Radeon RX 7900 XT
AMD Radeon RX 7900 XT: 20 GB of memory at 800 GB/s. Below: the open models it can run, estimated at 20 GB with an 8K context. 20 GB of VRAM for $899 at launch: enough for 27B-class models and 30B MoE models at 4-bit, using ROCm or Vulkan instead of CUDA.
Last updated: 2026-10-14
Specifications
What it can run
Best-quality quantization that fits with a 8K context, and the estimated generation speed. Estimates, not benchmarks. See the methodology.
| Model | Params | Best quant | Memory | tok/s (est.) |
|---|---|---|---|---|
| Gemma 4 31B IT | 30.7B | Q4_K_M | 19.8 GiB | 30 |
| Qwen3-30B-A3B (2507) | 30.5B (3.3B) | Q4_K_M | 19.3 GiB | 280 |
| Qwen3-Coder-30B-A3B-Instruct | 30.5B (3.3B) | Q4_K_M | 19.3 GiB | 280 |
| GLM-4.7-Flash (30B-A3B) | 30B (3B) | Q4_K_M | 18.7 GiB | 308 |
| Qwen3.5-27B | 27B | Q5_K_M | 19.8 GiB | 29 |
| Qwen3.6-27B | 27B | Q5_K_M | 19.8 GiB | 29 |
| Qwen3.8-27B | 27B | Q5_K_M | 19.8 GiB | 29 |
| Gemma 3 27B IT | 27B | Q4_K_M | 17.5 GiB | 34 |
| Gemma 4 26B A4B IT | 25.2B (3.8B) | Q5_K_M | 18.3 GiB | 207 |
| Mistral Small 3.2 24B Instruct (2506) | 24B | Q5_K_M | 18.4 GiB | 33 |
| Magistral Small 1.2 (2509) | 24B | Q5_K_M | 18.4 GiB | 33 |
| Devstral Small 2 24B Instruct (2512) | 24B | Q5_K_M | 18.4 GiB | 33 |
| gpt-oss-20b | 21B (3.6B) | Q6_K | 17.6 GiB | 189 |
| Phi-4 (14B) | 14B | Q8_0 | 16.6 GiB | 38 |
| Gemma 3 12B IT | 12B | Q8_0 | 13.8 GiB | 44 |
| Gemma 4 12B Unified IT | 11.95B | Q8_0 | 13.3 GiB | 44 |
| Qwen3.5-9B | 9B | FP16 | 18.4 GiB | 31 |
| Qwen3-8B | 8.2B | FP16 | 17.7 GiB | 34 |
| DeepSeek-R1-0528-Qwen3-8B | 8.19B | FP16 | 17.6 GiB | 34 |
| Llama 3.1 8B Instruct | 8B | FP16 | 17.1 GiB | 35 |
| Gemma 4 E4B IT | 8B | FP16 | 16.4 GiB | 35 |
| Llama 3.2 3B Instruct | 3.21B | FP16 | 7.7 GiB | 87 |
| Kimi K2.6 | 1000B (32B) | Doesn't fit | – | – |
| DeepSeek-R1-0528 (671B) | 671B (37B) | Doesn't fit | – | – |
| DeepSeek-V4-Flash | 284B (13B) | Doesn't fit | – | – |
| Qwen3.8-Flash-Next | 180B (6B) | Doesn't fit | – | – |
| Qwen3.5-122B-A10B | 122B (10B) | Doesn't fit | – | – |
| Mistral Small 4 119B (2603) | 119B (6.5B) | Doesn't fit | – | – |
| gpt-oss-120b | 117B (5.1B) | Doesn't fit | – | – |
| Llama 4 Scout 17B-16E Instruct | 109B (17B) | Doesn't fit | – | – |
| GLM-4.5-Air | 106B (12B) | Doesn't fit | – | – |
| Qwen3-Coder-Next (80B-A3B) | 80B (3B) | Doesn't fit | – | – |
| Llama 3.3 70B Instruct | 70B | Doesn't fit | – | – |
| Qwen3.5-35B-A3B | 35B (3B) | Doesn't fit | – | – |
| Qwen3.6-35B-A3B | 35B (3B) | Doesn't fit | – | – |
| Qwen3-32B | 32.8B | Doesn't fit | – | – |
| DeepSeek-R1-Distill-Qwen-32B | 32.76B | Doesn't fit | – | – |
Prices
No prices recorded for this device yet. All prices →
Frequently asked questions
Which LLMs can the Radeon RX 7900 XT run?
With 20 GB and an 8K context, the largest models that fit include: Gemma 4 31B IT, Qwen3-30B-A3B (2507), Qwen3-Coder-30B-A3B-Instruct, GLM-4.7-Flash (30B-A3B), Qwen3.5-27B. See the full table below.
How fast is the Radeon RX 7900 XT for local LLMs?
Generation speed is mostly set by memory bandwidth (800 GB/s). For Llama 3.1 8B Instruct at FP16, expect roughly 35 tokens/s; bigger models are slower.
Can the Radeon RX 7900 XT run 70B models?
Llama 3.3 70B Instruct does not fit on the Radeon RX 7900 XT (20 GB), even at Q4_K_M with an 8K context.
Sources
- https://www.amd.com/en/products/graphics/desktops/radeon/7000-series/amd-radeon-rx-7900xt.html
- https://www.amd.com/en/newsroom/press-releases/2022-11-3-amd-unveils-world-s-most-advanced-gaming-graphics-.html
- https://www.custompc.com/amd-radeon-rx-7900-xt-review
- https://rocm.docs.amd.com/en/latest/compatibility/compatibility-matrix.html