What are the best models to run on AMD RX 7900 XT (20GB)?

Based on model rankings and 20 GB usable VRAM: Top Overall: Qwen 3.8 27B (INT4) ~39 tok/s. Top Coding: Qwen3.5-27B (INT4) ~39 tok/s. Optimal Daily Driver: Phi-4 Reasoning Plus (INT4) ~51 tok/s. Maximum Model Size: K2 Horizon MoVA 36B A4B (INT4) ~126 tok/s.

How are VRAM, throughput, and latency calculated?

Calculations are determined using apxml.com modeling, combining LLM architecture analysis with GPU hardware physics. VRAM and Context Scaling: Memory requirements are determined from model architecture, including parameter count, layer configuration, and attention structure. The KV cache footprint scales deterministically based on context length, batch size, and cache precision. Throughput (TPS): Token generation is calculated in the memory-bandwidth-bound decode phase, measuring the time required to stream model weights and KV cache relative to hardware memory bandwidth. Latency (TTFT): Time to first token is calculated from the compute operations required during prompt prefill relative to hardware tensor compute capacity.

How accurate are these estimates?

Calculations are calibrated against empirical ground-truth benchmark runs across diverse architectures and hardware. Because the underlying mechanics are benchmark-calibrated, estimates remain accurate across different configurations without needing to run physical tests for every scenario. VRAM Accuracy: Memory estimates are deterministic and closely match actual allocation. Minor differences typically stay within 3% to 5% due to CUDA runtime overhead or memory allocator fragmentation. Speed and Latency Accuracy: Throughput and TTFT typically vary within 5% to 10% depending on the specific inference runtime, kernel implementations, and system power limits.

