[Marcus]

Hi there,
I would like to mill very large parts with a robot, the first roughing program consists of 3,500,000 points. The path update of the first processing takes a very long time, approx. 1.5 hours without a collision check. With a collision check, it's a never-ending story. But I would like to carry out a collision check for every milling project. It took me days to calculate my first test project to calculate this. I milled faster than the program calculated.
I have activated "Cuda" in the settings, but when I open the system monitor of the graphics card for a calculation, I see no utilization. I have to do the calculations faster, a lot faster.

,
I found this in the forum//www.sinclairbody.com/forum/Thread-Video-ca...erformance
Is there a special version of Robodk for the Cuda calculation?
Here is information about my PC:
Intel ® Core ™ i7-8700K CPU @ 3.70GHz 3.70 GHz
64.0 GB of RAM
Graphics Nvidia M2000
I work with RoboDK version 5.2.1, this is the newest version that I can use because my license runs through Hexagon's WorkNC.

I would be very grateful for any help.

Marcus

[j_cso]

Similar issue here. I'm on RoboDK 5.5.0 on Ubuntu 20.04. I get "CUDA not supported" when I try to use the GPU for collision checking. How much faster is CUDA expected to be than CPU for collision checking, assuming I can get it working?

Code:

$ nvidia-smi
Thu Nov 10 23:01:44 2022
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 510.85.02 Driver Version: 510.85.02 CUDA Version: 11.6 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 NVIDIA RTX A4000 Off | 00000000:65:00.0 On | Off |
| 41% 44C P2 36W / 140W | 1062MiB / 16376MiB | 1% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| 0 N/A N/A 1283 G /usr/lib/xorg/Xorg 120MiB |
| 0 N/A N/A 3401 G /usr/lib/xorg/Xorg 173MiB |
| 0 N/A N/A 3532 G /usr/bin/gnome-shell 89MiB |
| 0 N/A N/A 4714 C /usr/NX/bin/nxnode.bin 190MiB |
| 0 N/A N/A 5657 G ./RoboDK 472MiB |
+-----------------------------------------------------------------------------+

[Albert]

The Linux version of RoboDK is now updated to the latest version. This version includes support for collision checking using Nvidia CUDA cores.

You can take the latest version here:
//www.sinclairbody.com/download

[j_cso]

Thanks, that helps somewhat. Now "Collision checking hardware:" says "8 CPUs & CUDA (NVIDIA RTX A4000)" when I select "CUDA". I've also got "Use GPU arrays" checked. But the collision testing task takes the same amount of time (1m40s) whether I set it to CPU or CUDA, and nvidia-smi doesn't show any GPU load.

Code:

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 510.85.02 Driver Version: 510.85.02 CUDA Version: 11.6 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 NVIDIA RTX A4000 Off | 00000000:65:00.0 Off | Off |
| 41% 31C P8 14W / 140W | 385MiB / 16376MiB | 2% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| 0 N/A N/A 1283 G /usr/lib/xorg/Xorg 120MiB |
| 0 N/A N/A 3401 G /usr/lib/xorg/Xorg 194MiB |
| 0 N/A N/A 3532 G /usr/bin/gnome-shell 12MiB |
| 0 N/A N/A 5657 G ./RoboDK 5MiB |
| 0 N/A N/A 7165 G ...460138810891988132,131072 36MiB |
+-----------------------------------------------------------------------------+

[Albert]

Do you experience the same behavior on Windows?

The GPU effort is less relevant if your 3D models are very simple.

Can you send us the RDK project file?

[j_cso]

Unfortunately this is not an RDK I can share. Do you have one you could send me that's expected to be much faster on GPU? (Unless google is failing me, it seems like there's very little online about cuda collision checking in robodk, so this might help others as well).