How many times have I read that credits are only consistent within a project
not across projects.
My vote is keep BRP5 consistent for GPU work in this project, 4500 - 5500 per job not by hour, that way the relative performance of various GPU's is the only factor.
To keep the credits consistent with BRP4, I would suggest setting the credit for BRP5 tasks in the range of 4,450 and 4,750. This is based on the difference in runtime I have seen between BRP4 and BRP5 tasks and taking into consideration the BRP4 credit value of 500 per task. However, if the goal is to set Einstein to a similar credit return as Seti, the credit value per task may have to be reduced. Like Richard suggested, 4,000 credits per task may be a good starting point which can then be reduced or increased based on further observations.
BRP5 tasks are and will be much longer than BRP4 ones. The probability of errors during computation becomes much higher too. Many users with older GPUs will be forced to give up BRP5 tasks.
In my opinion these longer and more difficult tasks should be rewarded with some bonus points.
I've repeated my 'standard candle' calculations for my GTX 470 'Fermi'
For anyone checking my figures, I have doubled the raw values for SETI and Einstein because I run two tasks at once for those projects. The GPUGrid values are for 'short run' tasks, and only for tasks which earned the 'within 24 hours' completion bonus. GPUGrid pays even higher rates for 'long run' tasks, but I think should be excluded as an outlier.
Most GPUGrid is long run however and the long run WUs score over 2x your figures above. Why you're excluding it, I have no idea.
Real world choice to make for Einstein: if there is plenty of GPU power here and you don't want to attract any more, stick to a low GPU credit award system (lowest that I know of all BOINC projects except maybe SETI). POEM, MilkyWay, PrimeGrid, Collatz, Donate and DistrRTgen are all higher (some much higher). If you want more GPU power to be transferred here, raise the credit to be competitive with other projects. Like it or not that's the reality. We can all wax eloquent about the virtues of solely computing for science (and the projects each of us deems valuable, whatever they are) but the reality is that increased credits/hour attracts crunchers. So take your choice, I'll be here either way.
If we're sticking to nice round figures, may I propose 4,000? That would be 456 credits per hour on this host - slightly high, but not exaggeratedly so.
My HD6850 does BRP5s on average in 17,655 seconds, BRP4 in 2,300 seconds and 500 credit. ((17655/2300)*500)=3837 credit. (One BRP5 took 21K+ seconds, but that was because I was playing Minecraft at the same time. It appears, I need to set MC as an exclusive GPU app from now on. ;-) I did not add the 21K+ result in the average calculation.)
I think 4,000 is just about 'right' if you want to compare it to BRP4 payment.
Most GPUGrid is long run however and the long run WUs score over 2x your figures above. Why you're excluding it, I have no idea.
I was try to help Bernd with a 'standard candle' - a single host which was shared between two or more projects, and crunched for them under as close to identical conditions as possible. Although I do have a dual-Kepler host which crunches 'long run' GPUGrid tasks - so I'm fully aware of how much they pay - that host is being dedicated to a SETI Beta test of VLAR tasks on NVidia this weekend, and didn't have the necessary record of recent Einstein tasks to be helpful.
The Fermi host usually rotates through all three projects, but with the recent batch of GIANNI_VIL 'short' (ahem) tasks taking seven or eight hours, there's no way I could rotate three projects, crunch 'long' tasks, and report everything within 24 hours. I restricted my report to what I had available to report.
We are having this discussion because the current number of volunteers, and the general increase in computational power, exhausted the supply of raw data and stressed out the project servers sooner than the project team expected. They have no need to join the ever-spiralling credit race to attract yet more volunteers.
The new tasks require about 7.8 times more time on my GPU (GeForce GT 640) if it runs at full load. Single task calculations seem to cause a slightly lower GPU load than BRP4 tasks. The CPU usage seems somewhat higher than for the BRP4. So I agree that about 4000 credits per task would be right if it should be comparable to the BRP4 payment.
If you want more GPU power to be transferred here, raise the credit to be competitive with other projects. Like it or not that's the reality. We can all wax eloquent about the virtues of solely computing for science (and the projects each of us deems valuable, whatever they are) but the reality is that increased credits/hour attracts crunchers. So take your choice, I'll be here either way.
