TensorSharp hits 2x llama.cpp decode throughput on GLM-5.3-Flash
fuzhongkai · reddit · 2026-08-29
Benchmarks of GLM-5.3-Flash (unsloth GGUF UD-Q2KXL, 101GiB, dual GPUs) comparing TensorSharp vs llama.cpp (PR #27754, CUDA 12.8):
- Prefill is a wash: pp2048 2070 vs 2014 t/s; pp32768 1483 vs 1446 t/s.
- Decode doubles: tg64 73.5 vs 36.6 t/s (2.01×); 40.9 t/s after a 17.7K-token prompt, 28.2 t/s after 36K.
- Weights load in 17s from warm cache; CPU-MoE offload (first 10 layers' experts host-resident) decodes at 35–40 t/s.
The 2× comes from a shape-keyed LRU graph cache that replays CUDA-captured graphs instead of rebuilding per token — GLM's 34 KDA layers plus hyper-connections make per-token graphs deep in small ops, exactly where rebuild overhead hurts. Prefill is GEMM-bound, so both engines land within a few percent.
More from Infra
- Next-gen GraalVM benchmarks: 4x faster startup, 2x less memory, no build step needed — RealGeneKim · 2026-10-15
- How Microsoft optimized Windows for Nvidia's Arm-based RTX Spark AI laptops — tomwarren · 2026-10-15
- How Microsoft optimized Windows for Nvidia's Arm-based RTX Spark laptops — The Verge AI · 2026-10-15
- Global smartphone shipments plunge 22% in worst quarterly decline on record as AI demand fuels memory costs — Polymarket · 2026-10-15
- SK Hynix drops 11% again, now below 3x FY2027 earnings as AI memory selloff deepens — JOBhakdi · 2026-10-15
- ASML Sells Zero Lithography Machines into Europe as US Hosts 75%+ of AI Compute — NinaDSchick · 2026-10-15