# GPU computing in practice
The course notebook. Freely editable by the course participants. Here Filippo will have additional notes, here you put questions/comments.

* Sessions https://aalto.zoom.us/j/947220751
* Slides https://tinyurl.com/y7laxfyp
* Registration page https://scicomp.aalto.fi/training/scip/gpu-computing/
* if you want to be informed about this course progress, please, add your email through the registration page to the participants list, following the course with no registration is just fine as well
* This page URL https://hackmd.io/6G7aw9BiRFK2R57TOi3yDQ

Other useful links:
* Aalto Triton intro https://scicomp.aalto.fi/triton/
* GPU page https://scicomp.aalto.fi/triton/usage/gpu/
* accessing Aalto net remotely https://scicomp.aalto.fi/aalto/remoteaccess/
* UH wiki pages https://wiki.helsinki.fi/display/it4sci/Resources+for+Research
* TUNI https://wiki.eduuni.fi/display/tutsgn/TUT+Narvi+Cluster

## Session 1, Mon 1.12.2020 12-15
* Q: Ask here using this construction as a template. Or use Zoom's chat
* A: answer to your question will come here

* Q: I am not a gamer. Do gamer notice it when it is unrealistic?
* No :)
* good game design shifts the attention away from grpahics glitches

* Q: if I asked for an interactive node with GPU using this: `sinteractive -t 3:00:00 --gres=gpu:1 ` how I find which GPU type (out of 3) I am acually using?
* by default you get the first free, no way to predict which one, but you can specify the one you want, like `.. --gres=gpu:v100:1` or alike, see `sinfo -o '%30N %30f %30G'` (or `slurm features`) for the whole list of the available *gres* parameters
* `nvidia-smi`

* Q: I have compiled and make install everything under my environment, last line looks like this: `-- Installing: /home/krejcio1/wrkdir/Learning/GPUcomputing/env/lib/python3.6/site-packages/hoomd/example_plugin/update.py
(env) (base) [krejcio1@pe7 build]$`. But still I cannot find `hoomd` from python. `which python` : `/scratch/work/krejcio1/Learning/GPUcomputing/env/bin/python` .

>>> import hoomd
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/scratch/work/krejcio1/Learning/GPUcomputing/hoomd-blue/build/hoomd/__init__.py", line 13, in <module>
    from hoomd import version
  File "/scratch/work/krejcio1/Learning/GPUcomputing/hoomd-blue/build/hoomd/version.py", line 36, in <module>
    from hoomd import _hoomd
ImportError: cannot import name '_hoomd'

* Do you have any ideas please?
* working after removing the anaconda (base)

* For anyone trying this at home, I've found that under Linux Mint you need to specify where hoomd should look for the cuda libraries. In particular, I had to add the following line to `FindCUDALibs`: `SET(CUDA_LIB_PATH "/usr/lib/x86_64-linux-gnu/")`
* the ccmake in advanced mode (t) will show options for all paths to cuda libraries

## Session 2, Thu 3.12.2020 12-15
sides: http://tinyurl.com/yb85e2ep

- Q: Is this also possible to write this in Python, please?
- A: since we will move to GPU programming in CUDA, it is better to stick to C. For this assignment I will accept also python.

- Q: Isn't the example code missing a call to https://docs.nvidia.com/cuda/cusolver/index.html#cuSolverDNdestroy ?
- A: yes I forgot about it!

## Session 3, Tue 8.12.2020 12-15
sides: http://tinyurl.com/ybqu4jqr
files: https://www.dropbox.com/sh/40vguwoca3qgnlc/AABBJhWLHo0K3BQLFyj1Xfb0a?dl=0
Q: fgci-common repository does not have GCC version 6.5.0 available. This makes the compilation rather impossible for non-aalto students.
Eg. module load fgci-common
followed by loading of the required modules. At the moment the version on fgci-common is gcc 9.2.
A: other GCC versions will work as well. HOOMD complains about it if too new or too old. All other tutorials and assignments do not strictly require a particular GCC version.

## Session 4, Mone 14.12.2020 12-15
slides: http://tinyurl.com/yd53ayxf