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

Qwen 3 30B-A3B (MoE) on RTX 5090 32GB: Does It Fit, How Fast, What It Costs

Qwen 3 30B-A3B (MoE) at FP8 (E4M3) needs 30.0 GB for its weights and 36.4 GB in total at 8K context with 8 concurrent users. That means it needs 2× RTX 5090 32GB. Every figure is computed by tokcalc's open-source capacity model.

Fits?
Yes, 2×
36.4 GB needed / 64 GB available
Decode speed
990.1
tok/s per stream
Aggregate
23762
tok/s @ 8 concurrent
Cost / 1M tokens
$0.0117
2× RTX 5090 32GB

RTX 5090 32GB has no NVLink, so 2-way tensor parallelism runs over PCIe. tokcalc applies a flat 0.85× multi-GPU efficiency factor for this, which is realistic for NVLink-equipped datacenter cards but optimistic for PCIe scale-out — treat the throughput and cost figures for this configuration as an upper bound, and measure before committing.

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 GB36.4 GB0.0000.000——✗
2× RTX64 GB36.4 GB990.123762499.1 ms$0.0117✓
4× RTX128 GB36.4 GB198095048264.6 ms$0.005845✓
8× RTX256 GB36.4 GB3960380191147.3 ms$0.002923✓
Concurrency vs context length
On 2× RTX 5090 32GB at FP8 (E4M3). KV cache scales linearly with context, so this is the number of users your GPUs can actually serve.
ContextConcurrent usersKV per requestFits?
4K84412.3 MB✓ fits
8K42824.6 MB✓ fits
33K103.22 GB✓ fits
66K56.44 GB✓ fits
131K212.88 GB✓ fits
Other GPUs that run Qwen 3 30B-A3B (MoE)
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
AMD MI300X 192GB1×20670$0.0336
A100 40GB SXM41×6045$0.0414
AMD MI325X 288GB1×23400$0.0415
A100 80GB SXM41×7800$0.0427
Intel Gaudi 3 128GB1×12870$0.0432
H100 SXM5 80GB1×13065$0.0532
Frequently asked questions

Can Qwen 3 30B-A3B (MoE) run on RTX 5090 32GB?

Yes, but it needs 2× RTX 5090 32GB. At FP8 (E4M3) the total footprint is 36.4 GB (30.0 GB of weights plus 6.4 GB of KV cache at 8K × 8 concurrent users), against 64 GB of available VRAM across 2 GPUs.

How fast is Qwen 3 30B-A3B (MoE) on RTX 5090 32GB?

990.1 tokens/sec per stream and 23762 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 499.1 ms, and inter-token latency is 1.0 ms.

What does Qwen 3 30B-A3B (MoE) on RTX 5090 32GB cost per million tokens?

$0.0117 per 1M output tokens at 2× 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 30B-A3B (MoE) need on RTX 5090 32GB?

30.0 GB for the weights at FP8 (E4M3), plus 6.4 GB of KV cache for 8 concurrent requests at 8K context — 36.4 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is RTX 5090 32GB the right choice for Qwen 3 30B-A3B (MoE)?

It works. Compare $0.0117 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 4× RTX 3080 10GB at $0.005513, and the fastest is 1× Cerebras CS-3 (WSE-3) at 78000 tok/s.

Adjust the workload yourself

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

Open in calculator