Qwen3 8B GPU requirements: 4.1 GB at 4-bit, 16.4 GB at BF16
Qwen3 8B needs 16.4 GB of GPU memory for its weights at BF16, 8.19 GB at FP8 and 4.1 GB at 4-bit, plus 4.83 GB of KV cache for a 32K context. The cheapest node that runs it here is 1× RTX 3080 at 4-bit, for $0.03 an hour on spot.
Weights plus a 32K 16-bit KV cache, with 10% of the node's memory kept free. Billed per minute.
- Parameters
- 8.2BDense
- Weights, 4-bit
- 4.1 GBBF16: 16.4 GB
- KV cache, 32K
- 4.83 GBOne sequence, 16-bit cache
- Context window
- 128K40K native, extended with YaRN
How much VRAM Qwen3 8B needs, format by format.
Weights are parameters × bytes per parameter. Each context column adds the KV cache of one sequence of that length.
| Format | Weights | + 8Kcontext | + 32Kcontext | + 128Kcontext | Cheapest node, 32K |
|---|---|---|---|---|---|
| BF16Published weights, fine-tuning | 16.4 GB | 17.6 GB | 21.2 GB | 35.7 GB | 1× A30 · $0.05/h |
| FP8Hopper, Ada, Blackwell, MI300 | 8.19 GB | 9.4 GB | 13.0 GB | 27.5 GB | 1× L4 · $0.05/h |
| 4-bitAWQ, GPTQ, GGUF Q4 | 4.1 GB | 5.3 GB | 8.93 GB | 23.4 GB | 1× RTX 3080 · $0.03/h |
The serving engine adds its own buffers, and an 8-bit KV cache halves the context columns. We keep 10% of a node's memory free when we say a model fits. Speed is a separate question: it follows memory bandwidth and, for a mixture of experts, the active parameters.
GPUs that run Qwen3 8B, cheapest first.
For each format, the smallest node of each GPU model that holds the weights and a 32K context, sorted by spot price. One row per GPU model; larger nodes of the same GPU work too.
| Format | GPU | GPU memory | Spot | On-demand | 730 h spot | Action |
|---|---|---|---|---|---|---|
| 4-bit | 1× RTX 3080Ampere · 8.93 GB needed | 10 GB | $0.03 | $0.06 | $21.90 | Deploy |
| 4-bit | 1× RTX A4000Ampere · 8.93 GB needed | 16 GB | $0.03 | $0.05 | $21.90 | Deploy |
| 4-bit | 1× RTX 4070Ada Lovelace · 8.93 GB needed | 12 GB | $0.04 | $0.06 | $29.20 | Deploy |
| 4-bit | 1× RTX 3080 TiAmpere · 8.93 GB needed | 12 GB | $0.04 | $0.07 | $29.20 | Deploy |
| FP8 | 1× L4Ada Lovelace · 13.0 GB needed | 24 GB | $0.05 | $0.08 | $36.50 | Deploy |
| FP8 | 1× RTX 5080Blackwell · 13.0 GB needed | 16 GB | $0.06 | $0.09 | $43.80 | Deploy |
| FP8 | 1× RTX 5070 TiBlackwell · 13.0 GB needed | 16 GB | $0.06 | $0.09 | $43.80 | Deploy |
| FP8 | 1× RTX 4080 SUPERAda Lovelace · 13.0 GB needed | 16 GB | $0.06 | $0.09 | $43.80 | Deploy |
| BF16 | 1× A30Ampere · 21.2 GB needed | 24 GB | $0.05 | $0.08 | $36.50 | Deploy |
| BF16 | 1× L4Ada Lovelace · 21.2 GB needed | 24 GB | $0.05 | $0.08 | $36.50 | Deploy |
| BF16 | 1× RTX 3090Ampere · 21.2 GB needed | 24 GB | $0.05 | $0.08 | $36.50 | Deploy |
| BF16 | 1× RTX A5000Ampere · 21.2 GB needed | 24 GB | $0.06 | $0.09 | $43.80 | Deploy |
Spot capacity can be reclaimed with a 2-minute notice; for a live endpoint, use on-demand. How spot works · serving a language model.