Best Models to Run on AMD RX 7900 XT (20GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4	39 tok/s	~2.9s
#48	Agnes 3.0 Flash	33B	INT4	31 tok/s	~3.8s
#72	K2 Horizon MoVA 36B A4B	36B (4B active)	INT4	126 tok/s	~662 ms
#83	Qwen3.5-27B	27B	INT4	39 tok/s	~2.9s
#92	Sarvam-30B	32B (2.4B active)	INT4	164 tok/s	~428 ms
#94	Qwen3.6 35B A3B	35B (3B active)	INT4	143 tok/s	~540 ms
#96	Muse Glimmer 30B	30B	INT4	34 tok/s	~3.4s
#100	Phi-4 Reasoning Plus	14B	INT4 Q8	51 tok/s 33 tok/s	~1.5s ~1.6s
#101	MiMo V2 Flash	15B (309B active)	INT4	4 tok/s	~31.2s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4	148 tok/s	~516 ms
#109	Gemma 4 12B	11.95B	INT4 Q8	58 tok/s 38 tok/s	~1.3s ~1.4s
#113	Qwen3.5-9B	9B	INT4 Q8 FP16	96 tok/s 61 tok/s 30 tok/s	~978 ms ~984 ms ~1.2s
#114	Qwen3.5-35B-A3B	35B (3B active)	INT4	143 tok/s	~540 ms
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	115 tok/s 75 tok/s 42 tok/s	~766 ms ~771 ms ~781 ms
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	34 tok/s 20 tok/s 11 tok/s	~3.1s ~3.1s ~3.1s
#129	Gemma 4 31B	30.7B	INT4	33 tok/s	~3.5s
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	161 tok/s 113 tok/s 68 tok/s	~452 ms ~454 ms ~460 ms
#133	Qwen3-30B-A3B	30B (3B active)	INT4	153 tok/s	~490 ms
#136	GPT-OSS 20B	21B (3.6B active)	INT4	74 tok/s	~531 ms
#146	Granite 4.2 8B	8B	INT4 Q8 FP16	105 tok/s 67 tok/s 36 tok/s	~871 ms ~877 ms ~964 ms
#150	Gemma 4 E4B	8B	INT4 Q8 FP16	105 tok/s 67 tok/s 36 tok/s	~871 ms ~877 ms ~964 ms
#151	Nemotron 3.5 Lightning 30B A3B	30B (3B active)	INT4	153 tok/s	~490 ms
#153	Qwen3-32B	32B	INT4	32 tok/s	~3.7s
#158	GLM-4V	9B	INT4 Q8	72 tok/s 51 tok/s	~981 ms ~987 ms
#159	Gemma 3 27B	27B	INT4	39 tok/s	~2.9s
#160	Phi-4	14B	INT4 Q8	69 tok/s 42 tok/s	~1.5s ~1.5s
#162	DeepSeek-R1 32B	32B	INT4	33 tok/s	~3.7s
#164	Qwen2.5-32B	32B	INT4	32 tok/s	~3.7s
#168	Magistral Small	24B	INT4	32 tok/s	~2.8s
#169	OLMo 3 7B Think	7B	INT4 Q8 FP16	85 tok/s 62 tok/s 36 tok/s	~769 ms ~774 ms ~851 ms
#172	DeepSeek-R1 14B	14B	INT4 Q8	70 tok/s 42 tok/s	~1.5s ~1.5s
#173	Ministral 3 14B	14B	INT4 Q8	51 tok/s 33 tok/s	~1.5s ~1.6s
#177	Gemma 3 12B	12B	INT4 Q8	77 tok/s 48 tok/s	~1.3s ~1.3s
#181	Mistral-Small-2501	24B	INT4	44 tok/s	~2.5s
#182	Ministral 3 8B	8B	INT4 Q8 FP16	78 tok/s 56 tok/s 30 tok/s	~874 ms ~880 ms ~1.0s
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	219 tok/s 170 tok/s 113 tok/s	~243 ms ~244 ms ~247 ms
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	219 tok/s 170 tok/s 113 tok/s	~243 ms ~244 ms ~247 ms
#185	Qwen2.5-14B	14B	INT4 Q8	69 tok/s 42 tok/s	~1.5s ~1.5s
#186	Gemma 2 27B	27B	INT4	39 tok/s	~2.9s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	140 tok/s 95 tok/s 55 tok/s	~567 ms ~571 ms ~578 ms
#189	Qwen3-4B	4B	INT4 Q8 FP16	161 tok/s 113 tok/s 68 tok/s	~452 ms ~454 ms ~460 ms
#190	DeepSeek-R1 8B	8B	INT4 Q8 FP16	107 tok/s 68 tok/s 36 tok/s	~871 ms ~876 ms ~964 ms
#193	Ministral 3 3B	3B	INT4 Q8 FP16	134 tok/s 108 tok/s 73 tok/s	~349 ms ~351 ms ~355 ms
#195	Qwen3-14B	14B	INT4 Q8	69 tok/s 42 tok/s	~1.5s ~1.5s
#197	Gemma 2 9B	9B	INT4 Q8 FP16	96 tok/s 61 tok/s 30 tok/s	~978 ms ~984 ms ~1.2s
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	115 tok/s 75 tok/s 42 tok/s	~766 ms ~771 ms ~781 ms
#199	Qwen3-8B	8B	INT4 Q8 FP16	105 tok/s 67 tok/s 36 tok/s	~871 ms ~877 ms ~964 ms
#200	Gemma 3 4B	4B	INT4 Q8 FP16	161 tok/s 113 tok/s 68 tok/s	~452 ms ~454 ms ~460 ms
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	117 tok/s 76 tok/s 43 tok/s	~766 ms ~771 ms ~781 ms
