I reviewed some the overall top BOINC performers in terms of credit and ran across a guy that has received an insane amount of credit for the number of computers running. I dug into how he achieved it and discovered that he was mainly running Collatz Conjecture.

So, I attached to this project and began running WUs.

I got an eye-popping 8,255 credits for one Collatz WU which required 163 CPU seconds.
By comparison, I got 91 credits for one Seti@home WU which required 165 CPU seconds.
Also, I got 177 credits for one POGS WU which required 7,064 CPU seconds.

I'm not trying to blow a whistle here or be a party pooper, just trying to understand how or why there could be such credit award discrepancies, and why there isn't an enforced standard.

My stats: Win 7 Pro 64-bit, AMD 8-core CPU, 8GB memory, BOINC Manager version 7.2.28

Thanks.

There is a user MossyRock at SETI who has a computer just like that.

Significantly, it is also fitted with a NVIDIA GeForce GTX 650 graphics card.

The timing and credit awarded match Task 3253370173. That task ran in total (Run time) for 1,259.83 seconds, and the computation was done on the graphics card. I suspect you will find that something similar applies to the Collatz task: in these cases the run time (also described as elapsed time) is the significant value to compare, rather than the CPU time.

There is a formal definition for the value of a BOINC credit (in terms of the number of floating point calculations performed), but as a statement of fact: each project is independent, and is free to award credits as it sees fit. SETI tries (but doesn't always succeed) to follow the defined standard: I'll leave others to answer your question about 'why' the figures seem different.

Richard,

Yes, that's me, MossyRock (the originator of this post).

Thank you for the additional information.

It's just hard to fathom that I can be awarded 40% of my entire usual daily RAC from a single WU that ran in just a few minutes elapsed-time (my normal daily RAC is around 20K, and this one Collatz WU gave me over 8K).

No wonder people can easily rack up hundreds of thousands, if not millions, of credits a day with Collatz.

Collatz
DistrRTgen
GPUGrid
POEM
PrimeGrid

all have high credit GPU apps...not just Collatz

The commonality is in their use of the GPU that people have attached to their computers. All of these projects have optimized apps for the GPU and are running at high utilization. Other projects, on the other hand, often have not optimized their apps for the GPU. Even within a project there can be optimized and unoptimized GPU apps. There are also cr/effort differences between OSes as well as between ATI and Nvidia. [TMI... the floating point precision makes a difference for some projects.]

Furthermore, GPUs have a far greater number of cores to work with than CPUs. So GPU WUs and CPU WUs are not really comparable. In my case my GPU (GTX 660Ti) has 1344 cores. So while my CPU WU is processing a single operation, my GPU WU can be processing up to 1344 operations.

But going back to the OP, you only stated the CPU time not run time which, usually, is more important for GPU WUs. This is because most GPU apps use very little CPU processing. [If only BOINC would also list GPU time.]


Mid credit GPU projects include:
Albert
Einstein
Milkyway
Moo! Wrapper

Low credit GPU projects...
SETI
Seti Beta
You might as well run the cpu apps as SETI is either so unoptimized or their credit system is so messed up that they value a GPU at a fraction of its potential.

So branch out and test the other projects and see what they are like. I feel sorry for all those that join SETI and never venture further into the Boincverse.