BOINC came up with it recognizing the 9800 GT as a CUDA processor ... sadly GPU Grid insisted that there was no work for the Linux world ...

By the way Paul, did you have work units for other BOINC projects at the time you attempted to fetch work units for GPUGRID?

After adding another BOINC project and receiving some CPU work units, I noticed some weirdness where the BOINC manager (6.4.5 x86_64 Linux) would refuse to fetch additional GPUGRID work units to replace completed GPU work units until the CPU work units from the added BOINC project completed.

Ye, but that has not been a problem in the past ...

Well, I am back to having 3 in flight and a couple three four queued.

I sure will be glad when we get a reasonable version working that separates the running and fetching for CPU vs. GPU ... I was hoping that 6.6.3 would be it so I could start testing it ... every time I make a note on behavior of 6.5.0 that I am pretty sure will survive into the 6.6.x series I get chastized with the "But you aren't running the latest version..." brushoff ... yet the latest versions have been so bad that they are unusable for reasonable production use ...

I mean, I am doing wok for 40 some projects from newest alphas to the most established production projects ... and I can't afford to have my systems run dry because the developers can't code their way out of a wet paper sack ... :)

Oh, nuts, you meant for the linux system ...

Yes I did have some work queued ... not sure that I had just added a project or not ... well, it has been stable though I still have GPU Grid in the mix their is no GPU in the system. Wednesday when the the new GTX 295 card shows up I will be "bumping" cards down and the Linux box will get the 9800 GT ... I will see then if it will fetch any work for the GPU Grid ... if not, I will point it SaH Beta and see if it will pick up work in the rotation or not ... anyway ... early days for CUDA ...

I noticed some weirdness where the BOINC manager (6.4.5 x86_64 Linux) would refuse to fetch additional GPUGRID work units to replace completed GPU work units until the CPU work units from the added BOINC project completed.

I am running seti and gpugrid and to get around this problem I wrote a short cron job script that runs every 2 hours.

The script:
issues a suspend to seti
issues an update to gpugrid
sleeps 15 seconds
issues a resume to seti

The process doesn't get a download every time it should but it seems to keep my machine running at the max.

Curt

I sure will be glad when we get a reasonable version working that separates the running and fetching for CPU vs. GPU ...

It happened again. CPU-only work units blocked GPU work units from being fetched, due to so many seconds of CPU work units already present.

No more.
I've removed all CPU-only projects. They can run in their own private BOINC, away from the GPU tasks.

... and I can't afford to have my systems run dry because the developers can't code their way out of a wet paper sack ... :)

Oh, nuts, you meant for the linux system ...

Ha!

Apparently, the wet paper sack applies regardless of the system of deployment. :-o

@J.D.

I have "forced" my systems to run GPU tasks by simply figuring out a high enough resource share that the GPU tasks even when run continually can't solve the LTD build up... not the way it is supposed to work ...

And even sadder, it is entirely possible that the changes being made to the fetch algorithms may not make this better anytime soon ...

But, I will admit that I am basking in being back in the running to have a reasonable production per day ... even though I am down to 6 computers and three of them are very decripid ...

Well, the Linux box is just cranking on a couple of my low end projects for the heck of it ... while I try to get my scores up on a few of the "bigger" projects that I contribute to ... I have long wanted to have a bunch of them with more done than I did for SaH ... now if one or two of those other low end projects had a secret CUDA project ... :)