#202	Llama 3.1 8B	8B	INT4 Q8 FP16	105 tok/s 67 tok/s 36 tok/s	~871 ms ~877 ms ~964 ms
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	233 tok/s 185 tok/s 126 tok/s	~209 ms ~210 ms ~212 ms
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	319 tok/s 303 tok/s 271 tok/s	~38 ms ~39 ms ~39 ms
#208	Llama 3 8B	8B	INT4 Q8 FP16	105 tok/s 67 tok/s 36 tok/s	~871 ms ~877 ms ~964 ms
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	307 tok/s 274 tok/s 221 tok/s	~78 ms ~78 ms ~79 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	165 tok/s 116 tok/s 71 tok/s	~431 ms ~433 ms ~439 ms
#212	Ministral-8B-2410	8B	INT4 Q8 FP16	105 tok/s 67 tok/s 36 tok/s	~871 ms ~877 ms ~964 ms
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	243 tok/s 196 tok/s 137 tok/s	~185 ms ~186 ms ~188 ms
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	185 tok/s 135 tok/s 85 tok/s	~347 ms ~349 ms ~353 ms
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	290 tok/s 251 tok/s 194 tok/s	~103 ms ~103 ms ~104 ms
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	165 tok/s 116 tok/s 71 tok/s	~431 ms ~433 ms ~439 ms
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	134 tok/s 117 tok/s 87 tok/s	~262 ms ~263 ms ~266 ms
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	111 tok/s 72 tok/s 41 tok/s	~797 ms ~802 ms ~813 ms
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	273 tok/s 232 tok/s 173 tok/s	~129 ms ~129 ms ~131 ms
#220	Gemma 2 2B	2B	INT4 Q8 FP16	219 tok/s 170 tok/s 113 tok/s	~243 ms ~244 ms ~247 ms
#221	Phi-3-medium	14B	INT4 Q8	69 tok/s 42 tok/s	~1.5s ~1.5s
#222	OLMo 3 7B Instruct	7B	INT4 Q8 FP16	85 tok/s 62 tok/s 36 tok/s	~769 ms ~774 ms ~851 ms
#224	Qwen2-7B	7B	INT4 Q8 FP16	115 tok/s 75 tok/s 42 tok/s	~766 ms ~771 ms ~781 ms
#225	Phi-3-small	7B	INT4 Q8 FP16	115 tok/s 75 tok/s 42 tok/s	~766 ms ~771 ms ~781 ms
#226	Gemma 3 1B	1B	INT4 Q8 FP16	273 tok/s 232 tok/s 173 tok/s	~129 ms ~129 ms ~131 ms
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	111 tok/s 72 tok/s 41 tok/s	~797 ms ~802 ms ~813 ms
#229	Gemma 1 7B	7B	INT4 Q8 FP16	85 tok/s 62 tok/s 36 tok/s	~769 ms ~774 ms ~851 ms
#230	Gemma 1 2B	2B	INT4 Q8 FP16	229 tok/s 175 tok/s 116 tok/s	~243 ms ~244 ms ~247 ms
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	94 tok/s 69 tok/s 43 tok/s	~664 ms ~668 ms ~677 ms
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	94 tok/s 69 tok/s 43 tok/s	~664 ms ~668 ms ~677 ms
#233	ChatGLM-6B	6B	INT4 Q8 FP16	94 tok/s 69 tok/s 43 tok/s	~664 ms ~668 ms ~677 ms
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	317 tok/s 287 tok/s 237 tok/s	~65 ms ~66 ms ~66 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	317 tok/s 287 tok/s 237 tok/s	~65 ms ~66 ms ~66 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	313 tok/s 296 tok/s 262 tok/s	~42 ms ~42 ms ~43 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	313 tok/s 296 tok/s 262 tok/s	~42 ms ~42 ms ~43 ms
-	Falcon-1B	1B	INT4 Q8 FP16	283 tok/s 238 tok/s 176 tok/s	~128 ms ~129 ms ~131 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	273 tok/s 232 tok/s 173 tok/s	~129 ms ~129 ms ~131 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	243 tok/s 196 tok/s 137 tok/s	~185 ms ~186 ms ~188 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	243 tok/s 196 tok/s 137 tok/s	~185 ms ~186 ms ~188 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	233 tok/s 185 tok/s 126 tok/s	~209 ms ~210 ms ~212 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	228 tok/s 180 tok/s 122 tok/s	~219 ms ~220 ms ~223 ms
