What stroke me as odd today is the following:
Workunit 1(432593064) - Completed and validated 244.83sec 52.61(claimed) 38.27(received)
workunit 2 (432708709) Completed and validated 244.09sec 52.61(claimed) 52.6(received)

In the first case when the unit was crunced by my CUDA device and the CPU of the peer I got lower credit (what his part of the WU claimed).
In the second case both parts of the WU were done by CUDA and we both got the same claimed credit.

Both WUs in both cases and times seem pretty similar to me and i think this is not the first time this happened. Do you think this is normal?

For reasons that others can explain, CUDA tends to overclaim credit. you aren't losing credit you are just receiving the credit that should have been awarded to the WU

---

In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

There is no way (currently) for BOINC to accurately count GPU time, so its only able to count the amount of time the CPU loaded the app and sent it to the GPU, then saves the final result when the GPU finishes. Because of this, part of the credit claiming process has to do with Floating Point Operations Per Second (a unit of time measurement), and BOINC thinks that the CPU finished in a fraction of the time that it actually took to run the WU on the GPU, thus claiming much higher than it should.

This is a known issue and I'm not aware of any plan in the works to correct this until BOINC can accurately count GPU time.

---

All clear now, thanks for the answers

Hi, there!

For reasons that others can explain, CUDA tends to overclaim credit. you aren't losing credit you are just receiving the credit that should have been awarded to the WU

Does that mean I get same or less credits for 1hour cuda crunching as I get for 13min cuda crunching?
WU1
WU1
Thanks in advanced =) Would explain why some kill vlars?

---

During browsing berkeley's homepage I came along this statement that makes me say:
"ME!, ME!, ME!, wants to know, what patch was used and is it available for windows?"

That means that the more floating point operations performed per second, the more credits you get, but CUDA apps will always overclaim because there is no reliable way to count FPOPs on the GPU.

As I understand it, the reason why people kill VLARs is because they are too intensive on the GPU and tend to slow down the entire system while crunching.

---

Thanks for the quick reply, OzzFan!

I'll take this as a 'yes' ^^ However, I don't care if I get more or less credits for some WUs. Now I just have to lower my credit goal I want to archive before I go into hibernating mode again.

To keep my GPU running while watching recorded TV-Programs without having 'lag' I download some GPUGrid WUs - The deadline for those VLARs is perfect in this sense.

It's also good to have almost full CPU power for WCG or other projects on my Boinc-Manager.

P.S.: Sorry for my Denglish ^^

---

During browsing berkeley's homepage I came along this statement that makes me say:
"ME!, ME!, ME!, wants to know, what patch was used and is it available for windows?"

The WU's you pointed out are not VLAR. they do have a low angle but not VLAR. YOu'll know VLAR WU's because the credit is usually 60+/WU

---

In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

Hi, skildude!
You're right. My mistake. -_- Me seeing VLARs everywhere, therefore I will prepare to hibernate very soon.
60+/WU is about 5 times more claimed credits for 1 hour cuda crunching as it is for 13 min cuda crunching.
These WUs look much better.
WU2
WU2
SETI@home classic was so much easier. ^^

---

During browsing berkeley's homepage I came along this statement that makes me say:
"ME!, ME!, ME!, wants to know, what patch was used and is it available for windows?"