Our WUs are all supposed to require approximately half day on a 8800GT (114 shaders), around 6 hours for a 280GTX. For a bigger system, we reduce the number of iterations to meet this target or we increase it for bigger systems. This is not exact of course, allow for up to 50% error.

Something more, the time estimate of the boinc client is usually wrong, but the progress bar is EXACT. It's the application to tell the client how much it completed. You can do very accurate estimates from that.

gdf

Hmm, on my 9800 GT the "typical" run time was about 17 hours.

I am not complaining understand, but the uncertainty can cause angst for participants. IN that we don't have anything other than the progress bar ( and thank you for that tip) this can be troubling.

Understand that projects are not ivory tower isolated. So, when you have a project like Sztaki that changes the task so that the run time jumps to hundreds of hours, with intermittant progress and task hangs, well, it can reflect on how *GPU Grid* is perceived by participants in task uncertainty. And lest you think I am singling out one project, I just stopped Rosetta as a focus project because I had nearly 15 tasks in the space of 24 hours have serious problems. To me, for a project that is classed as "production" this is seriously troubling.

I concede that this project with regards to Nvidia processing is in early days and that grants some leeway, all I can say is that *MY* experience is that task execution time is fairly variable from just a few hours to over 90 ... I have covered this in other posts in other threads so will not revisit that here, but, only use that to illustrate the point ...

Predictability is highly desirable ... :)

Well, until there is some kind of perportionality to the credit granting scheme, I'm only going to run this project when I can't get work from others.

I had another case yesterday where 3 units that took ~19000 second got granted 1887, and one wu that ran for ~17,500 seconds got 2424. The 3232 credit wus were running ~22000.

Later guys.

Guys, relax. We all know the 1888 credit WUs are off and GDF just told us the reason and that a possible long-term solution should arrive within the next days.

Nognlite wrote:

If it is a more advanced algorithm, then why are the credits lower? "It's based on FLOPS!" That great. I have seen no proof of that. It seem to be based on the number of steps in a WU.

Advanced means they implemented advanced functionality by means of a new library. It's advanced science, not necessarily more efficient code. Could be that the estimate for the flops done by the library is off or that the new code is less efficient (less flops per time) or that calling functions from that library leads to more overhead.

Of course the credits are based on the number of steps in a WU. But there's also the basic complexity of the structure (e.g. how many atoms), which determines the credit value of each step. And there's the estimate of the number of flops per step, depending on the scientific model. This is where the new library gets into the game.

Nognlite wrote:

There is no reason that there cannot be 1 type of WU based on the 8800GT efficiency and leave it at that.

Well, we don't want to solve the same system over and over again, don't we? The application needs to be flexible and thus has to be able to solve systems with varying complexity, which inherently leads to different WUs. And by the way, people with slower cards want smaller WUs.. ;)

Paul wrote:

all I can say is that *MY* experience is that task execution time is fairly variable from just a few hours to over 90

I've seen fairly constant run times on my box.

MrS

---

Scanning for our furry friends since Jan 2002

I've seen fairly constant run times on my box.

MrS

*MOST* of them now seem to be stable ... but I did get lucky with a few ... which you can see if you rummage my account ...

I am NOT complaining, I was just commenting ... :)