What a long context costs in memory.
Grouped-query attention, 8 KV heads of 128 across 36 layers.
| Context | KV cache, 1 sequence | 8 sequences |
|---|---|---|
| 8K tokens | 1.21 GB | 9.66 GB |
| 32K tokens | 4.83 GB | 38.7 GB |
| 128K tokens | 19.3 GB | 155 GB |
Serving several users
KV cacheEvery concurrent request keeps its own cache, so memory for context grows with users × context length. A server batching 8 requests of 32K needs eight times the single-sequence figure — 38.7 GB here.
- Recommended node1× RTX 3080
- Weights at 4-bit4.1 GB
- Context that fits, 1 sequence32K
Paged attention (vLLM, SGLang) allocates the cache in blocks as requests grow, so a node holds more short requests than the worst case suggests.
Qwen3 8B, as published.
- Developer
- Alibaba QwenSmall dense Qwen3, common fine-tuning base
- Parameters
- 8.2B
- Architecture
- Dense model, 36 layersGrouped-query attention, 8 KV heads of 128 across 36 layers
- Context window
- 131,072 tokens40,960 native, the rest with YaRN scaling
- Published format
- BF16, 16.4 GB
- Licence
- Apache 2.0
- Released
- 29 April 2025
- Source
- Qwen/Qwen3-8Bconfig.json read 26 October 2026
Serve it with vLLM
template: vLLMOn 1× RTX 3080, from the instance terminal:
$ vllm serve Qwen/Qwen3-8B \ --tensor-parallel-size 1 \ --max-model-len 32768
Very recent architectures need a recent vLLM; the model card gives the minimum version and any extra flags.
Qwen3 8B on a GPU — the questions that come up
How much VRAM does Qwen3 8B need?
For the weights alone: 16.4 GB at BF16, 8.19 GB at FP8 and 4.1 GB at 4-bit. Add 4.83 GB of KV cache for one 32K-token context, or 19.3 GB at the full 128K window, and keep about 10% of the memory free for activations.
Can Qwen3 8B run on a single GPU?
One RTX 4090 (24 GB): yes, at BF16. One RTX 5090 (32 GB): yes, at BF16. One H100 SXM (80 GB): yes, at BF16. One H200 SXM (141 GB): yes, at BF16. One B200 SXM (180 GB): yes, at BF16. Those answers assume a 32K context and the widest format that still fits.
What is the cheapest way to run Qwen3 8B?
Here, 1× RTX 3080 at 4-bit: $0.03 per hour on spot and $0.06 on-demand, billed per minute — $21.90 for a 730-hour month on spot. That node has 10 GB of GPU memory for the 8.93 GB this setup needs.
How long a context fits on the recommended node?
With the weights at 4-bit on 1× RTX 3080, the remaining memory holds a 16-bit KV cache for about 33,264 tokens of context for one sequence (32K). An 8-bit KV cache doubles that; serving several users at once divides it between them.
Which format should I run it in?
BF16 keeps the published weights untouched and is the format for fine-tuning. FP8 halves the memory and runs in hardware on Hopper, Ada, Blackwell and MI300-class GPUs. 4-bit quantisation (AWQ, GPTQ, GGUF) quarters it, at a quality cost that is larger on small models than on big ones; check it against your own evaluation before relying on it.
How do I deploy it?
Pick the node in the table above and launch it with the vLLM template; the model downloads from Hugging Face (Qwen/Qwen3-8B) on first start. Spread the weights over the node with --tensor-parallel-size equal to the GPU count, and set --max-model-len to the context you need. New architectures need a recent vLLM: check the model card for the minimum version.
Sizing another model? GPU requirements for 64 open models · serving a language model · fine-tuning.
Qwen3 8B on 1× RTX 3080 for $0.03 an hour.
Launch the node with the vLLM template, point it at Qwen/Qwen3-8B, stop it when you are done. Prepaid, billed by the minute.
Billed per minute from the moment the instance is reachable. Minimum credit $30, no subscription.