3 ms·
Ask HN: Robotics engineers – how painful was setting up GPU sim infra?
I'm exploring a service that auto-runs robot physics simulations on cloud GPUs when you push code, think Vercel CI/CD but for Isaac Sim, Gazebo, or MuJoCo. GPU spins up on push, runs headlessly, posts metrics to a hosted results dashboard, tears down.
For those of you who've already solved the GPU infra problem: if you had to start over from scratch today with a new job, new team, new project, would you dread what lies ahead? Or was it a one-time annoyance you've forgotten about?
Genuinely trying to understand if this is painful enough to pay to avoid or just annoying for a few weeks and then fine.
(I have a landing page but no product yet, posting to validate before building!)
- nikhilol 6mo ago[dead]
- gus_massa 6mo agoAre they prototypes of IRL robots? Does the system generate some kind of video of the simulated robots?
- nikhilol 6mo agoYeah, a good one to look at is Isaac Sim from Nvidia. It's basically like a game engine but you connect up your robot system and hook nodes in the system up to a model in the engine. It simulates physics like friction etc as well.
- gus_massa 6mo agoSomething like https://developer.nvidia.com/isaac/sim?size=n_6_n&sort-field=featured&sort-direction=desc https://developer.nvidia.com/isaac/sim?size=n_6_n&sort-field... ? I think that if you can write blog phostthat linksto a repo with a demo of your product and a few images/videos of the test, it will get more upvotes from the general public.
- nikhilol 6mo agoThanks for this idea, I'm still trying to validate before building here, but I think you're right that something like this would really help paint the picture!
- ghostnexus 6mo agoFormer ML engineer here who ran IsaacGym and MuJoCo sims in the cloud for 2+ years. The pain is real and very specific: 1. Cold start latency killed iteration loops. Spinning up a GPU VM to test a 10-minute sim run took longer than the sim itself — you'd wait 3-5 min for the instance, run 8 min, tear down. That per-iteration overhead crushes exploration. 2. Idle billing. If you're grid-searching over reward functions, you want to fire 20 parallel runs, collect results, tune, repeat — but most providers bill per-hour so even a 12-minute run costs you a full hour. 3. Physics sim + CUDA dependencies. Custom CUDA kernels (warp sim, etc.) often need specific driver versions. Docker helps but image build/push overhead adds another 5-10 min to the loop. The "CI for sims" framing (push code → run on GPU automatically) directly addresses #1 and #3. Worth building. On the infrastructure layer: we built GhostNexus (https://ghostnexus.net https://ghostnexus.net) to address #1 and #2 — per-second billing, <30s cold starts on RTX 4090 hardware, Python SDK with 3 lines to submit a job. Might be worth using as the GPU backend if you don't want to manage the infra layer yourself. (Disclaimer: I'm the founder.)
- ghostnexus 6mo ago[dead]
- nikhilol 6mo agoThis is really interesting, thanks for this insight. Ghost nexus looks awesome, would love to chat some more!
- k8kraze 6mo ago[flagged]