitational wave tasks.
One thing some projects do, I do not know about Einstein in particular, is to give more credits for something they want more people to work on so they get that part of the research done quicker and for the 'credit whores' that works great.
I would rather work on what the researchers think is important. They seem to be sending mixed messages here, because the gamma ray pulsar app gives more credit, but I get mostly gravitation wave tasks when I don't disable it.
that's more a consequence of how BOINC decides what work to fetch than the project intentionally prioritizing gravitational wave tasks.
Still a lot in control by the project with their locality scheduling server version. I would state that is the main reason why a host with both sub-projects selected will get a preponderance of Gravitation work.
That GPU tasks are given more credit doesn't surprise me. In my case, I have a laptop with an NVidia GPU and a desktop with a much newer (current model) AMD GPU. I run 5 CPU and 1 GPU job on my Laptop, and 10 CPU and one GPU task on the desktop. I expect I get more science out of my 2 GPUs than the multiple tasks running on my CPUs. (18 cores (not threads) between my laptop and desktop) I am sure the GPUs are quite useful for SIMD calculations, which I assume all of this stuff is. The parallelism is massive in the GPUs! I am looking forward to upgrading to an AMD 7900XT when they come out for an even bigger bang! Dual GPUs I believe. Double+ the number of compute units.
I stopped the jobs running on the Intel GPU/APU on my laptop, the performance of which I found unimpressive.
I guess I am open to argument on that one. APU job or CPU job?
Indeed we had a problem with
)
Indeed we had a problem with the validator. It should be fixed now, I issued the 'validate error' tasks for re-inspection.
BM
Thanks Bernd!!
)
Thanks Bernd!!
My working assumption has
)
My working assumption has been that the points must be proportional to the number of floating point operations. If it aims to measure the value of the science, then I can't explain why gpu tasks are worth so much more than cpu tasks across different projects. Are cpu-only projects inherently less valuable? That doesn't make sense.
Neal Burns wrote: I seem to
)
One thing some projects do, I do not know about Einstein in particular, is to give more credits for something they want more people to work on so they get that part of the research done quicker and for the 'credit whores' that works great.
mikey wrote: One thing some
)
I would rather work on what the researchers think is important. They seem to be sending mixed messages here, because the gamma ray pulsar app gives more credit, but I get mostly gravitation wave tasks when I don't disable it.
Neal Burns wrote: mikey
)
that's more a consequence of how BOINC decides what work to fetch than the project intentionally prioritizing gravitational wave tasks.
_________________________________________________________________________
Ian&Steve C. wrote: Neal
)
Still a lot in control by the project with their locality scheduling server version. I would state that is the main reason why a host with both sub-projects selected will get a preponderance of Gravitation work.
That GPU tasks are given more
)
That GPU tasks are given more credit doesn't surprise me. In my case, I have a laptop with an NVidia GPU and a desktop with a much newer (current model) AMD GPU. I run 5 CPU and 1 GPU job on my Laptop, and 10 CPU and one GPU task on the desktop. I expect I get more science out of my 2 GPUs than the multiple tasks running on my CPUs. (18 cores (not threads) between my laptop and desktop) I am sure the GPUs are quite useful for SIMD calculations, which I assume all of this stuff is. The parallelism is massive in the GPUs! I am looking forward to upgrading to an AMD 7900XT when they come out for an even bigger bang! Dual GPUs I believe. Double+ the number of compute units.
I stopped the jobs running on the Intel GPU/APU on my laptop, the performance of which I found unimpressive.
I guess I am open to argument on that one. APU job or CPU job?
General consensus on running
)
General consensus on running both cpu and gpu tasks on Intel igpu is that it slows down BOTH application tasks.
Either one or the other or leave the gpu tasks off the igpu.
FGRPB1G work generator not
)
FGRPB1G work generator not running, no tasks ready to send.
_________________________________________________________________________