Initial condition

Here is the spec of the liquid-cooled GPU box that I recently purchased to meet the high-performance computing demands in my laboratory:

Motherboard:  Xeon E5-2600/1600 v3 C612 Chipset
CPU:               Intel® Xeon® Processor E5-2680 v4 (14-core, 35M Cache, 2.40 GHz)
Memory:         DDR4 ECC Reg SO-DIMM 128GB (= 4x32GB)
Storage:           2.5" SATA 6Gb/s Internal SSD 1TB
GPU:               2 x NVIDIA Tesla K80 24GB Passive Cooling PCI-E 3.0 x16 GPU
Cooling:          2-Phase Liquid Cooling Kit for GPU and CPU by Ebullient
OS:                  Ubuntu 16.04 LTS


Here are the problems that I found as soon as it arrived:

1. Can't login to Ubuntu using Unity
The vendor kindly installed Openbox which allows login to Ubuntu with no issues.

2. Older version of CUDA toolkit was installed
The current version of CUDA toolkit at the time of this writing is 8.0. However, toolkit 7.5 was installed.

3. Older version of CUDA driver was installed
The current version of CUDA driver for Tesla K80 at the time of this writing is 375. However, version 367 was installed.

4. MATLAB doesn't recognize NVIDIA Tesla K80
In MATLAB 2016b with Parallel Computing Toolbox;
>> gpuDeviceCount

ans =

     0
 

Oh, noooo!

The purpose of this blog is to document the solutions (and the struggles) so no one needs to waste their time trying to solve the same problems that I had.