Can I run it?

Can I run GLM 5.3 Flash on NVIDIA H100 SXM 80GB?

Partial offload

Partial offload on your NVIDIA H100 SXM 80GB: UD-IQ2_XXS at 4K context, 35 of 45 layers on the GPU and 20.2 GB in system RAM. Expect about 5.1–6.9 tok/s.

GLM 5.3 Flash details · NVIDIA H100 SXM 80GB page · How this estimate is made

Run commands

llama-server -hf unsloth/GLM-5.3-Flash-GGUF:UD-IQ2_XXS -c 4096 -ngl 35

ollama run glm-5.3-flash

lms get unsloth/GLM-5.3-Flash-GGUF
# Then load quant UD-IQ2_XXS

Assumptions

• Estimate from memory bandwidth, not a benchmark. Decode speed uses 55–75% of peak bandwidth.
• KV cache is FP16. Engine overhead assumes llama.cpp (0.3 GB + 1% of the weights). Ollama is closer to 0.9 GB + 3%, LM Studio closer to 1 GB + 3%.
• UD-IQ2_XXS size is the real GGUF file size from Hugging Face.
• CPU runs keep about 4 GB plus a 10% cushion for the OS.
• Speed is quoted at 4K context, not at the maximum context that still fits.
• This quant is IQ1 or IQ2. It can load, but quality is heavily degraded, so it is left off best-fit lists.
• About 35 of 45 layers stay on the GPU. 20.2 GB is read from system RAM.

Models that fit this machine more comfortably

• Qwen3 VL 235B A22B Instruct — Heavily degraded, UD-IQ1_M, 62–84 tok/s (estimate)
• MiniMax M2 — Heavily degraded, UD-IQ2_XXS, 57–78 tok/s (estimate)
• Step 3.7 Flash — Tight fit, UD-Q2_K_XL, 47–65 tok/s (estimate)

Updated 2026-10-24. Profile on this page is the device default plus 32 GB of DDR5-5600 system RAM.