I think one of the reasons that these WUs are so slow is that they run at 20% lower GPU than the other WUs. WinXP-64, v258.95 drivers, v6.11.6 BOINC.

These WUs made me overclock my CPUs. I think these WUs need more double precision calculations than other WUs, and this slows them down about 20%.

MilkyWay is heavily DP and it runs all the GPUs at 99% so I'm not too sure about that as a hypothesis. Good idea though.

I think one of the reasons that these WUs are so slow is that they run at 20% lower GPU than the other WUs. WinXP-64, v258.95 drivers, v6.11.6 BOINC.

These WUs made me overclock my CPUs. I think these WUs need more double precision calculations than other WUs, and this slows them down about 20%.

MilkyWay is heavily DP and it runs all the GPUs at 99% so I'm not too sure about that as a hypothesis. Good idea though.

But MilkyWay only runs on double precision GPUs, not on mine for example, while I get those monsters here.

---

Gruesse vom Saenger

For questions about Boinc look in the BOINC-Wiki

I think one of the reasons that these WUs are so slow is that they run at 20% lower GPU than the other WUs. WinXP-64, v258.95 drivers, v6.11.6 BOINC.

These WUs made me overclock my CPUs. I think these WUs need more double precision calculations than other WUs, and this slows them down about 20%.

MilkyWay is heavily DP and it runs all the GPUs at 99% so I'm not too sure about that as a hypothesis. Good idea though.

But MilkyWay only runs on double precision GPUs, not on mine for example, while I get those monsters here.

So then the WUs here must not be using the "DP units" of the GPU.

Maybe these are older type tasks that are not as readily optimized. Apparently the level of optimization they have now required them to be larger tasks, or they would be even slower to run if they were smaller.
They are for a paper, which may need a bit more work; when you try to publish a paper the moderator/publisher often gets back to you asking for you to include more results in a certain area.

NVidia’s are not great for dp, so they might be doing most of the dp work on the CPU here; most CPUs are reasonable good at it, but not all cards are. Saenger’s GT240 for example is not up to any dp work. It's only CC1.2; the MW requires CC1.3 or above cards as they have better dp ability.
The GF100 Fermis had much of this disabled by NVidia. Not sure what the GTX460's have to offer in terms of dp.

We will try to change the BOINC software to issue them to fermi only.

gdf

You haven't succeeded so far ;)
My last one is another of those monsters, but as I just became aware of it, after 13:30h of crunching already done, I will definitely not abort it.

x2... :-( Like Saenger, too far into this one (~ 11 hours) to abort.

---

Can anybody tell me what happened with this WU?
And this is a stange one too, but it's not TRYP, but it took so long, so that at first I thought it's one of them.

They ran and both finished, but while some tasks are longer than others, such as that Tryp task (for which you missed out on the 50% bonus, but picked up the 25% bonus), they both took too long and there was time issues.

When I looked at your other tasks I saw this TONI failure.

If you suspect there is a problem with a task (taking too long), look at the amount of GPU utilization and the temperatures (GPUZ). If they are not right, restart your computer. See this post.

# Time per step (avg over 297000 steps): 102.830 ms (This looks wrong - too long)
# Approximate elapsed time for entire WU: 82263.636 s !
called boinc_finish

The run time is different in two different places (for both tasks),
http://www.gpugrid.net/workunit.php?wuid=1889569 138,717.20
http://www.gpugrid.net/result.php?resultid=2982448 183265.009 s

http://www.gpugrid.net/result.php?resultid=2946115
http://www.gpugrid.net/workunit.php?wuid=1857373

I cant tell if the task (or even system) restarted a few times during its run, or if your GTX480's went into sleep mode, but I would guess the TONI task that failed messed things up, as it did with me and others.