Reference¶
Lookup material — abbreviations, bibliography, and reading order suggestions.
Pages¶
abbreviations— quick lookup for acronyms and shorthand used throughoutbibliography— citations to papers, NVIDIA docs, kernel-library source, and notable blog posts
For a glossary of terms (full prose definitions), see overview/glossary.
Suggested reading orders¶
"I want to understand why my MoE model is slow on workstation Blackwell"¶
overview/architecture— the central thesisblackwell/sm100-vs-sm120— the hardware differenceinterconnect/nvlink-vs-pcie— the bandwidth differenceinterconnect/moe-parallelism— how MoE expects to use that bandwidthcompatibility/ep-to-tp-rewriting— the fix
"I want to understand the GPU itself, end-to-end"¶
fundamentals/gpu-execution-modelfundamentals/memory-hierarchyfundamentals/cuda-pipelinefundamentals/tensor-coresfundamentals/number-formatsblackwell/(entire section)
"I want to learn about the kernel libraries"¶
fundamentals/(entire section)kernels/(entire section, in order)compatibility/translating-tcgen05— when you need to port onecompatibility/smem-budget-management— the constraint side
"I just want a TL;DR"¶
overview/architecture. Read that page; it summarizes the whole wiki in 5 minutes.