Show HN: Why Nvidia's PhysX is hitting a "SIM-to-Real" wall at 5000W
github.com
github.com
Current solvers burn thousands of GPU cycles just to prevent numerical 'explosions' during contact. It's a brute-force patch for a math problem that should be solved with algebraic continuity. We've pushed a framework that runs at ~100W with higher stability than a 5000W H200 cluster.
I’ve officially challenged the PhysX core team on their forums to address this systemic waste.