Re: compute capability

From: Josh Vermaas (vermaas2_at_illinois.edu)
Date: Mon Jan 25 2016 - 09:52:37 CST

Hi Stefano,

Drop the compute_11 from the compile line, as I think CUDA 7.5 dropped
support for the oldest stuff NVIDIA made. Its a bit hard to find
(arch/Linux-x86_64.cuda), but that is where the cuda generation code is.
Here for instance, are the contents of my file, which works for me with
version 7.0:

CUDADIR=/usr/local/cuda-7.0
CUDAINCL=-I$(CUDADIR)/include
CUDALIB=-L$(CUDADIR)/lib64 -lcudart_static -lrt
CUDASODIR=$(CUDADIR)/lib64
LIBCUDARTSO=
CUDAFLAGS=-DNAMD_CUDA
CUDAOBJS=$(CUDAOBJSRAW)
CUDA=$(CUDAFLAGS) -I. $(CUDAINCL)
CUDACC=$(CUDADIR)/bin/nvcc -O3 --maxrregcount 32 $(CUDAGENCODE)
-Xcompiler "-m64" $(CUDA)
*CUDAGENCODE= -gencode arch=compute_35,code=sm_35 -gencode
arch=compute_35,code=compute_35 -gencode arch=compute_50,code=sm_50
-gencode arch=compute_50,code=compute_50

*I've bolded the big change I had to make from the stock sources, which
should in theory be suitable for your machine too. Is there a particular
reason you are trying to compile NAMD 2.9 now? 2.11 is out and is awesomer!

-Josh Vermaas

On 01/24/2016 05:15 PM, Stefano Guglielmo wrote:
> dear all,
> I am trying to compile NAMD 2.9 with cuda (7.5) on a 32 core workstation
> running with centos 7 with an NVIDIA geforce gtx 750. During the make
> procedure I get the following error message:
>
> nvcc fatal : Unsupported gpu architecture 'compute_11'
>
> as the compute capability of the gpu is 5.0. Is there a way to fix?
>
> thanks in advance
>

This archive was generated by hypermail 2.1.6 : Sun Dec 31 2017 - 23:20:09 CST