Reverse-Engineering the RTX 4090: A Deep Dive into GPU Memory Reads
CatAstro_Piyush · x · 2026-08-14
This in-depth technical post reverse-engineers the complete execution path of a global memory load instruction (LDG) through GPU hardware, specifically using an RTX 4090.
- Instruction Tracing: Starting with a simple vector-add kernel, it traces how the compiled SASS instruction computes an address in a register and requests data from memory.
- Hardware Hierarchy: It breaks down how a 4-byte request across 32 thread lanes traverses cache lines, address translations, crossbars, and ultimately reaches L2 cache slices and DRAM chips.
- Reverse Engineering: Because NVIDIA poorly documents many of these low-level details, the author runs timing experiments directly on the hardware to infer and verify the undocumented mechanics.
More from Infra
- What Is the Thermodynamic Limit on Energy Per LLM Token? — prateekj · 2026-08-14
- Prime Flash MoE: Blackwell-Optimized CUDA Kernels for MoE Inference — sloppenheimer · 2026-08-14
- YC-Backed Dipole Labs Uses Optical Switches to Solve AI Compute Idle Time — ycombinator · 2026-08-14
- Cloudflare: The Last Hope of the Open Web Against AI Monopoly? — thedealdirector · 2026-08-14
- TensorSharp vs. llama.cpp: Benchmarking Muse Glimmer 30B Locally — fuzhongkai · 2026-08-14
- Cerebras Teams Up with OpenAI to Massively Accelerate GPT-5.6 Inference — pr337h4m · 2026-08-14