It seems that there are a batch of new work units that are double the run time length of those processed a week ago. Has there been any information regarding this.

For me the double length work units mean some 9400GT's that could just barely complete a work unit are going to need to find a new project home.

The 9600GT cards will still (for now) work. The 9800GT cards now are going to take about 35 hours instead of 17 hours. I'd note that several weeks ago, instead of 17 hours, they were completing work in 12 hours -- so it seems there is 'work unit inflation' going on here.

That's my fault, I'm afraid. As a new user, I dared to point out that the project estimates for new joiners were way out of line: estimate 1hr 8min, reality 16 hours.

Existing users won't have noticed this, of course, because BOINC automagically adapts to it. Looked at your Duration Correction Factor lately? Mine's 15.4471

So Admin are gradually adjusting the estimates - but not the actual runtime - so that theory and practice come gently back into line - see message 11557.

Your BOINC will find that the task finishes much quicker than the (new) estimate. So it will adjust (slightly) downwards, and show a smaller estimate for the next task. But it will still be too high, so BOINC will adjust a bit more when it's finished. And so on: in a month or so, BOINC will be guessing pretty well.

And then Admin can do another step in the estimate adjustment....

OK -- we will see if the work completes - I'm running one on a 9400GT at the moment.

Then again, at the moment it appears there is a bit of a work fetch problem. I'm getting:

08/06/09 16:29:31|GPUGRID|Sending scheduler request: To fetch work. Requesting 2854385 seconds of work, reporting 0 completed tasks
08/06/09 16:29:36|GPUGRID|Scheduler request completed: got 0 new tasks
08/06/09 16:29:36|GPUGRID|Message from server: No work sent
08/06/09 16:29:36|GPUGRID|Message from server: Full-atom molecular dynamics on Cell processor is not available for your type of computer.


On my workstations -- although, according to the server status, there is plenty of GPU work available (this is with my 9800GT work stations - they are not out of work, but as work completes, they are not getting new work).

Thanks for the response -- but I really am seeing 'double length' work units.

Like I posted to start this thread -- I was seeing the 15 hour work units on 9800GT's until last week. Now there appear to be a mix.

On one workstation as an example (AMD 940 + 9800GT), I currently have four work units, including one which just started.

39-GIANNI_BINDX119-36-100-RND7920_0 == 29 hours
65-KASHIU_HIVPR_twomons_far_ba7_17-100-RND3118_0 = 29 hours

Then -- 'old length' for two others:

128-GIANNI_DOPc-1-25-RND8720_0 - 14.5 hours
387-GIANNI_DOPc-1-25-RND4847_0 - 14.5 hours

I would really like to see some information about the apparent change. Note, this is not just this one work station. I have several workstations -- including one with a 9600GT which has a similar time increase with the 'new version' work units. As to my lowly 9400GT work stations -- I'm doing Aqua work units on them as the *new version*GPUGrid work units don't have a prayer of completing within 120 hours.

So please, can we get project information on these rather than hearing that my reports are phantoms?

Thanks

I'd note that several weeks ago, instead of 17 hours, they were completing work in 12 hours -- so it seems there is 'work unit inflation' going on here.

I don't think this is the norm. Last summer me 9800GTX+ needed about 11.5 - 12h per WU (and still does so by now), whereas 9800GTs needed ~15h. I have not seen any tasks with <10h, which would correspond to a 9800GT needing 12h.

MrS

What you see for a task just started is meaningless - almost all of the "to completion" column will be an estimate.

Please post us a task link when you see one finish with these extended run times.

To me, I am 100% sure, that some WU's run 150% longer than usually.

Previous WU runs 6:30 - 7:30 - now 8:00 - 10:00
____________
POLISH NATIONAL TEAM - Join! Crunch! Win!