Hi,

I'm running GPUgrid on a Debian Wheezy x64 box with a GTX950 (ID: 158410). Recently I've started getting missing GLIBC compat errors with acemd.848-65 and acemd.847-65, and most (if not all) GPU WUs fail.

$ LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/u/boinc/projects/www.gpugrid.net ./acemd.848-65.bin
./acemd.848-65.bin: /lib/x86_64-linux-gnu/libm.so.6: version `GLIBC_2.15' not found (required by ./acemd.848-65.bin)
./acemd.848-65.bin: /lib/x86_64-linux-gnu/libc.so.6: version `GLIBC_2.15' not found (required by ./acemd.848-65.bin)
./acemd.848-65.bin: /lib/x86_64-linux-gnu/libc.so.6: version `GLIBC_2.14' not found (required by ./acemd.848-65.bin)

$ strings /lib/x86_64-linux-gnu/libc-2.13.so |grep GLIBC
GLIBC_2.2.5
GLIBC_2.2.6
GLIBC_2.3
GLIBC_2.3.2
GLIBC_2.3.3
GLIBC_2.3.4
GLIBC_2.4
GLIBC_2.5
GLIBC_2.6
GLIBC_2.7
GLIBC_2.8
GLIBC_2.9
GLIBC_2.10
GLIBC_2.11
GLIBC_2.12
GLIBC_2.13
GLIBC_PRIVATE
GNU C Library (Debian EGLIBC 2.13-38+deb7u10) stable release version 2.13, by Roland McGrath et al.
$

The last acemd app running fine was 821-60.

$ LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/u/boinc/projects/www.gpugrid.net ./acemd.821-60.bin
# ACEMD Molecular Dynamics Version [3212M]
$

Can the apps be updated to support these *cough* legacy systems?

Thanks,
-w-

The acemd app is recompiled to incorporate cuda enhancements available from the cuda dev kit (as and when they do anything useful for the acemd app). Typically acemd updates facilitate new card generations or cuda updates which improve performance.

The most recent acemd app is 8.48 and supports cuda6.5. However cuda 6.0 is still supported via the 8.21 acemd app version (AFAIK).

You are presently trying to use the v8.48 (cuda65) app to run GPU tasks, and that's not working.

Suggest you first try to look for system updates and lib updates and do a cold start or two.
I'm getting the impression that your NV drivers are not properly installed, so you might want to try to reinstall them.
If you can't find/update any system/lib files or reinstalling the same drivers doesn't work, consider changing from your 358.16 drivers to 361.28 (released 2016.2.9) or as a last resort go back to the 8.21 drivers you were previously using (last choice and I'm assuming these still work with the present WU's).

Good luck,

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Hi,

On the lib updates, I have to slightly disagree - this is not a desktop box where I can do a dist-upgrade whenever I fancy, but a server-type machine where the OS release is somewhat set in stone. The fact that it has a GPU installed is to allow it to run 24/7 in a co-lo environment. If the requirement for newer GLIBCs come from the CUDA dev it, OK, fine, but then it sounds like a bad decision on nVidia's side.

For the driver version, the host had 361.28 installed, then reverted back to 358.16 - the acemd.848 app didn't work with it either obviously, as the GLIBC issue is in userland, and in my understanding it doesn't have to do anything with the kernel driver (apart from the supported CUDA version in the driver).

To move things forward, is there a way to force using the older acemd.821 binary?

Thanks,
-w-

In theory, if you only had an earlier driver then only 8.21 would try to run. Assuming you could remove the GPU drivers and install older versions you should be able to reset GPUGrid or detach and rejoin GPUGrid to get rid of the 8.48 app.
Unless you can build an app_info file to do so, I'm not sure how you would otherwise instruct to exclusively use 8.21 for GPUGrid. Maybe a dev could tell you.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help