GPUGRID sends me two work units at once, which makes some sense, but it's pushed me into a cycle of lateness. Right now, for example, I have a work unit due today. It only actually started yesterday because I was only working on the previous unit. So there's almost no chance it's going to finish in time.

I don't want to cancel it, since the initial replication is 1 and there's a chance that when the system realizes I'm late and issues it to other folks, I'll still be the fastest to return it. But if I finish it on the 15th or 16th, there's also a good chance that somebody else will have finished it before me. So my work is useless and I'll get no credit.

The problem is, this cycles. I've got another work unit due on the 18th. If I canceled the one due today, I'd finish that, no problem. But if I don't finish this one until the 16th, there's a good chance the next one will be late too. After all, that's why the current one is late.

I can't seem to find a way to force GPUGRID to only send me one work unit at a time. I think that would solve the problem.

There is no such option - sad but tru..... Eg, if you have quad and some like 8600GT - you can finish 1 max 2 task before dedline.... and rest is lost.... if the server will not terminate them you must do this manually....

---

POLISH NATIONAL TEAM - Join! Crunch! Win!

Try lowering your ressource share for GPU-Grid.

MrS

---

Scanning for our furry friends since Jan 2002

Thanks, guys. I have a 9800, so it's not terribly slow, but also not exactly intended to high performance computing. I've tried playing with the resource share, but with no effect. I guess I'll just cancel jobs I have no chance of finishing on time and let the server give it to somebody else.

Seems like a terrible solution, though.

I struggled with this one for a while - mine's a 9800GTX+.

In the end I set the cache size to 0.1 and from that point on I have got one download just over 2 hours before the running one completes (which fits when you think about it - 0.1 x 24hrs ....). It worked for me as that 2 hours gave me enough flexibility to get over "murphys law" which states when you need a download "now" your not getting it, like it or lump it rofl:)

The 2 hours meant I was not caught by the fiddly things in life that temporarily stop me from downloading, and time to "recover". Not been caught since. Its predicated on the GPUGRID server being reliable as in effect I have no cache, however only been caught once due to a server crash, so its worked out fine as the server is by in large very reliable.

Another disadvantage of that is if you run multiple projects (as I do), there is a need to manually fill the cache for the other projects, but as thats only once every few days its not too onerous. You could even set cache to zero if you manually fill other projects, as the download is quick, and would certainly minimise "dead" time, you'd only "lose" a couple of minutes whilst it downloaded another one.

I'm finding if I particularly want two or more, I just raise the cache a little to - say - 0.6 or whatever.

Regards
Zy

Oh, totally forgot to mention cache size. On my rig with a 9800GTX+ I set GPU-Grid to about 20% ressource share and to a very small cache, about a few hours. In that way BOINC only keeps a GPU-Grid cache of about 1 day, which is enough to return them quick enough to always get the credit bonus (running 24/7 though).

MrS

---

Scanning for our furry friends since Jan 2002