What are the best models to run on AMD RX 7900 XTX (24GB)?

Based on model rankings and 24 GB usable VRAM: Top Overall: Qwen 3.8 27B (INT4) ~46 tok/s. Top Coding: Qwen3.5-27B (INT4) ~46 tok/s. Optimal Daily Driver: Phi-4 Reasoning Plus (INT4) ~60 tok/s. Maximum Model Size: Falcon-40B (INT4) ~32 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 XTX (24GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4	46 tok/s	~2.4s
#48	Agnes 3.0 Flash	33B	INT4	39 tok/s	~2.9s
#72	K2 Horizon MoVA 36B A4B	36B (4B active)	INT4	149 tok/s	~514 ms
#83	Qwen3.5-27B	27B	INT4	46 tok/s	~2.4s
#92	Sarvam-30B	32B (2.4B active)	INT4	190 tok/s	~332 ms
#94	Qwen3.6 35B A3B	35B (3B active)	INT4	167 tok/s	~420 ms
#96	Muse Glimmer 30B	30B	INT4	42 tok/s	~2.7s
#100	Phi-4 Reasoning Plus	14B	INT4 Q8	60 tok/s 41 tok/s	~1.3s ~1.3s
#101	MiMo V2 Flash	15B (309B active)	INT4 Q8	5 tok/s 2 tok/s	~26.2s ~28.6s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4	165 tok/s	~434 ms
#109	Gemma 4 12B	11.95B	INT4 Q8	67 tok/s 47 tok/s	~1.1s ~1.1s
#113	Qwen3.5-9B	9B	INT4 Q8 FP16	110 tok/s 71 tok/s 38 tok/s	~821 ms ~826 ms ~908 ms
#114	Qwen3.5-35B-A3B	35B (3B active)	INT4	167 tok/s	~420 ms
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	130 tok/s 86 tok/s 50 tok/s	~644 ms ~648 ms ~656 ms
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	40 tok/s 24 tok/s 13 tok/s	~2.6s ~2.6s ~2.6s
#129	Gemma 4 31B	30.7B	INT4	42 tok/s	~2.7s
#131	Nemotron 3.5 Lightning	30B (3B active)	INT4	78 tok/s	~418 ms
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	177 tok/s 127 tok/s 78 tok/s	~380 ms ~382 ms ~387 ms
#133	Qwen3-30B-A3B	30B (3B active)	INT4	177 tok/s	~379 ms
#136	GPT-OSS 20B	21B (3.6B active)	INT4	90 tok/s	~411 ms
#146	Granite 4.2 8B	8B	INT4 Q8 FP16	119 tok/s 78 tok/s 44 tok/s	~731 ms ~736 ms ~746 ms
#150	Gemma 4 E4B	8B	INT4 Q8 FP16	119 tok/s 78 tok/s 44 tok/s	~731 ms ~736 ms ~746 ms
#151	Nemotron 3.5 Lightning 30B A3B	30B (3B active)	INT4	177 tok/s	~379 ms
#153	Qwen3-32B	32B	INT4	40 tok/s	~2.8s
#156	GLM-4	32B	INT4	28 tok/s	~3.3s
#158	GLM-4V	9B	INT4 Q8 FP16	83 tok/s 59 tok/s 34 tok/s	~824 ms ~829 ms ~911 ms
#159	Gemma 3 27B	27B	INT4	46 tok/s	~2.4s
#160	Phi-4	14B	INT4 Q8	79 tok/s 49 tok/s	~1.3s ~1.3s
#162	DeepSeek-R1 32B	32B	INT4	41 tok/s	~2.8s
#164	Qwen2.5-32B	32B	INT4	40 tok/s	~2.8s
#168	Magistral Small	24B	INT4	40 tok/s	~2.1s
#169	OLMo 3 7B Think	7B	INT4 Q8 FP16	98 tok/s 72 tok/s 44 tok/s	~646 ms ~650 ms ~658 ms
#170	OLMo 3.1 32B Instruct	32B	INT4	28 tok/s	~3.3s
#172	DeepSeek-R1 14B	14B	INT4 Q8	81 tok/s 50 tok/s	~1.3s ~1.3s
#173	Ministral 3 14B	14B	INT4 Q8	60 tok/s 41 tok/s	~1.3s ~1.3s
#177	Gemma 3 12B	12B	INT4 Q8 FP16	89 tok/s 56 tok/s 28 tok/s	~1.1s ~1.1s ~1.3s
#180	OLMo 3 32B Base	32B	INT4	28 tok/s	~3.3s
#181	Mistral-Small-2501	24B	INT4	51 tok/s	~2.1s