This comparison may be severely flawed for GPUs (what is a CPU second for a GPU App?), but these charts also include GPUGrid, which doesn't have a CPU App at all. So these can hardly be about CPU Apps only.
BM
Since I have my GPUGrid spreadsheet open in front of me, I can report that at GPUGrid, my GTX 470 is awarded in 'credit per CPU hour':
[pre]Median 24,607
Mean 45,006
Max 154,572
Min 23,995[/pre]
The reason for the huge difference between median and mean is that different types of task require different levels of CPU support - CPU time as a %age of runtime ranges from below 3% to nearly 17%.
Similarly, that host is averaging over 7,000 'credits per CPU hour' for BRP4 work on the GPU. It's not a meaningful statistic.
Hi Richard,
I think I'm with most project participants on this one: I'm not really concerned about it.
Except...
I would hate for Einstein to be the project that pays least, just for the project's own sake.
Using the dual 560Ti as a baseline, they are able to do dual BRP5s in just under 7 hours. So two every seven is 3.5 hours each, 24/3.5= 6.85 BRP5s per card, per day. At SETI each card does about 25k, so 25k/6.85 = 3650 "credits" per work unit to roughly equal SETI credit on those machines.
Einstein usually "pays" better than SETI, so 4,000 seems reasonable if you want to keep things as they have-been.
My HD 6770 (DDR3) doing two at a time, takes about 11.5 hours.
My HD 7770 doing two at a time takes just under 8 hours.
The 560 machine does two on each card in about 7.6 hours.
I've got a 660Ti machine doing work in about 4 hours, another in just under 5 (two SOC cards in one, two reference in the other).
The triple 670 machine's numbers are a little confusing because part of the time it was doing 3 at a time, then 2 at a time, then 1 at a time, and there was a mix of BRP4 and BRP5 for a while, but I *think* I'm seeing that it can do three at a time in just under 9 hours. So, a three 670 machine can do roughly one per hour x 24hrs = 96,000 "credits" per day @ 4,000/ credit which is within +/- 10% of that computer's "daily score" doing BRP4 work.
It really doesn't make any difference to me one way or the other because I'm only racing with myself...
...BUT, if I WERE racing with someone else that had two years worth of work built-up at a higher pay-rate than I can get, I wouldn't care for that. If I were the person being over-taken, I wouldn't like that a machine that's over-taking mine is now being paid much better than I was when I built-up my lead.
Of course, we both know there will never be any kind of childish behavior on this project.
By the way, regardless of what they pay, they need to start validating and paying something, right away. If you have influence over those sorts of decisions, we'd appreciate your lighting a fire. I now have 188 of those suckers Pending with 2002 In Progress.
We are having this discussion because the current number of volunteers, and the general increase in computational power, exhausted the supply of raw data and stressed out the project servers sooner than the project team expected.
Is that still true with the new Perseus Arm Survey and longer BRP5 WUs?
tbret:
Quote:
Of course, we both know there will never be any kind of childish behavior on this project.
I believe the BRP4 tasks were
)
I believe the BRP4 tasks were underscored by a factor of ~10 when comparing with other BOINC projects.
-----
How many times have I read
)
How many times have I read that credits are only consistent within a project
not across projects.
My vote is keep BRP5 consistent for GPU work in this project, 4500 - 5500 per job not by hour, that way the relative performance of various GPU's is the only factor.
To keep the credits
)
To keep the credits consistent with BRP4, I would suggest setting the credit for BRP5 tasks in the range of 4,450 and 4,750. This is based on the difference in runtime I have seen between BRP4 and BRP5 tasks and taking into consideration the BRP4 credit value of 500 per task. However, if the goal is to set Einstein to a similar credit return as Seti, the credit value per task may have to be reduced. Like Richard suggested, 4,000 credits per task may be a good starting point which can then be reduced or increased based on further observations.
RE: RE: BRP5 tasks are
)
Most GPUGrid is long run however and the long run WUs score over 2x your figures above. Why you're excluding it, I have no idea.
