Dear users, we are submitting several WUs (*EGF*) directly related with colorectal cancer research. We are collaborating with MDs & researchers at the Hospital del Mar here in Barcelona to characterize some previously unreported effects of a well-known drug for cancer. Thanks for your precious contribution.

Are these short, long or beta Wu's?

I'm currently processing a p27-IBUCH_4_wtEGF_110726-0-10-RND2134_1 WU, which is a long one.

I've got another piece of this very rare kind of long WU: p36-IBUCH_2_wtEGF_110726-0-10-RND6674_1

Hi, I have also tried one of these tasks lately, are longer and less credit. Greetings.

p1-IBUCH_2_cetumutEGFR_110912-0-10-RND2925_0
15 hours running time on my slower GTX480.
Got two others running, one of them will take 12h45m to complete on my GTX580 (my fastest card).
So these are really long workunits.

Application: Long runs (8-12 hours on fastest card)
So it's about 5% too long to fit the description - I think we can afford 'em a bit a slack on this one ;)
Anyway, it might take 12h exactly on a GTX580 optimized under Linux-x64.

There are many different task types so there will be variety in run lengths, and they don't all fit the same performance/credit scheme due to differences in the calculations and amount of CPU usage.

Application: Long runs (8-12 hours on fastest card)
So it's about 5% too long to fit the description - I think we can afford 'em a bit a slack on this one ;)

I'm not complaining about the running time, I wanted to say "at long last!"
But when this running time will become frequent, I have to reduce my task queue length, if I don't want to miss the 24h bonus. (and many others will have to do it too)

Anyway, it might take 12h exactly on a GTX580 optimized under Linux-x64.

Venec's Linux host completed two of these:
p10-IBUCH_4_cetumutEGFR_110912-0-10-RND9933_0 13h
p1-IBUCH_4_cetumutEGFR_110912-0-10-RND6052_0 13h04m
My (overclocked) GTX580 finished just as it was predicted:
p47-IBUCH_2_cetumutEGFR_110912-0-10-RND9228_0 12h45m

There are many different task types so there will be variety in run lengths, and they don't all fit the same performance/credit scheme due to differences in the calculations and amount of CPU usage.

Of course, but it would have been great to read about these really long workunits in the news section before they strike me (and others) with their running time.
Anyway, I did posted about them to welcome their length, and also to warn the others about it.

I share your concern about task lengths, after seeing several tasks take over 1day on a GTX260. However having a low cache has always been recommended for GPUGrid. Perhaps this time some notice would have been useful as these are so long but I am not sure we generally need an announcement every time a task has a different length. If we do then we need a table of different task types and expected completion times for different cards on different systems (XP or Win7). That would require a lot of maintenance.

Those 13h Linux tasks started and stopped 5 times during their runs, but I accept that if a task takes over 12h on the fastest card and operating systems then the default Boinc setting of a 0.5day cache would be undesirable even for the fastest card.

I usually keep a very low cache (0.01 to 0.20), even on my GTX470 Fermi systems. The only time I don't is when I am crunching the normal/short tasks, and even then I usually stick to 0.5 (even though it's not needed for GPUGrid, it stops having too many CPU tasks in the queue). When I notice Long tasks taking longer I drop back to between 0.01 and 0.05.

Perhaps this time some notice would have been useful as these are so long but I am not sure we generally need an announcement every time a task has a different length.

As I see this, in one end there are the enthusiast crunchers, who will adopt to anything the project they are crunching for will require without any communication from the project personnel. In the other end, there are the common crunchers, who like to be treated like a part of a community they are crunching for. I understand, that this project is more for the enthusiasts, than the common crunchers, but even an enthusiast like me likes to be treated as a part of a community from time to time :) So I think I would like to have announcements about new task types, even if it's only a single sentence.

If we do then we need a table of different task types and expected completion times for different cards on different systems (XP or Win7). That would require a lot of maintenance.

Good idea. I was about to recommend something like that. It wouldn't have to be complete, but it's a good basis for checking the performance of someone's system.

I think that the idea of Ignasi was to announce the new project EGF, not every single batch that it is put online for that project. Of course, this can be done as well. It should not require too much time.

gdf

Hello: I just finished one of these tasks (p20-IBUCH_5_cetumutEGFR. ..) on my GTX295 in more than 27 hours.

I think a commentary on these tasks (duration etc ...) before release, would have been interesting. Greetings.

...It wouldn't have to be complete, but it's a good basis for checking the performance of someone's system.

