Re: [AMBER] Error in Installing Amber 20 on GPU

From: Gustaf Olsson <>
Date: Thu, 2 Jul 2020 12:56:29 +0000

Both Cuda-9.0 cuda-10.0 are in this path: /usr/local
This sounds strange and could be a cause of problems.

    $ cat /usr/local/cuda/version.txt
    $ nvcc —version
    $ which nvcc
    $ ls -la /usr/local/cuda/nvcc

If both versions (9 and 10) are installed, which one is linked from (usr/local/cuda), and which has priority in your path?

Multiple versions can be installed though only one can be the “default”. If cuda 9 is linked or has priority in your path, "setting cuda 10” during compile might not actually produce the desired result as it seems that the config script still finds cuda 9.

Best regards
// Gustaf

On 2 Jul 2020, at 14:22, Suchetana Gupta <<>> wrote:

Version of /usr/local/cuda is cuda 9.0.176
Both Cuda-9.0 cuda-10.0 are in this path: /usr/local
We set CUDA_HOME by giving the command
export CUDA_HOME=/usr/local/cuda-10.0
Suchetana Gupta
IACS, Kolkata

On Thu, 2 Jul 2020, 17:37 David A Case, <<>> wrote:

On Wed, Jul 01, 2020, Suchetana Gupta wrote:

We tried a lot of options and tinkered with the cuda versions etc. We set
the cuda home variable to cuda 10. Yet while the installation happens, it
is calling cuda 9. And it is calling for gcc 6. Our machine has gcc 7.5. I
am attaching the cmake output file and make install file in text form.
Please help us solve this.

We just need more information. Exactly *how* did you "set the cuda home
variable to cuda 10"? What version is in /usr/local/cuda? Where is
cuda 9 and where is cuda 10?

Unfortunately, the people that really understand the ins and outs of
cuda building generally don't follow this email list.


AMBER mailing list<>

AMBER mailing list<>

AMBER mailing list
Received on Thu Jul 02 2020 - 06:00:04 PDT
Custom Search