Can Qwen3.8 Flash Next run on the H100 PCIe?
Not fully in VRAM. Even at 2-bit with 8K context, Qwen3.8 Flash Next needs about 76.4 GB, more than the 76.0 GB usable on the H100 PCIe. It can still run with offload (slower): Q4_K_M with part of the model in system RAM, about 16–22 tokens/sec (estimate). Details and command.
Runs with offload (slower)
Runs with offload 4-bit (Q4_K_M) at 8K context: all attention layers plus the experts of 24 of 48 MoE layers on the H100 PCIe (about 74.2 GB), and the experts of the other 24 layers in system RAM (about 37.4 GB). Needs a PC with at least 64 GB of system RAM. Estimated speed: 16–22 tokens/sec (usable for chat), versus a bandwidth ceiling of ~49 tok/s if it fit fully in VRAM.
llama-server -hf bartowski/Qwen3.8-Flash-Next-GGUF:Q4_K_M -c 8192 -ngl 999 --n-cpu-moe 24 --no-mmproj
Estimate, not a benchmark. The command downloads the GGUF from Hugging Face. The speed range spans dual-channel DDR4-3200 to DDR5-6000 system RAM; your CPU, RAM speed and llama.cpp version move it. If llama.cpp runs out of GPU memory, raise --n-cpu-moe by 1-2; if VRAM is left over, lower it for more speed. Recent llama.cpp builds can also pick the split automatically. How we estimate offload.
+ 0.8 GB with image input. The figures on this page are for text only. With image input enabled, the vision projector (mmproj) file loads alongside the model. llama.cpp's -hf downloads and loads it automatically unless you add --no-mmproj. Calculator with image input on.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | 32K ctx | 128K ctx | Max tok/s (est., 8K) |
|---|---|---|---|---|---|
| 2-bit (Q2_K) • | 76.3 GB No | 76.4 GB No | 77.0 GB No | 79.2 GB No | — |
| 3-bit (Q3_K_M) • | 86.6 GB No | 86.7 GB No | 87.3 GB No | 89.5 GB No | — |
| 4-bit (Q4_K_M) • | 112 GB No | 112 GB No | 113 GB No | 115 GB No | — |
| 5-bit (Q5_K_M) • | 126 GB No | 126 GB No | 127 GB No | 129 GB No | — |
| 6-bit (Q6_K) • | 158 GB No | 158 GB No | 158 GB No | 160 GB No | — |
| 8-bit (Q8_0) • | 176 GB No | 176 GB No | 177 GB No | 179 GB No | — |
| 16-bit (F16/BF16) • | 331 GB No | 331 GB No | 331 GB No | 334 GB No | — |
• real GGUF file size from bartowski/Qwen3.8-Flash-Next-GGUF. Speed is a bandwidth ceiling (2000 GB/s), not a benchmark. Methodology.
Hardware that does fit it
RTX PRO 6000 BlackwellDGX SparkRyzen AI Max+ 395 128GBMac M3 Max 128GBMac M4 Max 128GBH200
What fits instead
- Cheapest listed GPU that runs Qwen3.8 Flash Next: DGX Spark (128 GB, launch MSRP $3,999) at Q3_K_M, fully in VRAM at 8K context.
- Largest popular model that fits the H100 PCIe: gpt-oss 120B (117B, MoE) at F16, about 61.9 GB at 8K context.
"Runs" and "fits" mean the whole model in GPU memory at 8K context (4-bit or better for the model pick). Prices are US launch MSRPs, not today's street prices. "Popular" means the 60 most-downloaded models we track.
More
Qwen3.8 Flash Next on every GPU and Mac · Every model that runs on the H100 PCIe · Other models that fit 64 GB VRAM · Try other settings in the calculator
Guides: Best GPU for local LLMs by budget · Can I run gpt-oss, Llama 70B or Qwen3? · Mac vs PC for local LLMs
Data: Hugging Face Hub (Qwen/Qwen3.8-Flash-Next), NVIDIA specs. Model data last checked 2026-11-18; refreshed hourly.