Using hashcat with CUDA Toolkit 8.0 installed.
#1
I have a CentOS 6.8 machine that will need to be able to run CUDA code as well as Hashcat tasks.  We installed 2 Tesla P100s in the box and it works fine for CUDA e.g. HPL but errors out when hashcat is run.  I could go back to hashcat 2.x if that is the best route but was wondering if anyone has gotten 3.0 to work on a machine with the toolkit installed.


Can I recreate the makefile using nvcc as the compiler instead of gcc?
#2
Code:
1944:scc@mic:~/3/hashcat-3.10>./hashcat -m 500 -b
hashcat (v3.10) starting in benchmark-mode...

OpenCL Platform #1: Intel(R) Corporation
========================================
- Device #1: Intel(R) Xeon(R) CPU E5-2643 v3 @ 3.40GHz, skipped

OpenCL Platform #2: NVIDIA Corporation
======================================
- Device #2: Tesla P100-PCIE-16GB, 4070/16280 MB allocatable, 56MCU
- Device #3: Tesla P100-PCIE-16GB, 4070/16280 MB allocatable, 56MCU



ERROR: clCreateContext(): CL_UNKNOWN_ERROR

Is the output I get when running ./hashcat -b when compiling with gcc or icc.
#3
Can you run any other OpenCL application?
#4
We were able to run other OpenCL code on the P100s when compiling with GCC but not ICC.

We are using the 361.93.03 driver not sure if that is a problem?
#5
If clCreateContext() usually does not fail. It's a very first step to initialize the OpenCL runtime. I'd guess the P100-PCIE-16GB does not support OpenCL correctly or the driver, but this is not a hashcat error.
#6
Just to confirm I setup a fresh CentOS 7 install and only installed the GPU drivers and the required items in BUILD.md and still have the same issue.

Is this the extent of what would need to be done normally on a fresh system?