Revealing NVIDIA Closed-Source Driver Command Streams for CPU-GPU Runtime Behavior Insight
- Auto-discovered
Yuang Yan, Ian Karlin, Ryan Grant
Visit resource ↗Description
Revealing NVIDIA Closed-Source Driver Command Streams for CPU-GPU Runtime Behavior Insight recovers hardware command streams emitted by NVIDIA's closed userspace driver without source access.
Instrumentation on the open-source kernel driver watches GPU doorbell mappings to capture what the proprietary stack submits to the hardware.
Case studies cover CUDA data-movement DMA modes and how CUDA Graphs shrink launch overhead through smaller command footprints, aimed at systems researchers studying middleware behavior.
The work assumes advanced GPU driver knowledge, CUDA runtime internals, and comfort reading kernel and userspace interaction traces.
GPU systems venues frequently reference this style of reverse engineering when teaching performance attribution on closed stacks.
Link last verified accessible on 2026-09-28.
Related resources
- Drawing Thousands of Meshes with DrawMeshInstanced / Indirect (Toqoz)
Toqoz's November 2019 toqoz.fyi post walks through drawing very large mesh counts in Unity with Graphics.DrawMeshInstanced and Graphics.DrawMeshInstancedIndirect, with C# samples,
100% match · <code><advanced><cpu><data><gpu><kernel><last><movement> - High-Performance Real-Time Implicit Strand-Based Hair Rendering via Software Rasterization
High-Performance Real-Time Implicit Strand-Based Hair Rendering via Software Rasterization presents a deferred software rasterization pipeline for real-time strand-based hair…
100% match · <code><optimization><advanced><paper><accessible><hardware><last><link> - Optimizing CUDA like a Human: Micro-Profiling Tools as Expert Surrogates for LLM-Based GPU Kernel Optimization
Optimizing CUDA like a Human: Micro-Profiling Tools as Expert Surrogates for LLM-Based GPU Kernel Optimization.
100% match · <code><optimization><advanced><paper><accessible><cuda><gpu><hardware> - AMD GPU-Initiated I/O
AMD GPU-Initiated I/O is a March 2026 systems blog post by Abdelhadi exploring direct AMD GPU to NVMe communication instead of CPU-driven storage I/O. It explains NVMe queues…
100% match · <code><optimization><advanced><cpu><dma><gpu><hardware><mapping> - Ray Tracing using HIP
Ray Tracing using HIP is an AMD technical report (No.
100% match · <code><paper><accessible><cuda><gpu><hardware><last><link> - Making GPUs Actually Fast: A Deep Dive into Training Performance
Making GPUs Actually Fast: A Deep Dive into Training Performance is a Jane Street YouTube talk by Corwin and Sylvain on keeping GPUs busy while training neural nets in…
100% match · <code><optimization><cpu><cuda><gpu><launch><nvidia><performance> - CudaForge: An Agent Framework with Hardware Feedback for CUDA Kernel Optimization
CudaForge: An Agent Framework with Hardware Feedback for CUDA Kernel Optimization presents a training-free multi-agent workflow for CUDA kernel generation and optimization by…
100% match · <code><optimization><advanced><paper><accessible><cuda><gpu><hardware> - Practically Utilizing Neural Networks in CPU-based Production Rendering
Practically Utilizing Neural Networks in CPU-based Production Rendering presents a Journal of Computer Graphics Techniques paper by Steve Bako, Ryusuke Villemin, and Magnus…
100% match · <code><optimization><advanced><paper><accessible><cpu><gpu><last> - Optimizing spatiotemporal variance-guided filtering for modern GPU architectures
Optimizing spatiotemporal variance-guided filtering for modern GPU architectures.
100% match · <code><optimization><advanced><paper><accessible><gpu><hardware><last> - CUDA Tile IR
CUDA Tile IR is an NVIDIA open-source GitHub repository for an MLIR-based intermediate representation and compiler stack aimed at tile-based CUDA kernel optimization on NVIDIA…
100% match · <code><optimization><advanced><aimed><cuda><gpu><kernel><nvidia> - Towards Robust Agentic CUDA Kernel Benchmarking, Verification, and Optimization
Towards Robust Agentic CUDA Kernel Benchmarking, Verification, and Optimization presents a Sakana AI paper introducing robust-kbench and an agentic pipeline that translates…
100% match · <code><optimization><advanced><paper><accessible><cuda><gpu><hardware> - Variable-Rate Texture Compression: Real-Time Rendering with JPEG
Variable-Rate Texture Compression: Real-Time Rendering with JPEG presents companion OpenGL and CUDA source for a Computer Graphics Forum paper by Elias Kristmann, Michael Wimmer,…
100% match · <code><optimization><advanced><paper><accessible><cuda><last><link> - Ghost of Yōtei – tech deep dive
This PlayStation Blog post by Adrian Bentley, Director of Programming at Sucker Punch Productions, is a long-form tech deep dive into systems behind Ghost of Yotei on PlayStation…
100% match · <code><optimization><advanced><case><gpu><hardware><open><streams> - Inside NVIDIA GPUs: Anatomy of high performance matmul kernels
Inside NVIDIA GPUs: Anatomy of high performance matmul kernels is a long self-contained blog post by Aleksa Gordić that teaches how near-peak NVIDIA matrix-multiply kernels are…
100% match · <code><optimization><advanced><cuda><gpu><hardware><kernel><nvidia> - Towards GPU Parallelism Abstractions in Rust: A Case Study with Linear Pipelines
Towards GPU Parallelism Abstractions in Rust: A Case Study with Linear Pipelines presents a 2025 SBLP paper by Leonardo Gibrowski Fae and Dalvan Griebler (PUCRS) on high-level…
100% match · <code><advanced><paper><accessible><case><cuda><gpu><last> - Introducing Advanced Shader Delivery
Introducing Advanced Shader Delivery is a DirectX Developer Blog post by Wendy, Jesse, and Cassie announcing Microsoft's cloud-precompiled shader cache pipeline for PC games,…
100% match · <code><optimization><aimed><data><driver><gpu><hardware><launch> - Real-Time GPU Tree Generation
Real-Time GPU Tree Generation presents a High-Performance Graphics 2025 symposium paper by Bastian Kuth, Max Oberberger, Carsten Faber, Pirmin Pfeifer, Seyedmasih Tabaei, Dominik…
100% match · <code><optimization><advanced><paper><accessible><gpu><graphs><last> - GPU Programming and Language Design with Chris Lattner
GPU Programming and Language Design with Chris Lattner is a Software Unscripted podcast episode hosted by Richard Feldman with guest Chris Lattner on Mojo and GPU-oriented…
100% match · <code><advanced><cpu><cuda><gpu><hardware><kernel><open> - A Texture Lookup Approach to Bézier Curve Evaluation on the GPU
A Texture Lookup Approach to Bézier Curve Evaluation on the GPU is Muhammad Anas and Alan Wolfe's JCGT paper presenting a texture-based GPU method that evaluates Bézier curves…
98% match · <code><advanced><paper><accessible><gpu><last><link><performance> - Dual Contouring of Signed Distance Data
Dual Contouring of Signed Distance Data is a SIGGRAPH 2026 paper with a Columbia Graphics project page linking the PDF, source, and talk materials.
98% match · <code><advanced><paper><accessible><data><last><link><researchers>