Pinned Tweet
100k spheres. 3ms/frame. 1920x1080. 5060Ti. Vulkan compute (glsl path tracer real-time) and C.
Isaac Dobbins retweeted
Connect your agents to your friends/co-workers in 5 steps with no new interface
Isaac Dobbins retweeted
UPDATE - All your agents can finally talk to each other…
Crazy I got to this so early.
If you're someone who runs multiple agents across your company and has agents doing many things in different departments, got to try this tool out.
@bridger_penn shared the sauce
@openpod__
The natural progression of a custom physics engine is a physics-based sea battle prototype. 0.6ms/step for the physics. I am feeling a game idea coalescing.
Testing a few things here. I’ve moved to a simple forward rendering pipeline to give as much gpu time to physics as possible. This renders now in 0.7ms and physics takes about 4ms. It’s not as fancy as real time ray tracing but if I want to target lower end hardware it’s probably the way to go. All fracturing is dynamic. No pre-split voxel chunks.
2
27
992
GPU-resident custom physics solver I have spent the last few days iterating on in my real time ray tracer. Runs at 30hz. 4.73ms average frame time on my 4060Ti. (3.39ms p1, 5.16ms p99)
Time to write it from scratch with all the learnings.
The carpet bombings will continue in real time until I can figure out how to beat the performance issues out of this one that can only run at 30fps.
10
1,252
My gpu-based physics solver, unoptimized and very rough v1 runs the physics for this scene runs in 2ms. It is only 467 moving rigid bodies at peak but it is all dynamic fracturing. I love the AC-130 by the way.
9
1,235
A side affect of having my own raytracing and physics is now I spend way too much time putting little scenes like this together.
8
580
I finally got voxel destruction working in my real time path tracer. This one uses box3d but I think I am going to look into my own simplified physics solver because I want to do something a bit different.
1
13
685
Ray coherency diminishes performance after the first bounce in every single ray tracer I have built to date. You can see here how the directions of each ray change drastically from first bounce to second bounce. 26ms on the first, 79ms on the second bounce. 8spp but this verision is not quite as optimized.
I made a Discord for people who care about real-time ray / path tracing.
discord.gg/Rk7FaEdGGc
8.96ms/frame for 8 samples per pixel in my real-time path tracer. I spent a good chunk of time today re-implementing it from scratch because the old implementation was just wasteful. I would say the simplification efforts are massively worth it.
2
15
1,352
I'll eventually turn my ray-tracer into a playable game, but for now random dioramas are what I have to fire into the algorithmic void. This renders in ~10ms or so on my M1. I am trying to explore how much ray tracing I actually need to do to make something look decent.
















