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

Qwen3.8 27B (VLM) on RTX 5090 32GB: Does It Fit, How Fast, What It Costs

Qwen3.8 27B (VLM) at FP8 (E4M3) needs 27.8 GB for its weights and 33.3 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×
33.3 GB needed / 64 GB available
Decode speed
106.9
tok/s per stream
Aggregate
2566
tok/s @ 8 concurrent
Cost / 1M tokens
$0.1083
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 GB33.3 GB0.0000.000——✗
2× RTX64 GB33.3 GB106.925664.37 s$0.1083✓
4× RTX128 GB33.3 GB213.8102642.20 s$0.0541✓
8× RTX256 GB33.3 GB427.7410561.12 s$0.0271✓
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?
4K861.07 GB✓ fits
8K522.15 GB✓ fits
33K158.59 GB✓ fits
66K817.18 GB✓ fits
131K434.36 GB✓ fits
Other GPUs that run Qwen3.8 27B (VLM)
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
AMD MI300X 192GB1×2232$0.3111
A100 40GB SXM41×652.8$0.3830
AMD MI325X 288GB1×2527$0.3847
A100 80GB SXM41×842.3$0.3957
Intel Gaudi 3 128GB1×1390$0.3997
H100 SXM5 80GB1×1411$0.4922
Frequently asked questions

Can Qwen3.8 27B (VLM) run on RTX 5090 32GB?

Yes, but it needs 2× RTX 5090 32GB. At FP8 (E4M3) the total footprint is 33.3 GB (27.8 GB of weights plus 5.5 GB of KV cache at 8K × 8 concurrent users), against 64 GB of available VRAM across 2 GPUs.

How fast is Qwen3.8 27B (VLM) on RTX 5090 32GB?

106.9 tokens/sec per stream and 2566 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 4.37 s, and inter-token latency is 9.4 ms.

What does Qwen3.8 27B (VLM) on RTX 5090 32GB cost per million tokens?

$0.1083 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 Qwen3.8 27B (VLM) need on RTX 5090 32GB?

27.8 GB for the weights at FP8 (E4M3), plus 5.5 GB of KV cache for 8 concurrent requests at 8K context — 33.3 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is RTX 5090 32GB the right choice for Qwen3.8 27B (VLM)?

It works. Compare $0.1083 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 4× RTX 3080 10GB at $0.0511, and the fastest is 1× Cerebras CS-3 (WSE-3) at 8423 tok/s.

Adjust the workload yourself

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

Open in calculator