NVIDIA graphics card · current · data 2026-10-15

GeForce RTX 5090 for local AI.

Name: GeForce RTX 5090
Brand: NVIDIA

With 32 GB it runs Qwen3.8 27B (needs 22.7 GB). 75 models fit. Published speed: gpt-oss-20b at 318.4 tokens/sec.

Memory32 GB

vram

Bandwidth1792 GB/s

Higher means faster answers.

PriceFrom about $5,700

As of 2026-10-15. Check before buying.

Which models fit

By memory option.

Usable: 75% on Macs and shared-memory PCs, VRAM minus 1 GB on graphics cards (estimates).
Memory	Usable	Models that fit	Best three
32 GB	31 GB	75	Qwen3.8 27B (22.7 GB), K2 Horizon MoVA 36B-A4B (28.1 GB), Qwen3.5 27B (21.5 GB)

Speed

Published speeds.

Measured by others, with sources. Not our tests.

Model	Tokens/sec	Engine	Setup	Source
gpt-oss-20b	318.4	llama.cpp · MXFP4	RTX 5090 + Ryzen 7 9850X3D (chat-short workload)	community-measured · 2026-10-15

Questions

Short answers.

What is the best local LLM for the GeForce RTX 5090?: With 32 GB, Qwen3.8 27B (needs 22.7 GB, AA 33.7). 75 of 124 models fit.
How fast is the GeForce RTX 5090 for local AI?: Published: gpt-oss-20b at 318.4 tokens/sec (llama.cpp, RTX 5090 + Ryzen 7 9850X3D (chat-short workload)). These are other people's measurements, not ours.
How much does the GeForce RTX 5090 cost?: From about $5,700 (as of 2026-10-15). 2026 prices are inflated by the memory-chip shortage; check before buying.

Specs from NVIDIA – Compare GeForce graphics cards (memory, bandwidth, TGP) (2026-10-15). Prices are approximate and dated; 2026 prices are inflated by the memory-chip shortage.

AGENT VIEW · text/markdown · or send Accept: text/markdown

# GeForce RTX 5090 for local AI

URL: https://insidethemachine.ai/machines/rtx-5090
Type: NVIDIA graphics card. Chip: RTX 5090. Memory: vram, options 32 GB. Bandwidth: 1792 GB/s. Spec source: https://www.nvidia.com/en-us/geforce/graphics-cards/compare/ (2026-10-15).
Price: from about $5,700 (2026-10-15, approximate). Status: current.

## Which models fit

| memory GB | usable GB | models that fit | best 3 |
|---|---|---|---|
| 32 | 31 | 75 | Qwen3.8 27B (22.7 GB), K2 Horizon MoVA 36B-A4B (28.1 GB), Qwen3.5 27B (21.5 GB) |

## Published speeds (not our tests)

- gpt-oss-20b: 318.4 tok/s decode (llama.cpp, MXFP4, RTX 5090 + Ryzen 7 9850X3D (chat-short workload), community-measured) https://llm-speed.com/m/gpt-oss-20b (2026-10-15)

Notes: Street price is ~3.5× MSRP amid the 2026 DRAM shortage; rtx.live lists lowest in-stock at $5,700 (Micro Center).

## Questions

**What is the best local LLM for the GeForce RTX 5090?**
With 32 GB, Qwen3.8 27B (needs 22.7 GB, AA 33.7). 75 of 124 models fit.

**How fast is the GeForce RTX 5090 for local AI?**
Published: gpt-oss-20b at 318.4 tokens/sec (llama.cpp, RTX 5090 + Ryzen 7 9850X3D (chat-short workload)). These are other people's measurements, not ours.

**How much does the GeForce RTX 5090 cost?**
From about $5,700 (as of 2026-10-15). 2026 prices are inflated by the memory-chip shortage; check before buying.