-	Falcon-3B	3B	INT4 Q8 FP16	193 tok/s 139 tok/s 86 tok/s	~347 ms ~349 ms ~353 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	191 tok/s 167 tok/s 127 tok/s	~163 ms ~164 ms ~166 ms
-	Phi-1	1.3B	INT4 Q8 FP16	191 tok/s 167 tok/s 127 tok/s	~163 ms ~164 ms ~166 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	191 tok/s 167 tok/s 127 tok/s	~163 ms ~164 ms ~166 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	187 tok/s 136 tok/s 85 tok/s	~347 ms ~349 ms ~353 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	185 tok/s 135 tok/s 85 tok/s	~347 ms ~349 ms ~353 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	185 tok/s 135 tok/s 85 tok/s	~347 ms ~349 ms ~353 ms
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4	167 tok/s	~423 ms
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4	167 tok/s	~423 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	161 tok/s 113 tok/s 68 tok/s	~452 ms ~454 ms ~460 ms
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4	154 tok/s	~482 ms
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4	154 tok/s	~482 ms
-	Hy-MT2 30B A3B	30B (3B active)	INT4	153 tok/s	~490 ms
-	Phi-2	2.7B	INT4 Q8 FP16	140 tok/s 114 tok/s 78 tok/s	~318 ms ~319 ms ~323 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	134 tok/s 108 tok/s 73 tok/s	~349 ms ~351 ms ~355 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	134 tok/s 108 tok/s 73 tok/s	~349 ms ~351 ms ~355 ms
-	Falcon-7B	7B	INT4 Q8 FP16	119 tok/s 76 tok/s 43 tok/s	~766 ms ~771 ms ~781 ms
-	Falcon3-7B	7B	INT4 Q8 FP16	115 tok/s 75 tok/s 42 tok/s	~766 ms ~771 ms ~781 ms
-	Hy-MT2 7B	7B	INT4 Q8 FP16	115 tok/s 75 tok/s 42 tok/s	~766 ms ~771 ms ~781 ms
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	111 tok/s 72 tok/s 41 tok/s	~797 ms ~802 ms ~813 ms
-	LensVLM 9B	9B	INT4 Q8 FP16	96 tok/s 61 tok/s 30 tok/s	~978 ms ~984 ms ~1.2s
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8 FP16	96 tok/s 61 tok/s 30 tok/s	~978 ms ~984 ms ~1.2s
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	94 tok/s 69 tok/s 43 tok/s	~664 ms ~668 ms ~677 ms
-	Yi-6B	6B	INT4 Q8 FP16	94 tok/s 69 tok/s 43 tok/s	~664 ms ~668 ms ~677 ms
-	Falcon3-10B	10B	INT4 Q8	89 tok/s 56 tok/s	~1.1s ~1.1s
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4 Q8	89 tok/s 71 tok/s	~411 ms ~483 ms
-	Falcon2-11B	11B	INT4 Q8	86 tok/s 53 tok/s	~1.2s ~1.2s
-	Hunyuan Lite	7B	INT4 Q8 FP16	85 tok/s 62 tok/s 36 tok/s	~769 ms ~774 ms ~851 ms
-	OLMo 3 7B Base	7B	INT4 Q8 FP16	85 tok/s 62 tok/s 36 tok/s	~769 ms ~774 ms ~851 ms
-	SEA-LION-7B	7.1B	INT4 Q8 FP16	85 tok/s 61 tok/s 36 tok/s	~780 ms ~784 ms ~862 ms
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8 FP16	85 tok/s 61 tok/s 36 tok/s	~780 ms ~784 ms ~862 ms
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8 FP16	78 tok/s 56 tok/s 30 tok/s	~874 ms ~880 ms ~1.0s
-	Typhoon-2-8B	8B	INT4 Q8 FP16	78 tok/s 56 tok/s 30 tok/s	~874 ms ~880 ms ~1.0s
-	GLM-4-9B	9B	INT4 Q8	72 tok/s 51 tok/s	~981 ms ~987 ms
-	GLM-4-9B-Chat	9B	INT4 Q8	72 tok/s 51 tok/s	~981 ms ~987 ms
-	GLM-4-9B-Chat-1M	9B	INT4 Q8	72 tok/s 51 tok/s	~981 ms ~987 ms
-	Yi-9B	9B	INT4 Q8	72 tok/s 51 tok/s	~981 ms ~987 ms
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8	71 tok/s 50 tok/s	~1.0s ~1.0s
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8	71 tok/s 50 tok/s	~1.0s ~1.0s
-	GreenMind-14B-R1	14B	INT4 Q8	51 tok/s 33 tok/s	~1.5s ~1.6s

AMD RX 7900 XT (20GB)

VRAM Scaling by Context Length

Best Models to Run on AMD RX 7900 XT (20GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions