I was curious as to what would be the best ratio to use with the rescheduler? Right now I have 2155 CPU units and 345 GPU and keep getting GPU units that are VLar. I was thinking about transferring some CPU to GPU but am not sure what percentage to set it to so I don't make Boinc go crazy...

---

Official Abuser of Boinc Buttons...
And no good credit hound!

I figured out about how many WUs can be done by my CPU(s) in a day, and how many by my graphics card(s), and use that ratio for recheduling. For example, I have 2 8-core AMD machines, one with 2x GTS 250s and one with one GTS 250. On the dual GPU machine, I set GPU to be 75% and, for the single GPU machine, 60%. And that seems to work fine.

And a byproduct of the rescheduling is that all VLARs are sent to the CPUs, so there is no "cherry picking" and no GPUs potentially hung up on a VLAR for several hours.

---

Well I did 43% and see what happens. I am trying not to start a download frenzy but felt I may go into High Priority mode on CPU if I didn't get rid of some of them.

---

Official Abuser of Boinc Buttons...
And no good credit hound!

I am using 60%, which seemed to be working OK until my i7 920 went berserk last weekend and started downloading every WU it could get it's hands on....
But the rig had crashed, and may have boinked Boinc somehow to bork things up.

It seemed to keep enough loaded into the GPU to keep most of the work downloaded to the CPU, and then transferring it to the GPU as needed when I ran resched again.

---

"Freedom is just Chaos, with better lighting." Alan Dean Foster

I also use 60% on both my single and dual GPU machines, seems to work pretty well for me.

-Dave

---

I use 81%(?!). Big difference between my GPU and CPU.

I used the "Only VLar+VHar to CPU" option when I was running a cuda card. This seems to disable the % slider. However, I had left it at 75%. In general I thought I had to many GPU tasks all the time.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

I've got mine at 60%. I keep trying to tell the server my GPU is much faster than my CPU but it doesn't believe me. Every time I reschedule tasks to the GPU it reloads my CPUs. I usually just let it run on auto and move the VLARs over unless it needs more GPU work.

---

PROUD MEMBER OF Team Starfire World BOINC

I've got mine at 60%. I keep trying to tell the server my GPU is much faster than my CPU but it doesn't believe me. Every time I reschedule tasks to the GPU it reloads my CPUs. I usually just let it run on auto and move the VLARs over unless it needs more GPU work.

Same happens for me. (I think this is a BOINC bug, er, feature.) I think BOINC doesn't account for GPU WUs in deciding if you need more WUs (or does it incorrectly). Thus, a reschedule can cause a big download session.

For example, I just had an automatic reschedule from 536 CPU/143 GPU to 272/407, and BOINC d/l about 200 more WUs (almost all CPU).

---

So far with my 43% I haven't triggered a download...knocks on wood.

---

Official Abuser of Boinc Buttons...
And no good credit hound!

To tell your client the capabilities of your PC you must edit the app_info with the correct (or approx.) flops of your cpu and gpu.

You can find here a tutorial hou you can do that.

After you do that you will see different estimates for cuda and non-cuda tasks. Good luck!