Real world choice to make for Einstein: if there is plenty of GPU power here and you don't want to attract any more, stick to a low GPU credit award system (lowest that I know of all BOINC projects except maybe SETI). POEM, MilkyWay, PrimeGrid, Collatz, Donate and DistrRTgen are all higher (some much higher). If you want more GPU power to be transferred here, raise the credit to be competitive with other projects. Like it or not that's the reality. We can all wax eloquent about the virtues of solely computing for science (and the projects each of us deems valuable, whatever they are) but the reality is that increased credits/hour attracts crunchers. So take your choice, I'll be here either way.
RE: If we're sticking to
)
My HD6850 does BRP5s on average in 17,655 seconds, BRP4 in 2,300 seconds and 500 credit. ((17655/2300)*500)=3837 credit. (One BRP5 took 21K+ seconds, but that was because I was playing Minecraft at the same time. It appears, I need to set MC as an exclusive GPU app from now on. ;-) I did not add the 21K+ result in the average calculation.)
I think 4,000 is just about 'right' if you want to compare it to BRP4 payment.
RE: Most GPUGrid is long
)
I was try to help Bernd with a 'standard candle' - a single host which was shared between two or more projects, and crunched for them under as close to identical conditions as possible. Although I do have a dual-Kepler host which crunches 'long run' GPUGrid tasks - so I'm fully aware of how much they pay - that host is being dedicated to a SETI Beta test of VLAR tasks on NVidia this weekend, and didn't have the necessary record of recent Einstein tasks to be helpful.
The Fermi host usually rotates through all three projects, but with the recent batch of GIANNI_VIL 'short' (ahem) tasks taking seven or eight hours, there's no way I could rotate three projects, crunch 'long' tasks, and report everything within 24 hours. I restricted my report to what I had available to report.
We are having this discussion because the current number of volunteers, and the general increase in computational power, exhausted the supply of raw data and stressed out the project servers sooner than the project team expected. They have no need to join the ever-spiralling credit race to attract yet more volunteers.
The new tasks require about
)
The new tasks require about 7.8 times more time on my GPU (GeForce GT 640) if it runs at full load. Single task calculations seem to cause a slightly lower GPU load than BRP4 tasks. The CPU usage seems somewhat higher than for the BRP4. So I agree that about 4000 credits per task would be right if it should be comparable to the BRP4 payment.
RE: If you want more GPU
)
You are dam' right!
RE: RE: This comparison
)
Hi Richard,
I think I'm with most project participants on this one: I'm not really concerned about it.
Except...
I would hate for Einstein to be the project that pays least, just for the project's own sake.
Using the dual 560Ti as a baseline, they are able to do dual BRP5s in just under 7 hours. So two every seven is 3.5 hours each, 24/3.5= 6.85 BRP5s per card, per day. At SETI each card does about 25k, so 25k/6.85 = 3650 "credits" per work unit to roughly equal SETI credit on those machines.
Einstein usually "pays" better than SETI, so 4,000 seems reasonable if you want to keep things as they have-been.
My HD 6770 (DDR3) doing two at a time, takes about 11.5 hours.
My HD 7770 doing two at a time takes just under 8 hours.
The 560 machine does two on each card in about 7.6 hours.
I've got a 660Ti machine doing work in about 4 hours, another in just under 5 (two SOC cards in one, two reference in the other).
The triple 670 machine's numbers are a little confusing because part of the time it was doing 3 at a time, then 2 at a time, then 1 at a time, and there was a mix of BRP4 and BRP5 for a while, but I *think* I'm seeing that it can do three at a time in just under 9 hours. So, a three 670 machine can do roughly one per hour x 24hrs = 96,000 "credits" per day @ 4,000/ credit which is within +/- 10% of that computer's "daily score" doing BRP4 work.
It really doesn't make any difference to me one way or the other because I'm only racing with myself...
...BUT, if I WERE racing with someone else that had two years worth of work built-up at a higher pay-rate than I can get, I wouldn't care for that. If I were the person being over-taken, I wouldn't like that a machine that's over-taking mine is now being paid much better than I was when I built-up my lead.
Of course, we both know there will never be any kind of childish behavior on this project.
By the way, regardless of what they pay, they need to start validating and paying something, right away. If you have influence over those sorts of decisions, we'd appreciate your lighting a fire. I now have 188 of those suckers Pending with 2002 In Progress.
RE: We are having this
)
Is that still true with the new Perseus Arm Survey and longer BRP5 WUs?
tbret:
Not on this project and not on this planet ;-)