Two Minute Papers · 2026-07-03 · notable
Two Minute Papers: 'They Said This Will Never Run In Real Time'
Two Minute Papers walks through JGS2, a GPU solver for elastodynamic simulation that hits near-Newton convergence while staying parallel — the authors report 50 to 100 times faster convergence than existing GPU methods.
Two Minute Papers covers JGS2, a GPU elastodynamics solver that closes the gap between Newton-quality convergence and Jacobi-style parallelism.
What is it?
'They Said This Will Never Run In Real Time' is a Two Minute Papers video on JGS2, a new GPU solver for elastodynamic simulation from Lei Lan and co-authors. The paper title, 'Near Second-order Converging Jacobi/Gauss-Seidel for GPU Elastodynamics', describes exactly what the video shows: soft-body and contact physics that used to need an offline Newton solver, now moving toward interactive rates on a single GPU.
How does it work?
JGS2 keeps the fully parallel structure of a Jacobi/Gauss-Seidel iteration — every element updates without waiting on a global linear solve — but re-derives the update so it converges nearly quadratically, the way Newton's method does. The paper reports 50 to 100 times better convergence than existing GPU methods on the same problems, which is what unlocks the real-time claim in the video title.
Why does it matter?
Real-time deformable simulation with stable contact is a long-standing wall in computer graphics: Newton is accurate but serial, and parallel iterative solvers are cheap but crawl to a good answer. JGS2 argues you can have both, which matters for games, interactive VFX, and any robotics or digital-twin loop that needs soft-body physics at frame rates instead of overnight.
Who is it for?
Graphics and simulation researchers, game and VFX engineers, robotics and digital-twin teams
Try it
Read the paper: https://arxiv.org/abs/2506.06494