Specs probably too small

[rmast]

I have an old AMD X4 646 and 16GB. Running with CPU did not hog the CPU, so something went wrong.

My father has a new Windows 11 PC with 8GB. Running with CPU did not hog the CPU, so something went wrong.

Finally I decided to rent a Azure NC6_Promo with K80 cuda. That does the job after a reboot for installing the K80.

[rmast]

Even on that near empty K80 8GB is just too short

[rmast]

It's already running more than an hour on quadrupalizing a 300 dpi A4, so the 6 core K80 is no luxury.

[rmast]

After one and a half hour the memory footprint has more than doubled:

[rmast]

Next morning still a bit more memory... Glad I disabled Windows update...

[rmast]

Used memory still growing a bit:

[Djdefrag]

Wow! Thank you for the comprehensive tests that you did! I suppose that Pytorch-directml is not well optimized with CPU, in my tests i found that even when i set more cpus, it just use 1. The only thing i can say is that this library is still in alpha state, so maibe in future will be better optimized.

:D

[rmast]

The K80 contains a GPU, with CUDA codes sm_30, sm_35 and sm_37. It might be ignored by the used software.

[Djdefrag]

Yes, sadly the library just ignore this GPU

[rmast]

How long do you estimate a 300 dpi A4 will take with only 6 cores?

[rmast]

Do you know an Azure VM which supports DirectML?

[Djdefrag]

I estimate from 15 to 30 minutes

[Djdefrag]

Sorry, i m not pratical with Azure stuff

[rmast]

DirectML should support Kepler according to the readme: https://github.com/microsoft/DirectML

[Djdefrag]

Yes, but i suppose maibe they refers to commercial GPUs like GTX

[Djdefrag]

Just to be sure, can you try first a x0.5 upscale or x1 to check if everything works fine, this must be quick

[rmast]

Just broke off the long run and started x0.5

[Djdefrag]

oook thanks, i was thinking did you installed nvidia drivers on the vm?

[rmast]

Should the tool nvidia-smi be on Windows? It can't find it.

[rmast]

There was a popup stating K80 driver installed.

[rmast]

I'm now installing the Tesla-driver from the Nvidia-site.

[Djdefrag]

yes, maibe that will do the catch because when you screened the task manager i saw that there was no GPU panel. That usually happen when the driver is not installed

[rmast]

yes, maibe that will do the catch because when you screened the task manager i saw that there was no GPU panel. That usually happen when the driver is not installed

That feature only works with WDDM 2.0 I understand from this video. https://www.youtube.com/watch?v=gOo73cyeMUU

The K80 has 1.3.

[rmast]

I've restarted the x1 upscale after a reboot. nvidia-smi responds.

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 473.47       Driver Version: 473.47       CUDA Version: 11.4     |
|-------------------------------+----------------------+----------------------+
| GPU  Name            TCC/WDDM | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  Tesla K80           TCC  | 00000001:00:00.0 Off |                    0 |
| N/A   32C    P8    35W / 149W |      9MiB / 11448MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

CPU is high, nvidia-smi states "No running processes found"

[Djdefrag]

Really strange, so if you upscale on gpu it does nothing?

[Djdefrag]

Ah, also did you try install all Visual Studio runtimes? https://www.techpowerup.com/download/visual-c-redistributable-runtime-package-all-in-one/

[rmast]

I'm downloading/uploading those other VC runtimes. There might be some truth in this thread: https://discuss.pytorch.org/t/ubuntu-what-version-of-cuda-pytorch-etc-can-run-on-a-nvidia-gtx-680-compute-capability-3-0/118469/4

The files mention 3.7, however there might be some issues with 3.7. The Tesla-driver I downloaded is supporting Cuda 11.4, so that might be introducing other issues.