AMD · 24 GB GDDR6 · 960 GB/s

Yes. Radeon RX 7900 XTX fits Gemma 4 31B at Q4_K_M, the largest catalog model that clears the fit line at 8K context.

384-bit GDDR6 at 2500 MHz. 2500×8×384/8 = 960 GB/s. Speed uses raw GDDR bandwidth. Infinity Cache is far smaller than model weights, so it is not added. Source.

Recommended quant, 8,192 tokens.
ModelQuantWeightsTotalVerdictSpeedOllama
Gemma 4 31BQ4_K_M17.1 GiB19.9 GiBFits27–42 tok/sollama run gemma4:31b
Qwen3 30B-A3BQ4_K_M17.3 GiB19.5 GiBFits186–378 tok/sollama run qwen3:30b-a3b
Qwen3.8 27BUD-Q4_K_M15.3 GiB17.3 GiBFits31–47 tok/s—
Gemma 3 27B InstructQ4_K_M15.4 GiB17.8 GiBFits30–46 tok/sollama run gemma3:27b
Gemma 4 26B-A4BUD-Q4_K_M15.8 GiB17.5 GiBFits200–340 tok/sollama run gemma4:26b-a4b
Mistral Small 3.2 24B InstructQ4_K_M13.3 GiB15.9 GiBFits34–54 tok/s—
Devstral Small 2 24B InstructQ4_K_M13.3 GiB15.9 GiBFits34–54 tok/s—
gpt-oss-20bMXFP411.3 GiB12.6 GiBFits201–317 tok/sollama run gpt-oss:20b
Qwen2.5 14B InstructQ4_K_M8.37 GiB10.9 GiBFits50–85 tok/sollama run qwen2.5:14b
DeepSeek-R1 Distill Qwen 14BQ4_K_M8.37 GiB10.9 GiBFits50–85 tok/sollama run deepseek-r1:14b
Qwen3 14BQ4_K_M8.38 GiB10.6 GiBFits51–85 tok/sollama run qwen3:14b
Phi-4Q4_K_M8.43 GiB11.0 GiBFits49–85 tok/sollama run phi4:14b
Mistral Nemo 12B InstructQ4_K_M6.96 GiB9.15 GiBFits60–103 tok/sollama run mistral-nemo:12b
Gemma 4 12BQ4_K_M6.63 GiB7.95 GiBFits70–108 tok/sollama run gemma4:12b
Gemma 3 12B InstructQ4_K_M6.80 GiB8.52 GiBFits65–105 tok/sollama run gemma3:12b
Qwen3.5 9BQ4_K_M5.29 GiB6.39 GiBFits88–135 tok/sollama run qwen3.5:9b
Qwen3 8BQ4_K_M4.68 GiB6.62 GiBFits85–153 tok/sollama run qwen3:8b
Llama 3.1 8B InstructQ4_K_M4.58 GiB6.39 GiBFits88–156 tok/sollama run llama3.1:8b
Qwen2.5 7B InstructQ4_K_M4.36 GiB5.56 GiBFits102–164 tok/sollama run qwen2.5:7b
Gemma 4 E4BQ4_K_M4.64 GiB5.66 GiBFits101–154 tok/sollama run gemma4:e4b
Gemma 4 E2BQ4_K_M2.89 GiB3.69 GiBFits163–247 tok/sollama run gemma4:e2b
Gemma 3 4B InstructQ4_K_M2.32 GiB3.24 GiBFits190–308 tok/sollama run gemma3:4b
Phi-4-mini InstructQ4_K_M2.32 GiB4.01 GiBFits148–308 tok/sollama run phi4-mini
Llama 3.2 3B InstructQ4_K_M1.88 GiB3.42 GiBFits178–380 tok/sollama run llama3.2:3b
SmolLM3 3BQ4_K_M1.78 GiB2.99 GiBFits210–401 tok/s—
Llama 3.2 1B InstructQ4_K_M0.75 GiB1.58 GiBFits491–951 tok/sollama run llama3.2:1b
Qwen3.6 35B-A3BUD-Q4_K_M20.6 GiB22.4 GiBTight222–358 tok/s—
Qwen2.5 32B InstructQ4_K_M18.5 GiB22.0 GiBTight24–39 tok/sollama run qwen2.5:32b
DeepSeek-R1 Distill Qwen 32BQ4_K_M18.5 GiB22.0 GiBTight24–39 tok/sollama run deepseek-r1:32b
Qwen3 32BQ4_K_M18.4 GiB22.0 GiBTight24–39 tok/sollama run qwen3:32b
Llama 3.3 70B InstructQ4_K_M39.6 GiB44.7 GiBOffload1 tok/sollama run llama3.3:70b
gpt-oss-120bMXFP459.0 GiB62.8 GiBNo—ollama run gpt-oss:120b
Qwen2.5 72B InstructQ4_K_M44.2 GiB49.5 GiBNo—ollama run qwen2.5:72b

Exact answers

Qwen3.6 35B-A3Btight · UD-Q4_K_MQwen3 32Btight · Q4_K_MGemma 4 31Bfits · Q4_K_MQwen3 30B-A3Bfits · Q4_K_MQwen3.8 27Bfits · UD-Q4_K_MMistral Small 3.2 24B Instructfits · Q4_K_M

FAQ

What does the verdict for Radeon RX 7900 XTX mean?

Memory is the GGUF file, plus KV cache at 8,192 tokens, plus a 0.5 GiB engine allowance, then 5% headroom. Fits means that total is at most 85% of fast memory. Tight means it still fits. Offload means it needs system RAM. No means it does not fit in VRAM plus system RAM after an 8 GB Windows reserve. The formula is on the methodology page.

What memory did you count as usable?

Fast memory is the full 24 GB. System RAM is assumed to be 32 GB DDR5-5600, with 8 GB kept for Windows, and you can change that on the homepage calculator.

Why is speed a range?

Decode speed is memory bandwidth divided by the bytes read per token. The high end is 80% of that roofline on the weights alone. The low end is 55% once the KV cache at the working context is included. It is not a benchmark.

How should I run Qwen3.6 35B-A3B?

The table uses each model's recommended quant, usually Q4_K_M or the vendor's published quant. Qwen3.6 35B-A3B has no verified Ollama tag, so no command is shown.