Re: NAMD2.12b1 and CUDA

From: Giacomo Fiorin (giacomo.fiorin_at_gmail.com)
Date: Thu Nov 10 2016 - 11:09:04 CST

Hi Jeff, the error message seems quite clear. The new .arch file for
x86_64 only includes flags for CUDA capability 3.5 and later, you can
definitely recompile NAMD with 3.0 or earlier.

I tested this with two 680s on one node, and it's pretty awesome to be able
to use both GPUs and all CPU cores as well with good scaling.

Thank Antti-Pekka Hynninen at ORNL for the great work.

Giacomo

On Thu, Nov 10, 2016 at 10:09 AM, Jeff Comer <jeffcomer_at_gmail.com> wrote:

> NAMD2.12b1 runs on my GeForce GTX 960 GPUs but not on my GeForce GTX
> 770 GPUs. I tried adding "useCuda2 no" and "usePmeCuda no", but I
> continue to get this error:
>
> Charm++ fatal error:
> FATAL ERROR: CUDA error Error executing CUDA kernel
> read_CUDA_ARCH_kernel in file src/DeviceCUDAkernel.cu
> Error string: invalid device function
> Possible cause: Device compute capability is less than the compute
> capability the code was compiled for.
> on Pe 4 (conejo device 0)
>
> Is there a solution to this?
>
> Thanks,
> Jeff
>
>
> –––––––––––––––––––––––––––––––––––———————
> Jeffrey Comer, PhD
> Assistant Professor
> Institute of Computational Comparative Medicine
> Nanotechnology Innovation Center of Kansas State
> Kansas State University
> Office: P-213 Mosier Hall
> Phone: 785-532-6311
> Website: http://jeffcomer.us
>
>

-- 
Giacomo Fiorin
Associate Professor of Research
Institute for Computational Molecular Science (ICMS)
College of Science and Technology, Temple University
1925 North 12th Street (035-07), Room 704D
Philadelphia, PA 19122-1801
Phone: +1-215-204-4213
Scholar: http://goo.gl/Q3TBQU
Personal: http://giacomofiorin.github.io/
Lab page: https://icms.cst.temple.edu/members.html
*"As computer programmers we have a responsibility to make sure that we run
the computers instead of the computers running us."*  - Steve Oualline

This archive was generated by hypermail 2.1.6 : Tue Dec 27 2016 - 23:22:35 CST