What GPU do I need to run Qwen/Qwen3.8-2.4T-A95B-FP8?
2446.2B parameters, published in F8_E4M3. View on Hugging Face
Set up Qwen3.8-2.4T-A95B-FP8Qwen3.8-2.4T-A95B-FP8 is published by Qwen on Hugging Face, with 23,352 downloads and 233 likes to date. It's a Qwen3_5MoeForCausalLM model built for text-generation, published natively in F8_E4M3.
VRAM required & cheapest live GPU fit
Required VRAM = weight size at each precision times a flat 1.2 overhead for activations and fragmentation. The KV-cache grows with context and is not in that factor; it is listed per model below where the architecture is published. Full formula and assumptions: methodology.
| Precision | Weight size | Required VRAM | Cheapest live fit | GPUs needed | Est. $/hr (full fit) |
|---|---|---|---|---|---|
| FP8 (native) | 2278.2 GB | 2733.8 GB | No capable live offer found | – | – |
| INT4 (quantized) | 1139.1 GB | 1366.9 GB | AMD MI300X | 8 | $21.03/hr |
A GPU is only matched to a row if its hardware supports that precision, and the primary recommendation is always a single-GPU fit when one exists.
INT4 requires a quantized checkpoint actually published for this model, check its Hugging Face page before relying on this row.
Qwen3.8-2.4T-A95B-FP8: KV cache by context length
The KV cache is the memory the attention layers hold for every token of context, on top of the weights. It is not part of the flat 1.2x overhead in the table above, grows with context length and with every concurrent request, and is why a long-context deployment needs more VRAM than the table shows.
Hybrid: only the full-attention layers cache per token; the other layers keep a fixed-size state. Cached per token: 94,208 bytes at 16-bit.
| Context | KV cache, one sequence |
|---|---|
| 4K | 0.36 GB |
| 32K | 2.9 GB |
| 128K | 11.5 GB |
| 256K (model maximum) | 23.0 GB |
Computed from the layer, head and window counts in the model's own configuration (Qwen/Qwen3.8-2.4T-A95B-FP8), at 16 bits per cached value. Each concurrent sequence needs its own cache, on top of the weights.
Qwen3.8-2.4T-A95B-FP8: common questions
Can Qwen3.8-2.4T-A95B-FP8 run on a single GPU?
Not on a desktop card. At FP8 (native) it needs 2733.8 GB of VRAM, more than a single 32 GB desktop card holds. No card currently listed on the marketplace both supports FP8 (native) and has enough VRAM for it, so how many it would take is not something this page can answer today.
Is Qwen3.8-2.4T-A95B-FP8 already quantized?
Yes. It is published in FP8, one byte per parameter, so the 2733.8 GB figure is already a quantized footprint rather than a full-precision one. Only INT4 goes below it, at 1366.9 GB. A GPU without FP8 tensor cores cannot run it as published, which is why cards here are matched on precision support and not on VRAM alone.
Weight-to-VRAM math, the fit rules, and how live prices are normalized: full methodology.
More Qwen3.8 models
- Qwen3.8-2.4T-A95B (2446.2B, BF16)
- Qwen3.8-Whittle-MoE-27B-A17.8B (26.9B, BF16)
- Qwen3.8-Flash-Next (180.0B, BF16)
- Qwen3.8-Flash-Next-FP8 (180.0B, F8_E4M3)
- Qwen3.8-Flash-Next-FP8 (180.0B, F8_E4M3)
Lineage
Based on Qwen3.8-2.4T-A95B.
Alternatives at this size
Other models for text-generation within about a third of Qwen3.8-2.4T-A95B-FP8's 2446.2B parameters, from other model lines.
- DeepSeek-V4-Pro-0813 (1650.5B, FP4 + FP8 Mixed)
More on Qwen3.8-2.4T-A95B-FP8
Related reading: H100 pricing and specs, How much VRAM you need for LLMs, Serving LLMs with vLLM, vLLM vs TensorRT-LLM vs SGLang, and Best GPU for LLM inference.
