RTX 5090 (32GB) server with Ollama + Open WebUI, benchmarked vs RTX 4090
HOSTKEY swapped a 4090 for an RTX 5090 in an i9-14900K / 128GB server running Ollama and Open WebUI. Gemma3 27B ran at 48 tok/s on the 5090 vs 7 tok/s on the 4090, which spilled to CPU.
2-10 peopleCodingWriting assistantDocument search / chat with files$3k+Sensitivity: High (client/confidential)
Hardware
| Item | Approx. price (as of 2026-10-16) |
|---|---|
| On sale −6% MSI GeForce RTX 5060 Ti 16GB Ventus 3X OC (Newegg, related current offer) | $579.99 |
| Server: Intel Core i9-14900K, 128GB RAM, 2TB NVMe, NVIDIA GeForce RTX 5090 (32GB) | Not stated in source |
Source prices are as stated by the source. Retailer offers: prices checked 2026-10-16, may change. Some links may be affiliate links; we may earn a commission at no extra cost to you. This never changes what we list.
Software stack
- Ubuntu 22.04
- Docker + NVIDIA passthrough
- Ollama
- Open WebUI
Models used
- DeepSeek R1 14B / 32B
- Gemma3 27B
Good at
- DeepSeek R1 14B: 104.5 tok/s vs 65.8 on 4090
- 32GB VRAM fits 27B/32B models with large context
Weak at
- At test time, Open WebUI knowledge-base creation and web search failed, and ComfyUI needed newer PyTorch builds for Blackwell
- Whole system peaked at 830W
Rough total cost
Not stated in source.
Getting started
- Install Ubuntu 22.04, NVIDIA driver and CUDA 12.8+.
- Install Docker with NVIDIA GPU passthrough.
- Install Ollama on the host and Open WebUI as a container.
- Check Blackwell support in any CUDA-dependent tools before relying on them.
Source & credit
Jensen Huang, seriously!? Testing the NVIDIA RTX 5090 in AI workflows — Hugging Face blog (HOSTKEY)
Summarized from the source; quotes are the source author's words. Check the original before buying.