I have 2 GTX 260 216 190.62, 6.6.38 Vista 64 and things have been perfect since update.

I allowed my machine with dual GTX260's to pickup a couple of work units now we have the acemd 6.71 app. I will report how it goes.

I noticed that it uses plan class cuda now and the cuda 2.2 dll's instead of the cuda23 plan class (along with appropiate dll's). Is this an intended part of the update?

---

BOINC blog

yes,
cuda2.3 did not solve any real bug of cuda fft.

gdf

yes,
cuda2.3 did not solve any real bug of cuda fft.

gdf

It doesn't seem that 2.2 does either. I had a 50% success rate. One worked and one failed.

---

BOINC blog

no, cuda2.2 does not. At the end is probably a driver issue.
We will see if one day they decide to solve it or we have to wait for gtx260 to phase out...

gdf

no, cuda2.2 does not. At the end is probably a driver issue.
We will see if one day they decide to solve it or we have to wait for gtx260 to phase out...

gdf

4-5 years wait ... I mean they may stop selling them in a year, but they are likely to be about in people's systems for far longer ...

Heck, I am kinda waiting for the next generation to come out so that the 260 cards and the like get pushed down to replace the 9800 series which seemed to be about the low end card that Frys was selling this weekend ...

Looks like the new application version also broke GTS250 support for me. Now every work unit fails with

"Cuda error: Kernel [pme_fill_charges_overflow] failed in file 'fillcharges.cu' in line 97 : unknown error."

or some other unknown error after few hours runtime. Only thing that changed was the application version.

Oh well, I guess I disable this project until your application gets fixed or there's a stable version out for the recommended BOINC version (assuming that'll help). :(

i will admit, 190XX+ are unstable for public use...

no, cuda2.2 does not. At the end is probably a driver issue.
We will see if one day they decide to solve it or we have to wait for gtx260 to phase out...

gdf

4-5 years wait ... I mean they may stop selling them in a year, but they are likely to be about in people's systems for far longer ...

Heck, I am kinda waiting for the next generation to come out so that the 260 cards and the like get pushed down to replace the 9800 series which seemed to be about the low end card that Frys was selling this weekend ...

Expect the next generation to start being available in the next few weeks, at least for the basic chip (GT300), although in rather small numbers. I've seen guesses that it will be early next year before many cards using it are available.

Nvidia has announced that they plan to start offering some new compilers at about the same time, including some for C++ and Fortran, so expect some competition from other BOINC projects that decided to wait until compilers for those languages were available. I haven't seen a clear statement yet on whether those new compilers will also work for the older chips they already sell.

A recent version of acemd from GPUGRID and a recent version of rosetta_beta from Rosetta@home may have a compatibility problem; if not, the rosetta_beta graphics portion appears to have frozen by itself.


9/24/2009 4:39:22 PM CUDA device: GeForce 9800 GT (driver version 19038, compute capability 1.1, 1024MB, est. 60GFLOPS)

9/24/2009 4:39:35 PM rosetta@home Restarting task Rossmann2X3_002_14911_14657_0 using rosetta_beta version 598
9/24/2009 4:39:38 PM GPUGRID Restarting task PMEno54-OTTO_HERG4-10-40-RND5579_0 using acemd version 671

Today, I saw the graphics portion of a rosetta_beta workunit freeze in a way that kept it from ending its screensaver function when I used the keyboard and mouse.

Have now seen essentially the same problem with a second rosetta_beta workunit, and also with a workunit from QMC@home.

A new guess on what the problem is: The newer versions of BOINC come with a screensaver that automatically displays graphics from one of the workunits that are running, waits a while, then moves those graphics to a different place on the screen. So far, the three times I've seen this problem, it's been while showing graphics that cover the whole screen, leaving no place to move them to. Therefore I'll guess that the new screensaver is unable to deal with situations when it is time to move the graphics, but the graphics are too big to leave anywhere else to move them.

If this is correct, Docking@home is yet another project likely to trigger this problem, even though I haven't seen any Docking@home workunit trigger it yet. Also, many of the GPUGRID participants who also participate in any of these three other BOINC projects are likely to need instructions on how to keep the new screensaver from running at all.