What GPU do I need to run Qwen/Qwen3-235B-A22B-Instruct-2507-FP8?
A 235B (MoE) language model for chat and instruction-following. 235.1B parameters, published in F8_E4M3. View on Hugging Face
Qwen3-235B-A22B-Instruct-2507-FP8 is published by Qwen on Hugging Face, with 176,542 downloads and 149 likes to date. It's a Qwen3MoeForCausalLM model built for text-generation, published natively in F8_E4M3.
What Qwen3-235B-A22B-Instruct-2507-FP8 is
Qwen3-235B-A22B-Instruct-2507-FP8 is a 235B-parameter mixture-of-experts language model published by Alibaba (Qwen) on Hugging Face. It is released under Apache 2.0.
License note: fully permissive, no gating and no usage restrictions. Facts in this section are sourced from Qwen3-235B-A22B-Instruct-2507-FP8's Hugging Face model card, not benchmarked by Aquanode.
What it's used for
- Chat assistants
- Instruction following
- Synthetic data generation
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) | 219.0 GB | 262.8 GB | RTX PRO 6000 | 3 | $4.13/hr |
| INT4 (quantized) | 109.5 GB | 131.4 GB | AMD MI300X | 1 | $2.63/hr |
| cheaper alt. | RTX A5000 | 6 | $1.06/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.
Cheapest way to run Qwen3-235B-A22B-Instruct-2507-FP8 at its published (F8_E4M3) precision: 3× RTX PRO 6000, at $1.38/hr per GPU ($4.13/hr total). Quantizing to FP8 or INT4 (rows above) can cost less, but requires a compatible quantized checkpoint to exist for this model.
Qwen3-235B-A22B-Instruct-2507-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.
Grouped-query attention: every layer caches keys and values for a few shared KV heads. Cached per token: 192,512 bytes at 16-bit.
| Context | KV cache, one sequence |
|---|---|
| 4K | 0.73 GB |
| 32K | 5.9 GB |
| 128K | 23.5 GB |
| 256K (model maximum) | 47.0 GB |
Computed from the layer, head and window counts in the model's own configuration (Qwen/Qwen3-235B-A22B-Instruct-2507-FP8), at 16 bits per cached value. Each concurrent sequence needs its own cache, on top of the weights.
Qwen3-235B-A22B-Instruct-2507-FP8: common questions
Can Qwen3-235B-A22B-Instruct-2507-FP8 run on a single GPU?
No. At FP8 (native) it needs 262.8 GB of VRAM, more than a single 32 GB desktop card holds. The cheapest capable card in the live feed is a 95.0 GB RTX PRO 6000, and it takes 3 of them.
Is Qwen3-235B-A22B-Instruct-2507-FP8 already quantized?
Yes. It is published in FP8, one byte per parameter, so the 262.8 GB figure is already a quantized footprint rather than a full-precision one. Only INT4 goes below it, at 131.4 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.
How many GPUs do I need to run Qwen3-235B-A22B-Instruct-2507-FP8?
3 at FP8 (native). It needs 262.8 GB of VRAM and the cheapest capable live offer is a 95.0 GB RTX PRO 6000, so 3 of them come to $4.13/hr in total.
Does quantizing Qwen3-235B-A22B-Instruct-2507-FP8 lower the GPU bill?
Yes. At FP8 (native) the cheapest live fit is 3 RTX PRO 6000 cards at $4.13/hr. At INT4 (quantized) it drops to one AMD MI300X at $2.63/hr, provided a quantized checkpoint exists for it.
How to run Qwen3-235B-A22B-Instruct-2507-FP8
Run Qwen3-235B-A22B-Instruct-2507-FP8 with vLLM
From Qwen/Qwen3-235B-A22B-Instruct-2507-FP8's own deployment docs.
vllm serve Qwen/Qwen3-235B-A22B-Instruct-2507-FP8 --tensor-parallel-size 4 --max-model-len 262144Source: https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507-FP8/raw/main/README.md
Deploy Qwen3-235B-A22B-Instruct-2507-FP8 on Aquanode
Aquanode has no one-click deploy template for Qwen3-235B-A22B-Instruct-2507-FP8; you install the inference engine yourself with the commands below. Aquanode sells GPU pods billed per second, not a hosted inference API.
- Launch a bare GPU pod sized to the requirement above (3× RTX PRO 6000 or larger).
- Open a terminal on the pod, or save one of the commands above as a startup script so it runs automatically the first time the pod boots.
- Run the command and connect to the resulting endpoint.
Launch a GPU podWeight-to-VRAM math, the fit rules, and how live prices are normalized: full methodology.
More Qwen3 models
- Qwen3-235B-A22B-Thinking-2507-FP8 (235.1B, F8_E4M3)
- Qwen3-235B-A22B-FP8 (235.1B, F8_E4M3)
- Qwen3-235B-A22B (235.1B, BF16)
- Qwen3-235B-A22B-Instruct-2507 (235.1B, BF16)
- Qwen3-235B-A22B-Thinking-2507 (235.1B, BF16)
Lineage
Based on Qwen3-235B-A22B-Instruct-2507.
Alternatives at this size
Other models for text-generation within about a third of Qwen3-235B-A22B-Instruct-2507-FP8's 235.1B parameters, from other model lines.
- MiniMax-M2.7 (228.7B, F8_E4M3)
- MiMo-V2.5 (310.8B, F8_E4M3)
- Step-3.5-Flash (199.4B, BF16)
- Hy3-preview (298.8B, BF16)
- DeepSeek-V2 (235.7B, BF16)
More on Qwen3-235B-A22B-Instruct-2507-FP8
Fits on a 288 GB GPU at FP8: every model that fits in 288 GB.
Fits on a 141 GB GPU at INT4: every model that fits in 141 GB.
Related reading: RTX PRO 6000 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.
