NVIDIA GeForce RTX 5090 for local LLMs
Written by Jakub Rusinowski · Last updated
With 32 GB at 1,792 GB/s, the RTX 5090 runs 113 catalogued models at Q4_K_M with 8K context. The largest that fits is Gemma 3 27B Instruct (~25.4 GB), and the top pick is K2 Horizon MoVA 36B-A4B at 133–276 tok/s.
The fastest consumer GPU for local AI. 32 GB VRAM fits 32B-class models such as Qwen 3 32B at Q4_K_M with room for long context; Llama 3.3 70B does not fit on one card, even at Q3. About 78% more memory bandwidth than the RTX 4090.
Models that run on the RTX 5090
Q4_K_M, 8K context, 32 GB usable. Ranked by quality and speed.
| Model | Memory · marker = 32 GB | VRAM | Speed |
|---|---|---|---|
| K2 Horizon MoVA 36B-A4B K2 Horizon | 25 GB | 133–276 tok/s | |
| Qwen 3.6 35B-A3B Qwen 3.6 | 22.1 GB | 183–381 tok/s | |
| Nex-N2.5 mini Nex-N2.5 | 21.9 GB | 139–289 tok/s | |
| Nex-N2 mini Nex-N2 | 21.9 GB | 139–289 tok/s | |
| Qwen 3.5 35B-A3B Qwen 3.5 | 21.9 GB | 139–289 tok/s | |
| Laguna XS 2.1 33B-A3B Poolside Laguna XS 2.1 | 20.7 GB | 140–291 tok/s | |
| Qwen 3 32B Qwen 3 | 22.8 GB | 39–81 tok/s | |
| Aya Expanse 32B Aya Expanse | 21.6 GB | 41–85 tok/s | |
| EXAONE 3.5 32B EXAONE 3.5 | 22.3 GB | 40–83 tok/s | |
| Cogito v1 32B Cogito v1 | 22.3 GB | 40–83 tok/s | |
| Granite 4.0 Small-H 32B-A9B IBM Granite 4.0 | 20.1 GB | 93–192 tok/s | |
| Olmo 3 32B Olmo 3 | 20.1 GB | 40–84 tok/s |
Buy it or rent the same memory
Buy the card, or rent a GPU with the same memory by the hour to try models first.
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
Speed vs other GPUs
Llama 3.1 8B, Q4_K_M. Estimated ranges. How this is calculated
Specifications
Specs last updated 2026-09-29.
- Memory
- 32 GB
- Memory bandwidth
- 1,792 GB/s
- Architecture
- Blackwell GB202
- Series
- RTX 50-series
- Board power
- 575 W
- Release year
- 2025
- Launch price
- $1,999
- Usable for models
- 32 GB
Similar GPUs
Frequently asked questions
Can the NVIDIA GeForce RTX 5090 run local LLMs?
Yes. With 32 GB (32 GB usable by a model) it runs 113 of the catalogued models at Q4_K_M with 8K context; the largest is Gemma 3 27B Instruct, needing about 25.4 GB.
How fast is the NVIDIA GeForce RTX 5090 for AI inference?
It is estimated to run Llama 3.1 8B at 113–234 tok/s at Q4_K_M. Llama 3.3 70B does not fit: it needs about 44 GB against 32 GB usable. These are modelled estimates from memory bandwidth, not measurements; the methodology page shows the formula.
What LLMs can I run on 32 GB?
Among the best that fit: K2 Horizon MoVA 36B-A4B, Qwen 3.6 35B-A3B, Nex-N2.5 mini, Nex-N2 mini, Qwen 3.5 35B-A3B. The quickest start is Ollama: ollama run llama4:scout.
Keep going
Can I run it on the RTX 5090?
More for this card
Tools