GLM-5.3 hits 21.4x speedup on RTX PRO 6000 via Kimi-Linear Decode kernel
teortaxesTex · x · 2026-08-22
KernelBench-Mega results show GLM-5.3 achieves 21.4x speedup over PyTorch baseline on RTX PRO 6000 using Kimi-Linear Decode, improving from GLM-5.2's 11.1x. The optimization uses a single CUDA loadinline launch, persistent 512-thread CTAs, and 30 grid barriers. It keeps Int4 dequantization inside GEMV and absorbs MLA to avoid materializing the fat KV cache. Benchmark comparisons show GLM-5.3 performing strongly against other leading models.
More from Infra
- Matryoshka Framework: Train Model Suites 36% Cheaper with Nested Architecture — TheTuringPost · 2026-08-22
- Stanford CS336 wraps up with deep dive into GPU programming and frontier inference — stanfordnlp · 2026-08-22
- How Pi handles context compaction for long coding sessions — bibryam · 2026-08-22
- Gwangyang Steel Works reveals the poverty of the data center energy debate — AndyMasley · 2026-08-22
- Benchmark Shows Unified Memory Significantly Boosts Local LLM VRAM Efficiency — Pablo_the_brave · 2026-08-22
- OpenAI Acquires Backend Startup Instant to Bolster AI App Infrastructure — testingcatalog · 2026-08-22