Can anyone tell me what other projects (e.g. SETI) grant on average per GPU hour?
Difficult to tell. I presume you mean stock app, not optimized 3rd party?
CUDA, CUDA23, CUDA_Fermi (Multibeam), ATI_OpenCL_100, CUDA_OpenCL_100 or OpenCL_NVIDIA_100 (Astropulse)? And by Wednesday the new v7 applications, thus all the aforementioned can go out the window? ;-)
And even then it's difficult as it changes according to the client the users use. Seti uses CreditNew, so it really is anyone's guess how it gives out credit. ;-)
Can anyone tell me what other projects (e.g. SETI) grant on average per GPU hour?
There are a lot of variables involved such as GPU load (#of simultaneous tasks), application version, Cuda or CL version, OS. Here are some Seti numbers for my pc's:
[Can anyone tell me what other projects (e.g. SETI) grant on average per GPU hour?
Ho Bernd. Don't know about seti, creditnew is bizarre and erratic to say the least. Probably one of the reasons for the mass exit from the project.
GPUGrid grants about 8250 credits/hour on a relatively slow GTX 650 Ti. Double that on a GTX 670. POEM, MilkyWay, Collatz, Donate and DistrRTgen are all higher (some much higher) for both ATI/AMD and NVidia GPUs. The WCG GPU app is gone but it granted very low credits (but higher than Einstein ;-).
BTW, thanks for a great project and one that seems to value the crunchers and keeps us informed. Thumbs up.
I completed my first BRP5 work unit in 25 hours 12 minutes on my GT 610, which is one of the slowest GPU's on this project.
Computer usability was no different compared to running BRP4 work.
As for credits, if I had spent that time running Collatz Conjecture, I would have earned 41,000 credits.
Basically I think E@H doesn't give much credit for tasks.
Though it's relative, according to my stats, GPUGrid for example gives about 12600 points/hour on a 660 Ti. This however includes a bonus for returning task within 24 hours (long runs).
Milkyway@Home gives about 13600 points/hour on a HD7950 (double-precision).
BRP4 tasks are ~ 1500 pts/hour on 660 Ti running 2 tasks in parallel.
From the approximately 40 validated tasks currently showing at each project, I'm getting:
SETI - average 320 credits per hour (range 447 to 140 - that's CreditNew for you)
Einstein - average 441 credits per hour (range 490 to 379)
The host has just completed its second BRP5 task, in an average of ~31,500 seconds (call it 8.75 hours)
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.
NB - the host does run third-party apps at SETI, but on older GPUs (9800GT), there's not a great deal - if any - optimisation to be squeezed out.
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.
That is, I think, good starting point for further discussion.
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.
@EDIT
I'd like to resolve one doubt - can we determine how many times more computational intensive are BRP5 tasks comparing to BRP4 (in FLOPS let's say)? If we knew this (absolute) value, it would be easier to talk about credits/WU.
I think you'll find that Collatz has changed recently. Solo_Collatz WUs now average around 8000 credits apiece.
Solo_Collatz WUs take 5 hours on my GT 610, and I have been getting 8,200 credits each. In 25 hours that's 41,000.
My guess for BRP5 work units is that they will be granted around 5,000 credits each, unless someone can come up with a better estimate.
RE: Can anyone tell me what
)
Difficult to tell. I presume you mean stock app, not optimized 3rd party?
CUDA, CUDA23, CUDA_Fermi (Multibeam), ATI_OpenCL_100, CUDA_OpenCL_100 or OpenCL_NVIDIA_100 (Astropulse)? And by Wednesday the new v7 applications, thus all the aforementioned can go out the window? ;-)
And even then it's difficult as it changes according to the client the users use. Seti uses CreditNew, so it really is anyone's guess how it gives out credit. ;-)
RE: Can anyone tell me
)
There are a lot of variables involved such as GPU load (#of simultaneous tasks), application version, Cuda or CL version, OS. Here are some Seti numbers for my pc's:
560Ti Linux: 833 cr/hr
560Ti Win 7: 958 cr/hr
580 Win 7: 1500 cr/hr
570 Linux: 1000 cr/hr
Just finished a GPUGrid task
)
Just finished a GPUGrid task in 19660 seconds on a GTX580 on Linux and that yielded 31050 credits. That equals 5685 cr/hr...
RE: [Can anyone tell me
)
Ho Bernd. Don't know about seti, creditnew is bizarre and erratic to say the least. Probably one of the reasons for the mass exit from the project.
GPUGrid grants about 8250 credits/hour on a relatively slow GTX 650 Ti. Double that on a GTX 670. POEM, MilkyWay, Collatz, Donate and DistrRTgen are all higher (some much higher) for both ATI/AMD and NVidia GPUs. The WCG GPU app is gone but it granted very low credits (but higher than Einstein ;-).
BTW, thanks for a great project and one that seems to value the crunchers and keeps us informed. Thumbs up.
I completed my first BRP5
)
I completed my first BRP5 work unit in 25 hours 12 minutes on my GT 610, which is one of the slowest GPU's on this project.
Computer usability was no different compared to running BRP4 work.
As for credits, if I had spent that time running Collatz Conjecture, I would have earned 41,000 credits.
Basically I think E@H doesn't
)
Basically I think E@H doesn't give much credit for tasks.
Though it's relative, according to my stats, GPUGrid for example gives about 12600 points/hour on a 660 Ti. This however includes a bonus for returning task within 24 hours (long runs).
Milkyway@Home gives about 13600 points/hour on a HD7950 (double-precision).
BRP4 tasks are ~ 1500 pts/hour on 660 Ti running 2 tasks in parallel.
-----
RE: As for credits, if I
)
I think you'll find that Collatz has changed recently. Solo_Collatz WUs now average around 8000 credits apiece. Test run times for my GPUs were:
HD 7770: 45:21 (minutes:seconds)
HD 7790: 35:38
HD 7850: average of 56:00 for 2 WUs, so 28:00/WU
So even a lowly HD 7770 does about 10,500 credits/hour.
An HD 7850 mid-range GPU now does over 16,000 credits/hour.
RE: Can anyone tell me what
)
May I offer this host as a standard candle?
SETI host 3751792 (valid GPU tasks)
Einstein host 1001562 (valid BRP4 tasks on GPU)
From the approximately 40 validated tasks currently showing at each project, I'm getting:
SETI - average 320 credits per hour (range 447 to 140 - that's CreditNew for you)
Einstein - average 441 credits per hour (range 490 to 379)
The host has just completed its second BRP5 task, in an average of ~31,500 seconds (call it 8.75 hours)
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.
NB - the host does run third-party apps at SETI, but on older GPUs (9800GT), there's not a great deal - if any - optimisation to be squeezed out.
RE: If we're sticking to
)
That is, I think, good starting point for further discussion.
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.
@EDIT
I'd like to resolve one doubt - can we determine how many times more computational intensive are BRP5 tasks comparing to BRP4 (in FLOPS let's say)? If we knew this (absolute) value, it would be easier to talk about credits/WU.
RE: I think you'll find
)
Solo_Collatz WUs take 5 hours on my GT 610, and I have been getting 8,200 credits each. In 25 hours that's 41,000.
My guess for BRP5 work units is that they will be granted around 5,000 credits each, unless someone can come up with a better estimate.