I'm not sure if this is a bug or not (or indeed where to post if it is a bug).

I updated one of my crunchers from 7.0.28 to 7.0.44
The update appeared to go well, the benchmarks were run, and crunching restarted, with the elapsed and remaining times appearing to increment correctly. A couple of S@H cpu tasks reported and I left the room (bad move). Coming back I saw that one of the GPU tasks had ended in an error - exceeded time!
A quick glance down the list showed that all the GPU tasks were scheduled to take an optimistic 50s, so when a normal task take about 8-12 minutes it is obvious why the error was logged.

System is an AMD FX 8350 processor, 16GB ram, with a GTX 690 running Windows 7 64 bit. I'm running optimised apps, but only 1 WU per GPU (two total).

I've just reverted to version 7.0.28 and the estimated times have returned to normality and no more errors.

In 7.0.31 there was this addition:

client: when estimating FLOPS for an anonymous-platform app version for which no estimate has been supplied by user, use (CPU speed)*(cpu_usage + 10*gpu_usage) (--> add the 10*)

This means that all GPU work present in cache will err on the time exceeded 'buglet', due to its resource values being 10 times too little to allow for the tasks to finish in time.

Workarounds:
1. While running 7.0.28, set NNT on all projects, run your cache empty, upload & report everything. Next do the update to a later BOINC, then allow new work. The new work will not have this bug.

2. Only for the super advanced. Exit BOINC completely. Open client_state.xml file. Find the entry for Seti. Find the entry for Seti. Manually edit the <rsc_fpops_est/> and <rsc_fpops_bound/> values so they are 10 times their original size (add a zero). Do this for all GPU work in cache. Save file, restart BOINC.

If you really want to go test a later BOINC version, I'd go with workaround 1. It's the easiest.

Edited to correct what Claggy wrote below. So the advice is in one place.

---

2. Only for the super advanced. Exit BOINC completely. Open client_state.xml file. Find the entry for Seti. Manually edit the <rsc_fpops_est/> so they are 10 times their original size. Do this for all GPU work in cache. Save file, restart BOINC.

You'll also have to do that to <rsc_fpops_bound> too, otherwise tasks will continue to be on the verge of erroring out.

Claggy

Ah thanks, Claggy, for picking that one up. :-)

---