Massive AI performance engineering repo launches, featuring full GPU MODE lectures
ycombinator · x · 2026-10-08
waferai launched what it calls the most comprehensive AI performance engineering repo. Part 9 collects GPU MODE's lectures, slides, notebooks and code covering profiling with PyTorch/Nsight, kernel memory coalescing, shared-memory tiling, numerical precision, GPU communication and runtime scheduling — a full curriculum for diagnosing slow training and inference.
More from Infra
- Blockway Ships Agens Volundr 32B: Only 18 Layers Keep KV Cache for Long Context — lmoroney · 2026-10-08
- Yaroslav Bulatov Traces the Origins of the Memory Wall in New Research Writeup — yaroslavvb · 2026-10-08
- Two Years Building a Local Agent OS: Multi-Node Routing, Memory Arbitration and Failure Governance — Budget_One_8784 · 2026-10-08
- DARPA QBI finalists revealed: six quantum approaches, up to $300M each through 2029 — hardimanjames · 2026-10-08
- DCVC recalls backing Atom Computing from pitch deck to DARPA QBI finalist — hardimanjames · 2026-10-08
- Rust+Vulkan Inference Backend Goes Cross-Vendor: Pick Kernels by CPU ISA, Not GPU Vendor — PhysicsDisastrous462 · 2026-10-08