Skip to main content

​
GPU Docs

GPU Docs is a documentation assistant that helps you navigate CUDA, PTX, CuTe DSL, and CUTLASS concepts. Ask questions in natural language and get answers with citations to official documentation.

​
What is it?

GPU programming documentation is vast and scattered across multiple sources:
  • NVIDIA CUDA Programming Guide
  • PTX ISA Reference
  • CUTLASS documentation
  • CuTe DSL guides
  • Architecture whitepapers
GPU Docs brings this knowledge together in a conversational interface. Ask questions, get answers, and follow citations to dive deeper.

​
Features

Natural Language

Ask questions the way you’d ask a colleague—no special syntax needed.

Citations

Every answer includes links to source documentation so you can verify and learn more.

Multi-Turn Chat

Continue conversations with follow-up questions and context from previous messages.

File Context

Attach code files to get targeted help with your specific implementation.

​
What You Can Ask About

​
CUDA Programming

  • Memory management (global, shared, constant, texture memory)
  • Thread hierarchy (grids, blocks, warps, threads)
  • Synchronization primitives
  • CUDA runtime and driver APIs
  • Performance optimization techniques

​
PTX ISA

  • Instruction reference and encoding
  • Register usage and constraints
  • Memory operations and addressing modes
  • Modern features (tensor cores, MMA operations)

​
CuTe DSL

  • Layout concepts and operations
  • Tensor abstractions
  • Tiling strategies
  • Integration with CUTLASS

​
CUTLASS

  • Matrix multiplication primitives
  • Epilogue operations
  • Warp-level operations
  • Performance tuning

​
GPU Architecture

  • Streaming multiprocessor (SM) structure
  • Memory hierarchy
  • Warp scheduling
  • Architecture-specific features (Ampere, Hopper, Blackwell)

​
Requirements

RequirementDetails
Wafer AccountSign in with GitHub
Local InstallationNone required—works immediately
GPU Docs works out of the box with no local installation required. Just sign in and start asking questions.

​
Quick Start

  1. Select GPU Docs from the Wafer tool dropdown
  2. Type your question in the chat input
  3. Press Enter or click Send
  4. View the answer with inline citations

Using GPU Docs

Learn tips for getting the best answers →