No, the posted difference referes to the E@H app. I wanted to take further tests, but currently two wu faild after ~75 min.
This is the reason why I want to replace my unloved GTX260.
Anyway, nice to hear that there is no difference with the ATI cards. I cannot test that since my ATI-System is not dual boot and I don't want to do too many experiments with it; I need this system correct running.

It's really interesting, that so many issues regarding nVidia seem to be not existend on the ATI-side. Lets see what's coming up with the OpenCL apps.

Alexander

It's really interesting, that so many issues regarding nVidia seem to be not existend on the ATI-side. Lets see what's coming up with the OpenCL apps.

The MW app for ATI is hand-crafted. This is possible because the code is comparably simple. You can't do this with an app as complex as GPU-Grid, or not within a finite time. That's why you have to rely on external libraries here, which makes things driver dependent and more vulnerable to errors which you can not control / fix yourself.
Try doing the same with ATIs and you'll probably run into similar problems. At least ;)

MrS

---

Scanning for our furry friends since Jan 2002

Hello everyone,

after my FERMI GTX 480 finally seems to really having picked up the pace on XP X64 with Swan_Sync=0, I have some questions though:
1. Why is this performance not possible until now under Linux X64? Is this solely because the drivers aren't so well-crafted for Linux? Used the newest available from NVidia and the one coming from Linux itself.

2. Now under XP, BOINC suddenly reports 1345 GFlops peak, instead of 778 GFLops before. The latter can't simply be driver overhead on Win 7 X64, can it?

3. Why does a WU running XP X64 crash seriously like this one? Just thought I'd contact you as the stderr output says to do so. Annoying, as it was nearly finished...

Thx and cheers,

1. You ran two Cuda 3.0 tasks under Linux (6.06) and not CUDA 3.1 tasks for Fermi. You used the 2.6.32-24-generic driver. The latest is 256.44.

2. Boinc 6.10.17 (old Linux ver) vs 6.10.58 (latest Win ver). You did not have your GTX480 on W7.

3. Your error message, "The system can not find the path specified" appears to be a one off event. Perhaps the message will help the developers.

Thanks,

Hi skgiven!

1. Yeah, I only ran two tasks under Linux X64 and I know that is not representative at all. But I was fairly disappointed with the performance, that's why I tried XP instantly. I hate to contradict you, but I used both the generic drivers (later and not at the same time) and NVidia's 256.44 first, cause i knew that the Fermi has a performance boost with new drivers.
I tend to look first, if there's a new driver for any of my hardware. Both were disappointing, to say the least.

2. To be honest, I'm not a Linux geek, more a Linux newbie. So I didn't figure, how to install the newest BOINC for Linux AND getting it to run. The manual to install on Linux from the BOINC website wasn't helpful at all for me.
And I did have my GTX 480 running under Win 7, because I don't have any other performant NVidia-Graphics-Adapter aside from an ION, that crunched exactly a single WU, but then was too slow to report a second WU in time.

3. I hope, that this is helpful in further development; so no prob and you're welcome!

Cheers,
Christopher

Hi Christopher Herr,

I made this N00B guide, I use it myself when I install/reinstall or upgrade Mint Linux 8 (Ubuntu 9.10). If UR using a higher version of Mint or Ubuntu, the Nvidia driver upgrade steps won't work, but the steps to upgrade BOINC still works. I myself can't get higher versions of Mint/Ubuntu to run stable on my hardware. I'm told that the steps I use are time consuming, (but it's just clip n paste so it's not that time consuming).

---

Hello liveonc,

Thanks a lot! Will try that the next time I tinker with a full Linux install.

All the best,
Christopher

---