NVIDIA32 GB VRAM1,792 GB/s$0.5/hr

Qwen 3 8B on RTX 5090 32GB: Does It Fit, How Fast, What It Costs

Qwen 3 8B at FP16 needs 16.0 GB for its weights and 25.7 GB in total at 8K context with 8 concurrent users. That means it fits on a single RTX 5090 32GB. Every figure is computed by tokcalc's open-source capacity model.

Fits?
Yes, 1 GPU
25.7 GB needed / 32 GB available
Decode speed
72.80
tok/s per stream
Aggregate
873.6
tok/s @ 8 concurrent
Cost / 1M tokens
$0.1590
1× RTX 5090 32GB
What changes as you add RTX 5090 32GBs
Tensor-parallel degree vs VRAM, throughput, and cost. Cost scales linearly with GPU count; decode throughput scales near-linearly too, so cost per token stays roughly flat.
GPUsTotal VRAMVRAM neededDecode tok/sAggregate tok/sTTFTCost / 1M tokFits?
1× RTX32 GB25.7 GB72.80873.62.54 s$0.1590✓
2× RTX64 GB25.7 GB123.829701.29 s$0.0935✓
4× RTX128 GB25.7 GB247.511881658.2 ms$0.0468✓
8× RTX256 GB25.7 GB495.047524344.1 ms$0.0234✓
Concurrency vs context length
On 1× RTX 5090 32GB at FP16. KV cache scales linearly with context, so this is the number of users your GPUs can actually serve.
ContextConcurrent usersKV per requestFits?
4K26618.5 MB✓ fits
8K131.21 GB✓ fits
33K34.83 GB✓ fits
66K19.66 GB✓ fits
131K019.33 GB✗ does not fit
Other GPUs that run Qwen 3 8B
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
AMD MI300X 192GB1×2584$0.2688
V100 SXM2 32GB1×438.8$0.3166
A100 40GB SXM41×755.6$0.3309
AMD MI325X 288GB1×2925$0.3324
A100 80GB SXM41×975.0$0.3419
Intel Gaudi 3 128GB1×1609$0.3453
Frequently asked questions

Can Qwen 3 8B run on RTX 5090 32GB?

Yes. Qwen 3 8B at FP16 needs 25.7 GB including the KV cache, and RTX 5090 32GB has 32 GB — so it fits on a single GPU.

How fast is Qwen 3 8B on RTX 5090 32GB?

72.80 tokens/sec per stream and 873.6 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 2.54 s, and inter-token latency is 13.7 ms.

What does Qwen 3 8B on RTX 5090 32GB cost per million tokens?

$0.1590 per 1M output tokens at 1× RTX 5090 32GB ($0.5000/hr). That is the marginal hardware cost — add idle capacity, ops, and egress for the real figure.

How much VRAM does Qwen 3 8B need on RTX 5090 32GB?

16.0 GB for the weights at FP16, plus 9.7 GB of KV cache for 8 concurrent requests at 8K context — 25.7 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is RTX 5090 32GB the right choice for Qwen 3 8B?

It works. Compare $0.1590 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 4× RTX 3080 10GB at $0.0441, and the fastest is 1× Cerebras CS-3 (WSE-3) at 9750 tok/s.

Adjust the workload yourself

Same formulas, your context length, concurrency, and quantisation.

Open in calculator