Side-by-side simulation · recommended settings

2× RTX 3090 vs RTX 5090 Local LLM Comparison

If putting a large model is your priority, two 3090s are right, and if single GPU speed and simple configuration are your priorities, one 5090 is right for you.

Choose devices and a model

Left deviceRight device

Model to run on both

Comparison mode

Expand comparison settings

Input length

Same 117output tokens. Only evidence-qualified acceleration is selected automatically; runtime formats are shown on each device.

Left

Ready

2× RTX 3090 48GB (NVLink)

First token1.3 to 3.6 seconds

Prefill1134.1 ~ 3122.7 tok/s

Decode67.1 ~ 184.9 tok/s

Press Start above to generate the same answer on both devices.

Strata 0.1.40 · IQ3_S · MTP-4 · pipeline-windows 2 · SSD PLE0.0s · 0/117

Right

Ready

RTX 5090 32GB

First token0.46 to 0.94 seconds

Prefill4375.9 ~ 9044.5 tok/s

Decode127.7 ~ 263.9 tok/s

Press Start above to generate the same answer on both devices.

Strata 0.1.41 · IQ3_S · MTP-6 · INT8 KV · SSD PLE0.0s · 0/117

With these settings, RTX 5090 32GB finishes the full answer About 2.4× faster.

Compare from input processing to completion of 117 token response.

Which one fits my needs?

Answer completion

RTX 5090 32GB

At the estimate midpoint, it finishes the same answer about 2.4× faster.

Prompt processing

RTX 5090 32GB

The estimated input-processing midpoint is higher at 6,710 tok/s.

Image and video generation

Check the media model

LLM speed does not tell you media-generation performance. Choose the image or video model and resolution separately.

Runtime

Settings for your hardware

Strata 0.1.40 · IQ3_S · MTP-4 · pipeline-windows 2 · SSD PLE / Strata 0.1.41 · IQ3_S · MTP-6 · INT8 KV · SSD PLE. When runtime paths differ, the speed gap also includes software and model-format differences.

Estimated power draw

RTX 5090 32GB

Estimated LLM power draw is lower at about 620 W. Noise and physical size need a separate check.

Local pricing

2× RTX 3090 48GB (NVLink) consists of 2 GPUs, while RTX 5090 32GB is a single GPU price. When comparing single GPU prices, the base cost of the GPU itself must be added.

Before choosing hardware

Measure	2× RTX 3090 48GB (NVLink)	RTX 5090 32GB
Available fast memory	45GB	30.5GB
Memory bandwidth	1872GB/s	1792GB/s
Estimated total time	2.9 s	1.2 s
Estimated sustained LLM load	about 780W	about 620W
Korean price range (KRW)	478–478 × KRW 10,000	899–929 × KRW 10,000

2× RTX 3090 48GB (NVLink) details (Korean)

Set up this device Image and video preview

RTX 5090 32GB details (Korean)

Set up this device Image and video preview

Estimated ranges for one user. Multi-GPU and multi-device results depend on the runtime and interconnect.