What LLMs Can the Radeon RX 7900 XT 20GB Run?

The Radeon RX 7900 XT 20GB has 20 GB of VRAM and 800 GB/s of bandwidth. It runs 35 of 58 models fully at Q4_K_M with 8K context, up to Gemma 4 31B, and generates about 79 tokens/s on an 8B model.

AMD · RDNA 3 · Released 2022-12 · Updated October 14, 2026

Your GPU or Mac

Quantization

Context length

KV cache precision

35 of 58 models fit

Radeon RX 7900 XT 20GB (20.0 GB) at Q4_K_M with 8K context. Largest that fits fully: Gemma 4 31B.

Model	Needs	Status	Speed	Best quant that fits
Qwen 3 30B-A3B	18.8 GB	Fits	~97 tok/s	Q4_K_M
Qwen3-Coder 30B-A3B	18.8 GB	Fits	~97 tok/s	Q4_K_M
Gemma 3 27B	17.6 GB	Fits	~25 tok/s	Q4_K_M
Qwen 3.5 27B	16.9 GB	Fits	~26 tok/s	Q4_K_M
Qwen 3.6 27B	16.9 GB	Fits	~26 tok/s	Q4_K_M
Qwen 3.8 27B	16.9 GB	Fits	~26 tok/s	Q4_K_M
Mistral Small 3.1 24B	15.6 GB	Fits	~29 tok/s	Q5_K_M
Gemma 4 26B-A4B	15.6 GB	Fits	~112 tok/s	Q5_K_M
gpt-oss-20b	12.9 GB	Fits	~132 tok/s	MXFP4 (official)
Qwen 2.5 14B	10.7 GB	Fits	~44 tok/s	Q8_0
DeepSeek R1 Distill Qwen 14B	10.7 GB	Fits	~44 tok/s	Q8_0
Phi-4 14B	10.6 GB	Fits	~44 tok/s	Q8_0
Qwen 3 14B	10.5 GB	Fits	~45 tok/s	Q8_0
Ministral 3 14B	9.7 GB	Fits	~49 tok/s	Q8_0
Mistral Nemo 12B	9.0 GB	Fits	~53 tok/s	Q8_0
Gemma 3 12B	8.8 GB	Fits	~54 tok/s	Q8_0
Ministral 3 8B	6.7 GB	Fits	~75 tok/s	FP16 / BF16
Qwen 3 8B	6.6 GB	Fits	~77 tok/s	FP16 / BF16
Qwen 3.5 9B	6.4 GB	Fits	~76 tok/s	FP16 / BF16
Llama 3.1 8B	6.3 GB	Fits	~79 tok/s	FP16 / BF16
DeepSeek R1 Distill Llama 8B	6.3 GB	Fits	~79 tok/s	FP16 / BF16
Mistral 7B v0.3	5.7 GB	Fits	~87 tok/s	FP16 / BF16
Gemma 4 E4B	5.7 GB	Fits	~86 tok/s	FP16 / BF16
Qwen 2.5 7B	5.6 GB	Fits	~88 tok/s	FP16 / BF16
DeepSeek R1 Distill Qwen 7B	5.6 GB	Fits	~88 tok/s	FP16 / BF16
Qwen 3 4B	4.2 GB	Fits	~138 tok/s	FP16 / BF16
Phi-4 Mini	4.1 GB	Fits	~146 tok/s	FP16 / BF16
Gemma 4 E2B	4.0 GB	Fits	~134 tok/s	FP16 / BF16
Gemma 3 4B	3.7 GB	Fits	~149 tok/s	FP16 / BF16
Qwen 3.5 4B	3.7 GB	Fits	~153 tok/s	FP16 / BF16
Ministral 3 3B	3.5 GB	Fits	~167 tok/s	FP16 / BF16
Llama 3.2 3B	3.5 GB	Fits	~173 tok/s	FP16 / BF16
Llama 3.2 1B	1.7 GB	Fits	~460 tok/s	FP16 / BF16
Gemma 3 1B	1.6 GB	Fits	~587 tok/s	FP16 / BF16
Gemma 4 31B	19.8 GB	Tight fit	~22 tok/s	IQ4_XS
GLM-4.5-Air	62.1 GB	30% on GPU	~7.0 tok/s	–
Qwen3-Next 80B-A3B	46.3 GB	41% on GPU	~29 tok/s	–
Qwen 2.5 72B	44.5 GB	42% on GPU	~1.5 tok/s	–
Llama 3.3 70B	43.2 GB	43% on GPU	~1.6 tok/s	–
DeepSeek R1 Distill Llama 70B	43.2 GB	43% on GPU	~1.6 tok/s	–
Mixtral 8x7B	28.0 GB	67% on GPU	~12 tok/s	Q2_K
Qwen 2.5 32B	21.4 GB	88% on GPU	~9.7 tok/s	Q3_K_M
Qwen 2.5 Coder 32B	21.4 GB	88% on GPU	~9.7 tok/s	Q3_K_M
DeepSeek R1 Distill Qwen 32B	21.4 GB	88% on GPU	~9.7 tok/s	Q3_K_M
Qwen 3 32B	21.4 GB	88% on GPU	~9.7 tok/s	Q3_K_M
Qwen 3.5 35B-A3B	21.0 GB	90% on GPU	~78 tok/s	IQ4_XS
Qwen 3.6 35B-A3B	21.0 GB	90% on GPU	~78 tok/s	IQ4_XS
Kimi K2.6	566 GB	Too small	–	–
DeepSeek R1 671B	380 GB	Too small	–	–
Llama 3.1 405B	234 GB	Too small	–	–
Qwen 3.5 397B-A17B	226 GB	Too small	–	–
DeepSeek V4 Flash	161 GB	Too small	–	–
Qwen 3 235B-A22B	135 GB	Too small	–	–
MiniMax-M2	133 GB	Too small	–	–
Qwen 3.5 122B-A10B	70.2 GB	Too small	–	–
Mistral Small 4 119B	68.2 GB	Too small	–	–
gpt-oss-120b	67.2 GB	Too small	–	–
Llama 4 Scout	64.0 GB	Too small	–	–

Radeon RX 7900 XT 20GB for local AI

20 GB is an uncommon size that fits 24B models at Q5 and 27B models at Q4 with short contexts. Bandwidth is high, so token generation is fast in llama.cpp with ROCm or Vulkan.

Our picks for the Radeon RX 7900 XT 20GB

The model we would run first for each job, chosen from our recommendations by what fits fully with an 8K context. "Best quant" is the highest quality that fits; "max context" is at Q4_K_M or the native format.

Task	Model	Best quant	Max context	Speed
Best all-rounder	Qwen 3.6 27B	Q4_K_M	38K	~26 tok/s
Coding	Qwen 3.6 27B	Q4_K_M	38K	~26 tok/s
Fastest replies	gpt-oss-20b	MXFP4 (official)	128K	~125 tok/s
Long documents (64K+)	Gemma 4 26B-A4B	Q5_K_M	170K	~112 tok/s
Images in, text out	Gemma 4 26B-A4B	Q5_K_M	170K	~112 tok/s
Image generation	Qwen-Image	GGUF Q6_K	–	–

Spec	Value
Memory	20 GB
Memory bandwidth	800 GB/s
Architecture	RDNA 3
Released	2022-12
Llama 3.1 8B Q4_K_M speed	~79 tok/s
Largest model that fits	Gemma 4 31B

Image and video models on the Radeon RX 7900 XT 20GB

Peak VRAM with model swapping (ComfyUI default on smaller cards).

Model	FP8	FP8 fit	GGUF Q4	Q4 fit
Stable Diffusion 1.5	1.8 GB	Fits	1.7 GB	Fits
SDXL 1.0	4.1 GB	Fits	4.1 GB	Fits
Stable Diffusion 3.5 Medium	6.4 GB	Fits	6.4 GB	Fits
Stable Diffusion 3.5 Large	8.6 GB	Fits	6.4 GB	Fits
FLUX.1 [dev]	12.1 GB	Fits	7.3 GB	Fits
FLUX.1 [schnell]	12.1 GB	Fits	7.3 GB	Fits
Qwen-Image	20.4 GB	92% on GPU	12.3 GB	Fits
Wan 2.2 A14B (T2V / I2V)	18.8 GB	Tight fit	13.2 GB	Fits
Wan 2.2 TI2V 5B	8.0 GB	Fits	6.5 GB	Fits
HunyuanVideo	18.3 GB	Fits	13.1 GB	Fits
FLUX.1 Kontext [dev]	12.3 GB	Fits	7.5 GB	Fits
FLUX.2 [dev]	31.0 GB	61% on GPU	23.6 GB	80% on GPU
Z-Image Turbo	6.8 GB	Fits	4.9 GB	Fits
Qwen-Image-Edit 2509	20.5 GB	91% on GPU	12.4 GB	Fits
LTX-2	19.2 GB	Tight fit	12.6 GB	Fits

Upgrade path

The next step up is the Intel Arc Pro B60 24GB with 24 GB.

Radeon RX 7900 XT 20GB for LLMs: common questions

Can the Radeon RX 7900 XT 20GB run Llama 3.3 70B?

Only partly. It needs about 43.2 GB at Q4_K_M, so about 43% fits in VRAM and the rest runs from system RAM, which is slow.

Can the Radeon RX 7900 XT 20GB run Qwen 3 32B?

Only partly. It needs about 21.4 GB at Q4_K_M, so about 88% fits in VRAM and the rest runs from system RAM, which is slow. Q3_K_M fits fully.

Can the Radeon RX 7900 XT 20GB run Qwen 3 14B?

Yes. Qwen 3 14B needs about 10.5 GB at Q4_K_M with 8K context, and the best quality that fits is Q8_0.

How fast is the Radeon RX 7900 XT 20GB for LLMs?

About 79 tokens per second for an 8B model at Q4_K_M, based on its 800 GB/s memory bandwidth. Larger models scale down roughly in proportion to their size.