RunsOnMy ModelsGPUs & MacsBy VRAMAPI pricesVRAM calculatorNew

Can Meta Llama 3.1 8B Instruct run on the H200?

Yes. At 8K context the best fit is F16, using about 16.6 GB of the 134 GB usable on the H200. Estimated speed ceiling: ~299 tokens/sec.

Memory by quant and context

Quant4K ctx8K ctx32K ctx128K ctxMax tok/s (est., 8K)
2-bit (Q2_K)3.9 GB Fits4.4 GB Fits7.4 GB Fits19.4 GB Fits1594
3-bit (Q3_K_M)4.8 GB Fits5.3 GB Fits8.3 GB Fits20.3 GB Fits1223
4-bit (Q4_K_M) •5.7 GB Fits6.2 GB Fits9.2 GB Fits21.2 GB Fits975
5-bit (Q5_K_M) •6.5 GB Fits7.0 GB Fits10.0 GB Fits22.0 GB Fits837
6-bit (Q6_K) •7.3 GB Fits7.8 GB Fits10.8 GB Fits22.8 GB Fits728
8-bit (Q8_0) •9.1 GB Fits9.6 GB Fits12.6 GB Fits24.6 GB Fits562
16-bit (F16/BF16) •16.1 GB Fits16.6 GB Fits19.6 GB Fits31.6 GB Fits299

• real GGUF file size from Saxo/Linkbricks-Horizon-AI-Korean-llama3.1-sft-rlhf-dpo-8B. Speed is a bandwidth ceiling (4800 GB/s), not a benchmark. Methodology.

More

Meta Llama 3.1 8B 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 (NousResearch/Meta-Llama-3.1-8B-Instruct), NVIDIA specs. Data as of 2026-11-17.