AMD Radeon RX 7900 XT for local LLMs
Written by Jakub Rusinowski · Last updated
With 20 GB at 800 GB/s, the RX 7900 XT runs 103 catalogued models at Q4_K_M with 8K context. The largest that fits is Nemotron-Cascade 2 30B-A3B (~19.9 GB), and the top pick is Qwen 3.6 27B at 17–34 tok/s.
20 GB VRAM is a unique sweet spot — fits all 14B models with a larger context than 16GB cards. Great for AMD users who want more VRAM than the 7800 XT.
Models that run on the RX 7900 XT
Q4_K_M, 8K context, 20 GB usable. Ranked by quality and speed.
| Model | Memory · marker = 20 GB | VRAM | Speed |
|---|---|---|---|
| Qwen 3.6 27B Qwen 3.6 | 17.6 GB | 17–34 tok/s | |
| Qwen 3.5 27B Qwen 3.5 | 17.1 GB | 17–35 tok/s | |
| 1-bit Bonsai 27B Bonsai 27B | 17.1 GB | 17–35 tok/s | |
| Ternary Bonsai 27B Bonsai 27B | 17.1 GB | 17–35 tok/s | |
| Ternary Bonsai 2 27B Ternary Bonsai 2 | 17.6 GB | 18–38 tok/s | |
| Trinity Mini Trinity | 16.9 GB | 90–186 tok/s | |
| Gemma 4 26B-A4B Gemma 4 | 16.5 GB | 57–118 tok/s | |
| Devstral Small 2505 24B Devstral | 16.6 GB | 19–40 tok/s | |
| Devstral Small 2 24B Devstral | 16.6 GB | 19–40 tok/s | |
| Magistral Small 24B Magistral Small | 15.3 GB | 19–39 tok/s | |
| Mistral Small 3 (24B) Mistral Family | 17.2 GB | 19–39 tok/s | |
| LFM2.5-24B-A2B LFM2.5 | 15.5 GB | 97–202 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.
or compare on Vast.ai from $0.35/hr (typical low · varies)
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 2026-07-12.
- Memory
- 20 GB
- Memory bandwidth
- 800 GB/s
- Architecture
- RDNA 3 Navi 31
- Series
- RX 7000-series
- Board power
- 315 W
- Release year
- 2022
- Launch price
- $899
- Usable for models
- 20 GB
Similar GPUs
Frequently asked questions
Can the AMD Radeon RX 7900 XT run local LLMs?
Yes. With 20 GB (20 GB usable by a model) it runs 103 of the catalogued models at Q4_K_M with 8K context; the largest is Nemotron-Cascade 2 30B-A3B, needing about 19.9 GB.
How fast is the AMD Radeon RX 7900 XT for AI inference?
It is estimated to run Llama 3.1 8B at 47–98 tok/s at Q4_K_M. Llama 3.3 70B does not fit: it needs about 44 GB against 20 GB usable. These are modelled estimates from memory bandwidth, not measurements; the methodology page shows the formula.
What LLMs can I run on 20 GB?
Among the best that fit: Qwen 3.6 27B, Qwen 3.5 27B, 1-bit Bonsai 27B, Ternary Bonsai 27B, Ternary Bonsai 2 27B. The quickest start is Ollama: ollama run qwen3:14b.
Keep going
Can I run it on the RX 7900 XT?
More for this card
Tools