It seems a bit unfair
13 May 2023 20:10:13 UTC
Topic 229510
(moderation:
I recently upgraded my GPU and now my RAC is going down! My rx 6600 has twice the memory my old rx 560 had so now I get only gravity wave work units. Previously I was getting gamma ray pulsar units which run much faster. I'm not in it for bragging rights but it is kind of a bummer since I was expecting a higher RAC.
Oh well.
Why don't you change your
)
Why don't you change your preference settings to run only the type of work that suits you best?
Cheers,
Gary.
Had a similar issue recently.
)
Had a similar issue recently. Went from an nvidia 1650super to a 3080TI but the RAC stayed about the same. makes no sense as one card is almost 10x the power of the other.
Had to go in and tweak my preferences to work on the jobs best suited to my card. RAC quadrupled after that.
I haven't ran the GW -GPU
)
I haven't ran the GW -GPU tasks since shortly after they came out. They only had parts of the calculation moved over to the GPU. Between that and whatever bottlenecks they have on the GPU itself (an early GPU port NVidia attempted ended up being slower than the CPU version because the code did tons of random memory access - something that GPUs are much worse at than CPUs) with the result that while faster than the CPU version they've never been able to fully utilize anything beyond a relatively low end GPU.
I think a few arguments
)
I think a few arguments describe different things here:
For some GPUs only a (large) part of the work is done on the GPU. A final calculation of top lists is done on the CPU, if the GPU is not capable of FP64 (double precision) arithmetics (see log file outputs of the science apps). The RAC value is calculated from credits. The credits for different searches (science apps FGRP5, O3MD1, BRP4 etc.) do not perfectly reflect the computing effort. Some apps generate more credits per day than others. It also depends on the CPU, the GPU, etc. As Gary said, everyone can choose their preferred app in the configuration.
Finally: Einstein@home is not about achieving high RAC values, but about processing the necessary work units for the scientific output. During the years there were approaches for a new credit mechanism in BOINC, which again had other disadvantages. That's why Einstein@home uses the classic credits. There have been endless discussions about it here and in other projects. It is like it is.
I would be satisfied with a new GPU card if it performs the same tasks faster. At the moment a lot of O3MD1 CPU and GPU tasks are distributed. They generate less credits, less RAC than e.g. FGRP1BG (Fermi Gamma Ray Pulsar search). Doesn't matter.
Ben Scott wrote: I recently
)
When I look at your pc's I see one pc with a 6600 and one pc with a 1060, the one with the 1060 has not done any tasks at all while the one with the 6600 is doing GW tasks that only give out 1,000 credits per tasks, while the GRP#1 tasks give out 3,465 credits for each task, ie you will get more credits if you run the GRP#1 tasks and your RAC will go up faster. Now if Science is the goal then it doesn't matter which tasks you run as both types are needed to be gone thru, but if credits and RAC is one of your goals then running tasks like the GRP#1 tasks will help you achieve that goal.
As a volunteer paying for the
)
As a volunteer paying for the electricity and the expensive GPU's. The only thing I feel we get in return is the credit numbers, maybe a cert if one of our machines is "the one" to work a task that finds a new pulsar.
It would be nice for the seemingly expensive to run O3 tasks would deliver more credit. The flat 1,000 points per unit does not reflect the time or the electricity it costs to execute them. At least compared to other tasks we can choose to run
If its points your after (which I am). It takes a little reading of the forums to stitch together which tasks are the best for RAC. Which gives those who do that leg work a huge advantage if they're trying to build up points and rise through the ranks. The competitive aspect of points is part of the fun for some. Maybe you want it this way? having to do the leg work + experimentation to find that sweet spot. But if your a volunteer and all you want to do is attach to the project and let it run. You'll probably find yourself in the backseat for RAC
yummycheese wrote: As a
)
One thing I have done over the years is to come to a new for me Project with a zero resource share so I get one task and just before it finishes get a 2nd task etc etc, that way I have a chance to try out the carious apps at each Project before I jump in with both feet and start running tasks with a cache of a day or so. I also always set my cache settings to 0.1 and 0.1 that way I don't get too many tasks until I'm ready then I go to my normal day or so cache settings.
It is true that 1000 credits
)
It is true that 1000 credits does not reflect the time and electricity used to run the O3 tasks but that is not the point of credit. The point is to show how much work is done not how hard you worked. Originally GWs were CPU only and were correctly assigned 1000 credits. Later a GPU app was developed and the same data could be run faster and the same 1000 credits remain because the same work was done.
Betreger wrote: It is true
)
do you think FGR and BRP are doing 3x the work?
_________________________________________________________________________
Bernd does and in my opinion,
)
Bernd does and in my opinion, he is very conscious of credit value