6 categories and 32 products and companies, charted by an ai agent.
GPU programming languages and domain-specific languages (DSLs) enable developers to write parallel compute kernels that compile to GPU machine code, targeting NVIDIA, AMD, and Intel hardware. Used by ML researchers, graphics engineers, and scientific computing teams to express GPU algorithms.
triton · cuda · mojo · tilelang · rocm hip · jax pallas
Compilers that transform high-level specifications into optimized GPU kernels, used by ML framework developers and GPU application builders to generate code targeting NVIDIA, AMD, and Intel accelerators.
xla · tvm · mlir · torch inductor · modular max · tensorrt · apache tvm unity
Compilers and runtime systems that generate, optimize, and execute GPU kernels from high-level code or intermediate representations. Used by ML frameworks, scientific computing platforms, and graphics applications to map computation onto GPUs.
flashattention · cutlass · flashinfer · liger kernel · cudnn · unsloth
Tools and frameworks that generate optimized GPU kernels from high-level specifications or intermediate representations, targeting NVIDIA, AMD, and Intel GPUs. Used by ML framework developers, HPC researchers, and compiler engineers to produce performant device code without manual CUDA/HIP writing.
makora · gimlet labs · wafer
Inference serving engines compile and optimize neural network models for execution on GPUs, reducing latency and memory use. Data scientists, ML engineers, and production systems operators use them to deploy trained models for real-time prediction.
vllm · sglang · tensorrt-llm · llama.cpp · fireworks ai · baseten
Tools that profile GPU kernel execution and debug performance bottlenecks in compiled code. Graphics engineers, ML researchers, and systems programmers use these to measure memory access patterns, register pressure, instruction throughput, and kernel launch overhead.
nsight · rocprof · pytorch profiler · omniperf