It should be possible to create a calculation for this, so that people could enter their GPU and work out how long a task should take; the GFlops peak are static constants. So with a known task type and given steps someone could create a calculator to display this. You could even account for OC/UC'ing, as this has a linear influence.

Hello: I just finished one of these tasks (p20-IBUCH_5_cetumutEGFR. ..) on my GTX295 in more than 27 hours.

This is my concern. A high end (GTX200 series) GPU taking over 24h to complete a task. It's too long to claim the 50% bonus. If I had known a few weeks in advance I might have even sold my GTX260 and replaced it with a Fermi, so that I could finish said tasks within 24h.

All that said, I'm sure Ignasi did not mean to create tasks that would not return within 24h or 12h for respective GTX200/GTX500 series GPU's, and there are limited numbers of these Looong Cancer tasks, so I don't want to bang on about it. In fact, as Zoltan said, I'm happy to see them in the Long format, if that helps, and I will adapt accordingly.

Good luck,

Presuming appropriate optimizations/recommended settings are in place, and stock cards.
Estimated runtime for X-IBUCH_X_cetumutEGFR tasks (Linux/XP, Vista/W7), ball-park estimates:

GTX 590 GF110 40nm CC2.0 2488 BoincGFlops peak (3359) 17.7h, 19.7h
GTX 580 GF110 40nm CC2.0 1581 BoincGFlops peak (2134) 14h, 15.5h
GTX 570 GF110 40nm CC2.0 1405 BoincGFlops peak (1896) 15.8h, 17.4h
GTX 480 GF100 40nm CC2.0 1345 BoincGFlops peak (1816) 16.5h, 18.2h
GTX 295 GT200b 55nm CC1.3 1192 BoincGFlops peak (1669) 27.3h, 30.4h
GTX 470 GF100 40nm CC2.0 1089 BoincGFlops peak (1470) 20.3h, 22.5h
GTX 465 GF100 40nm CC2.0 855 BoincGFlops peak (1154) 27.7h, 30.7h
GTX 560 GF114 40nm CC2.1 1263 BoincGFlops peak (1136) 26.3 29.1h
GTX 285 GT200b 55nm CC1.3 695 BoincGFlops peak (973) 23.9h, 26.4h
GTX 275 GT200b 55nm CC1.3 674 BoincGFlops peak (934) 24.96h, 27.3h
GTX 260-216 GT200b 55nm CC1.3 596 BoincGFlops peak (834) 27.9h, 30.4h
GTX 460 GF104 40nm CC2.1 907 BoincGFlops peak 768MB (816) 35.8h, 40.5h
GTX 460 GF104 40nm CC2.1 907 BoincGFlops peak 1GB (816) 36.6h, 40.5h
GTX 550 GF116 40nm CC2.1 691 BoincGFlops peak (622) 48h, 53h
GTS 450 GF106 40nm CC2.1 601 BoincGFlops peak (541) 55h, 61h

Note that dual GPU cards will return each task slower than the highest single GPU cards.
PS. Initially, these were going by old comparisons on old apps, so there was some discrepancies. I have since better adjusted for the GTX200 series cards. Drivers might also influence the situation. Linux on 6.14 with SWAN_SYNC might be slightly faster than on XP without (different app).

My XP machine with a GTX 285 card without the SWAN_SYNC 0 finished in 86,172.64 seconds or just under 24 hours.

http://www.gpugrid.net/workunit.php?wuid=2700413

After looking at a few tasks I think the chart might be out by about 27% for the GTX200 series cards. I will correct it below.

Bedrich Hajek, your task just about finished before the 24h mark, but because of your cache settings and possibly not using report tasks it did not get reported within 24h, so you only got 25% credit rather than 50%.

GL

Dear all,

I have stopped the *cetumutEGFR* batch. Sorry for that.
There was clearly an error in the length that made them twice as intended.

ignasi

*EDIT*

I am submitting *1_cmEGFR* only 51 WU literally half as long. Confirm that you see the reduction please and I'll submit the 200 remaining for this calculation.

By the way, under an experiment/project objective there may exist long and short WUs. Cancer-related workunits *EGF* have been run either in short or long acemd depending on system size and resurces availability basically.

Should we abort the *cetumutEGFR* tasks?

Abort them if they haven't started or you don't want to keep crunching them.
I am going to use anyway the ones that completed.

Thanks

I am submitting *1_cmEGFR* only 51 WU literally half as long. Confirm that you see the reduction please and I'll submit the 200 remaining for this calculation.

I got one running on my slowest GTX480 for 1h47m21s and it's at 27.376%, so it'll take 6h32m to complete.