Can I run Qwen3.8-Flash-Next-GGUF on an NVIDIA H200?
Yes, just
Its UD-Q4_K_XL file (111 GB) fits in the 127 GB an NVIDIA H200 can use, with room for ≈ 256K tokens of context.
Full analysis of Qwen3.8-Flash-Next-GGUF
Everything that runs on NVIDIA H200
Get the UD-Q4_K_XL file on Hugging Face
llama-server -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XLFiles and context on NVIDIA H200
Each file of Qwen3.8-Flash-Next-GGUF (and of GGUF versions the site has analyzed), and how much context fits next to it, with the context cache in 16-bit or 8-bit.
| File | Size | On NVIDIA H200 | Context (16-bit) | Context (8-bit) |
|---|---|---|---|---|
| BF16 | 354 GB | Doesn't fit | — | — |
| Q8_0 | 188 GB | Doesn't fit | — | — |
| UD-Q6_K_XL | 169 GB | Doesn't fit | — | — |
| UD-Q5_K_XL | 158 GB | Doesn't fit | — | — |
| UD-Q4_K_XL | 111 GB | Tight | 256K | 256K |
| UD-IQ4_XS | 93.7 GB | Fits | 256K | 256K |
| UD-Q3_K_XL | 89.5 GB | Fits | 256K | 256K |
| UD-IQ3_XXS | 83.6 GB | Fits | 256K | 256K |
| UD-Q2_K_XL | 79.6 GB | Fits | 256K | 256K |
| UD-IQ2_XS | 75.8 GB | Fits | 256K | 256K |
| Qwen3.8-Flash-Next-UD-Q2_0 | 70.5 GB | Fits | 256K | 256K |
| UD-IQ1_M | 68.0 GB | Fits | 256K | 256K |
| UD-IQ1_S | 66.4 GB | Fits | 256K | 256K |