RunsOnMy ModelsGPUs & MacsBy VRAMAPI pricesVRAM calculatorNew

Can Llama 3.3 70B Instruct run on the H200?

Yes. At 8K context the best fit is Q8_0, using about 73.0 GB of the 134 GB usable on the H200. Estimated speed ceiling: ~64 tokens/sec.

Memory by quant and context

Quant4K ctx8K ctx32K ctx128K ctxMax tok/s (est., 8K)
2-bit (Q2_K)26.5 GB Fits27.8 GB Fits35.3 GB Fits65.3 GB Fits181
3-bit (Q3_K_M)34.0 GB Fits35.3 GB Fits42.8 GB Fits72.8 GB Fits139
4-bit (Q4_K_M)42.1 GB Fits43.3 GB Fits50.8 GB Fits80.8 GB Fits111
5-bit (Q5_K_M)48.7 GB Fits50.0 GB Fits57.5 GB Fits87.5 GB Fits95
6-bit (Q6_K)55.8 GB Fits57.0 GB Fits64.5 GB Fits94.5 GB Fits83
8-bit (Q8_0)71.7 GB Fits73.0 GB Fits80.5 GB Fits110 GB Fits64
16-bit (F16/BF16)133 GB Tight135 GB No142 GB No172 GB No—

• real GGUF file size. Speed is a bandwidth ceiling (4800 GB/s), not a benchmark. Methodology.

More

Llama 3.3 70B Instruct on every GPU and Mac · Every model that runs on the H200 · Other models that fit 64 GB VRAM · Try other settings in the calculator

Data: Hugging Face Hub (meta-llama/Llama-3.3-70B-Instruct), NVIDIA specs. Data as of 2026-11-17.