cuOpt software for Jetson

AI & Data Science NVIDIA cuOpt cuOpt
1

Hello,

I can success to run cuOpt with docker image(amd64) of bellow NGC.
https://ngc.nvidia.com/containers/ea-reopt-member-zone:ea-cuopt

Question:
Are there any plan for cuOpt software for Jetson?

We cannot find cuOpt docker image for Jetson in below site.
https://ngc.nvidia.com/containers/ea-reopt-member-zone:ea-cuopt

If you have plan for cuOpt software for Jetson, please tell us.

Best regards,

2

Hi,

cuOpt on Jetson is on the roadmap and there is no firm date yet. Please share additional details on your use case with Jetson so we can re-assess.

Thanks,

3

Hi,

Thank you very much for your kindness.
I really appreciate your help.

We are in promotion phase.
So, we 'd like to prepare FAQ answer for our promotion like this post.
And we’d like to share our customer’s use-case in our promotion activity.

P.S. We also consider to prepare below log description.

The software showed a 2.96 percent accuracy average gap, compared to the world’s best known solution for the number of vehicles, in early tests on the Gehring & Homberger benchmark.

Question:
Are there any clue to explain like below question?
-Why is so high-accuracy result? (Are there special technique in cuOpt?)
-How to measure the 2.96 percent accuracy?

Best regards,

4

Hi,
cuOpt is built on CUDA primitives which executes solutions in parallel leveraging the power of GPU.

• Multiple parallel initial solutions are generated, and an initial route is generated for each solution, also in parallel.
• A CUDA block is assigned for each of the different solutions. To find the best insertion, we iterate over the current route and compare each of the insertion positions.
• Heuristics and metaheuristics are used to further improve the solution to reduce the total distance as well as prevent restrictive greedy approach.
• Local searches can start from different positions in the search space so space is explored thoroughly and structural constraints of data are handled easily.

For further details on accuracy - please refer to the GTC tech talk:
Accelerated Vehicle Routing Algorithms [S41708]

Thanks,

5

Hi,
Thank you very much for your kindness.
I really appreciate your detail explanation.

I continue to study cuOpt from your information.
And I think “Local searches can start from different positions” is one of uniqe Key-feature of cuOpt.
So, I have one question like below;

Question:
I found below description in “HOMBERGER DATASET EXAMPLE”
“set_number_of_climbers(2048)”
It means " Local searches can start from 2048 different positions."

Is this understanding correct?

Best regards

6

Yes, your understanding is correct.

7

Hi,
Thank you very much for your kindness.

Best regards,

8

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.