implementation detail · filed under inference & serving
FP8 GEMM
Matrix multiplication performed using FP8 arithmetic.
Also called FP8 matrix multiplication (GEMM).
- source
- 1
- model
- 1
- labs adopt it
- 0
- strongest
- evaluated
How sources treat it
One count per evidence span, weakest treatment to strongest.
evaluated 1
Documented in
Evidence
1 span quoted from the sources, strongest treatment first.
FP8 matrix multiplication (GEMM) is one of the most compute-intensive parts of large model inference, and also one of the most difficult to optimize.
evaluatedsoftware implementationin MiniMax-M3MiniMax
Filed alongside
Other methods under inference & serving :: inference kernel.
Batch-invariant deterministic kernelsCUDA GraphCUDA Graph capture size reductionFlashAttention 3Kernel fusionMoE-side chunkingSynchronization-free static-shape MoE executionTensorRT-LLM multi-head attention backendAvoiding split-KDeepGEMM-based batch-invariant matrix multiplicationDistributed shared memory for cross-SM attention data exchangeDual-kernel batch-invariant attention decodingDynamic load balancingExp-free TopK kernelExpert-optimized Triton kernelsFA4 sheared-bias attention kernelFlashAttentionFlashAttention 4FlashKDAFused AttnRes merge and RMSNorm kernelFused latent down-projection and MoE-router GEMMFused mHC kernelsFused QSA kernelFused RoPE-attention-RoPE-cast kernel