florianmattana

florianmattana / sass-king

Public

Reverse engineering NVIDIA SASS instruction dictionary, kernel audits and pattern recognition across GPU architectures.

83
2
100% credibility
Found Apr 20, 2026 at 83 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Cuda
AI Summary

An educational roadmap and studies reverse-engineering NVIDIA GPU instructions to correlate code patterns with real-world performance measurements.

How It Works

1
🔍 Discover SASS King

You come across this GitHub page while curious about the hidden workings of NVIDIA graphics cards.

2
📖 Read the Starting Guide

You dive into the introduction and first article to understand the basics of studying GPU speed secrets.

3
🎯 Explore Early Lessons

You follow simple examples that show how small tweaks make graphics tasks run much faster.

4
📋 Check the Learning Path

You review the completed studies and future plans for deeper dives into common computing patterns.

5
🖥️ See Covered Graphics Cards

You learn which popular NVIDIA cards are featured and why they power everyday AI and gaming.

🏆 Unlock GPU Insights

You now grasp the inner magic of graphics performance, empowering smarter projects and experiments.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 83 to 83 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is sass-king?

Sass-king reverse engineers NVIDIA SASS instructions across GPU architectures like SM120 on RTX 5070, pairing cuobjdump disassembly and gpuasm.com visualizations with Nsight Compute profiling to link code to real performance metrics. Built in CUDA, it delivers kernel studies on topics from FMA fusion and scoreboards to tensor cores and warp reductions, solving the black-box pain of optimizing GPU kernels without vendor docs. Think github reverse engineering meets king of sass: empirical breakdowns that reveal instruction latencies, stalls, and patterns.

Why is it gaining traction?

Unlike scattered SASS dumps or opaque profiler outputs, it correlates disassembly with measured throughput and stalls, making github reverse commits and kernel audits actionable for tuning. The roadmap hooks devs with planned audits of flash-attn, cutlass, and llama.cpp kernels, plus cross-arch ports to H100 and B200—rare systematic reverse engineering ki in a sea of high-level CUDA advice.

Who should use this?

CUDA kernel writers debugging slow global memory loads or register spills on consumer GPUs like RTX 4090. ML engineers auditing transformer libs for SASS inefficiencies in inference. Reverse engineering app enthusiasts dissecting production binaries from xformers or tinygrad.

Verdict

Watchlist for serious GPU hackers—83 stars and 1.0% credibility score reflect early-stage maturity with solid docs but no shipped code yet. Dive into the Part 1 blog now; contribute kernels to accelerate Phase 2.

(178 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.