credit inequities
24 Feb 2013 13:52:49 UTC
Topic 196823
(moderation:
I don't know f this has been addressed in the past.
I usually do not pay much attention to credit but when my 4 core machine was getting more than the 8 core I started to wonder what was up.
The 8 core was getting almost all Gamma ray Pulsar Search.
times 8-9.5K cpu times abt the same....credit 80
4 core largely Binary Radio Pulsar Search
times 5-6.5 cpu time 500.............credit 500
Also a few Gravitational Wave S6 Searches in both
times 27k cpy time 26k................credit 250
I fail to see how this is fair for anyone?????
M Henry
credit inequities
)
correction .......5-6.5K
RE: I don't know f this has
)
Generally speaking each project gives credits based on how much work is done for each of its projects, rarely giving the same credits for each type of unit. On the website under your account and then Einstein@home preferences, you can then select which type of units you want to run. You can run different units for the different groups you can put your pc's in, school, work, home etc. Selecting the best type of units for your goals is the idea, you don't have to settle for what the project sends you, BUT this also means you COULD run out of units if there are not enough of the kind you choose. That is why there is ALSO the next line that lets run units from other type is your favorite type is out at the moment. The BRP units are done by a gpu, not the cpu, this means they are doing roughly 10 times the amount of work in the same amount of time as a cpu can. The credits awarded seem to reflect that difference.
RE: I don't know f this has
)
Ive made a similar question a long time ago, and the explanation for the difference in credits between the 2 native CPU apps was that on Intel CPUs the GW searches run faster then the Fermi searches while on AMD its the opposite... so if they change the credits to make them more similar on Intel then there will me much more difference on AMD... so they were giving credits based in a kind of compromise value to make it fair for Intel and AMD users...
BRP tasks are intended to be run on GPU, they can be crunched on CPU also but that option was added to allow people without GPU to been able to contribute to that project too... I guess they didnt give more credits to this tasks when they are crunched on CPU for several reassons, but mainly, because this is not the main project and they preffer that people uses their CPU to run the other searches.
If your goal is to maximize the credits received, you should set your prefferences to not accept the BRP tasks on CPU, and in the CPU accept only the app that gives you the best ratio of credits/time.
By the way, I think that as long as every user gets the same credits for the same work it's allways fair, no matter if one apps gives more or less credits than another... Unfair will be if they were giving more credits to people that pay for them or something in that line...
Guys - look at his hosts and
)
Guys - look at his hosts and task lists before posting, please.
He is running BRP on GPUs (graphics co-processor, running in CUDA general-purpose computing mode).
He has a NVIDIA GeForce GT 220 on the 8-core machine, and a NVIDIA GeForce 8800 GTS on the 4-core machine. Although the 8800GTS is a much older card design, it's computationally more powerful, which explains the shorter run times.
Also, read the 4-core times carefully. They are
Elapsed time ~5K to 6.5K seconds
CPU time ~500 seconds
Another clue that the bulk of the computation is not being done on the CPU.