I've noticed that within the past few days, my average runtime for GPU units on my HD4670 has dropped from around 2 hours to 18 minutes. I'm not saying that's a bad thing, just curious as to why such a sudden and drastic change would crop up.

I've noticed on the Server Status page, that there are workunits of varying sizes, is there info anywhere on what constitutes one size of workunit versus another? And, is there a way to determine which size a particular workunit is?

Lastly, it would be great if I could get this project working on my Mac OS X clients; the GeForce 320M does nothing but error out and the HD6750M never gets any work.

---

Click here to see My Detailed BOINC Stats

I've noticed that within the past few days, my average runtime for GPU units on my HD4670 has dropped from around 2 hours to 18 minutes. I'm not saying that's a bad thing, just curious as to why such a sudden and drastic change would crop up.

I've noticed on the Server Status page, that there are workunits of varying sizes, is there info anywhere on what constitutes one size of workunit versus another? And, is there a way to determine which size a particular workunit is?

Lastly, it would be great if I could get this project working on my Mac OS X clients; the GeForce 320M does nothing but error out and the HD6750M never gets any work.

Over at Collatz, a project that always seems to test the next thing Boinc is working on, the units are multiples of each other. ie a basic unit is 32 times as large as an Android unit, 16 times as big as a basic cpu unit, 8 times as big as another kind, etc. I am going to guess that Moo is now doing that too, so it would be based on your own machine and how it handles the units. There is no chart saying if you have this gpu send that unit, but instead more of a trail and error thing where it tries to find an optimal size for your gpu, from among the fixed sizes. The project picks the size of the units and the distance between them, so you could have a basic unit that takes 3 hours and a mini unit that only takes 20 minutes and no other choices, or multiple other choices. Collatz is also beginning the process of letting users choose which size unit they want to crunch, that could expand everywhere too. That would mean you could chose either the long or short units to crunch. The credits vary based on the size of the unit being crunched of course.

I should also point out, that Moo! does not seem to make full use of my GPU, it hovers around 90% +/- 5% utilization. I'm not saying that's a completely terrible thing, as it allows me to run GPU computation all the time without a lag in the UI, but perhaps it could be improved on so the project runs more efficiently?

---

Click here to see My Detailed BOINC Stats