Side-by-side simulation · recommended settings
One vs two RTX 5090s: Is a second card worth it for local LLMs?
A second card helps when a model exceeds one card or when requests can be divided. If the model already fits, compare splitting it across two against running it on one first.
Choose devices and a model
Expand comparison settings
Same 117output tokens. Only evidence-qualified acceleration is selected automatically; runtime formats are shown on each device.
Left
ReadyRTX 5090 32GB
Press Start above to generate the same answer on both devices.
Right
Ready2× RTX 5090 64GB (PCIe)
Press Start above to generate the same answer on both devices.
With these settings, total timeis similar on both devices.
Compare from input processing to completion of 117 token response.
Which one fits my needs?
Answer completion
Similar
Total answer time is similar with these settings.
Prompt processing
Similar
Prefill performance is similar at this input length.
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 0.1.41 · IQ3_S · MTP-6 · INT8 KV · SSD PLE / Strata 0.1.41 · IQ3_S · MTP-6 · INT8 KV · SSD PLE · GPU 1/2. 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
RTX 5090 32GB is the price of GPU separately, 2× RTX 5090 64GB (PCIe) is the price of 2 GPUs. For individual GPUs, you need to add the cost of the main body before comparing.
Before choosing hardware
| Measure | RTX 5090 32GB | 2× RTX 5090 64GB (PCIe) |
|---|---|---|
| Available fast memory | 30.5GB | 61GB |
| Memory bandwidth | 1792GB/s | 3584GB/s |
| Estimated total time | 1.2 s | 1.2 s |
| Estimated sustained LLM load | about 620W | about 1100W |
| Korean price range (KRW) | 899–929 × KRW 10,000 | 1,798–1,858 × KRW 10,000 |
Estimated ranges for one user. Multi-GPU and multi-device results depend on the runtime and interconnect.
Choose between splitting a model and dividing requests
Splitting one large model makes GPUs exchange data during computation. Keeping a small model on each card lets you route requests separately. The first serves capacity, the second concurrent work. After the single-answer comparison here, test the concurrency your actual server needs.
Check PCIe slots and physical spacing first
Two 5090s require checking actual motherboard lane wiring, card thickness and cooling clearance. The PSU and cables must accommodate both cards. Add any component replacements to the second GPU's cost and compare the complete build. Confirm that the case and system can house both before buying.
Check waiting time after moving to a larger file
If the goal of a second card is a larger model, comparing only the old model's speed is insufficient. Select the larger model and context you will actually use, then inspect first response and completion. Consider both whether it gives the required answer and whether the wait fits your workflow.