#182	Ministral 3 8B	8B	INT4 Q8 FP16	90 tok/s 65 tok/s 38 tok/s	~734 ms ~738 ms ~812 ms
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	235 tok/s 187 tok/s 128 tok/s	~205 ms ~206 ms ~208 ms
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	235 tok/s 187 tok/s 128 tok/s	~205 ms ~206 ms ~208 ms
#185	Qwen2.5-14B	14B	INT4 Q8	79 tok/s 49 tok/s	~1.3s ~1.3s
#186	Gemma 2 27B	27B	INT4	46 tok/s	~2.4s
#187	OLMo 3.1 32B Think	32B	INT4	28 tok/s	~3.3s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	156 tok/s 108 tok/s 65 tok/s	~477 ms ~479 ms ~486 ms
#189	Qwen3-4B	4B	INT4 Q8 FP16	177 tok/s 127 tok/s 78 tok/s	~380 ms ~382 ms ~387 ms
#190	DeepSeek-R1 8B	8B	INT4 Q8 FP16	121 tok/s 79 tok/s 44 tok/s	~731 ms ~736 ms ~746 ms
#193	Ministral 3 3B	3B	INT4 Q8 FP16	150 tok/s 122 tok/s 84 tok/s	~294 ms ~295 ms ~299 ms
#195	Qwen3-14B	14B	INT4 Q8	79 tok/s 49 tok/s	~1.3s ~1.3s
#197	Gemma 2 9B	9B	INT4 Q8 FP16	110 tok/s 71 tok/s 38 tok/s	~821 ms ~826 ms ~908 ms
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	130 tok/s 86 tok/s 50 tok/s	~644 ms ~648 ms ~656 ms
#199	Qwen3-8B	8B	INT4 Q8 FP16	119 tok/s 78 tok/s 44 tok/s	~731 ms ~736 ms ~746 ms
#200	Gemma 3 4B	4B	INT4 Q8 FP16	177 tok/s 127 tok/s 78 tok/s	~380 ms ~382 ms ~387 ms
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	132 tok/s 87 tok/s 50 tok/s	~644 ms ~647 ms ~656 ms
#202	Llama 3.1 8B	8B	INT4 Q8 FP16	119 tok/s 78 tok/s 44 tok/s	~731 ms ~736 ms ~746 ms
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	248 tok/s 201 tok/s 142 tok/s	~176 ms ~177 ms ~179 ms
#205	Command R	35B	INT4	26 tok/s	~3.6s
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	327 tok/s 313 tok/s 283 tok/s	~34 ms ~34 ms ~34 ms
#208	Llama 3 8B	8B	INT4 Q8 FP16	119 tok/s 78 tok/s 44 tok/s	~731 ms ~736 ms ~746 ms
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	316 tok/s 286 tok/s 237 tok/s	~66 ms ~67 ms ~67 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	182 tok/s 132 tok/s 82 tok/s	~362 ms ~364 ms ~369 ms
#212	Ministral-8B-2410	8B	INT4 Q8 FP16	119 tok/s 78 tok/s 44 tok/s	~731 ms ~736 ms ~746 ms
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	258 tok/s 213 tok/s 153 tok/s	~157 ms ~157 ms ~159 ms
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	202 tok/s 151 tok/s 97 tok/s	~292 ms ~294 ms ~297 ms
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	300 tok/s 265 tok/s 211 tok/s	~87 ms ~88 ms ~89 ms
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	182 tok/s 132 tok/s 82 tok/s	~362 ms ~364 ms ~369 ms
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	150 tok/s 132 tok/s 100 tok/s	~220 ms ~221 ms ~224 ms
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	126 tok/s 83 tok/s 48 tok/s	~670 ms ~674 ms ~683 ms
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	286 tok/s 247 tok/s 190 tok/s	~109 ms ~110 ms ~111 ms
#220	Gemma 2 2B	2B	INT4 Q8 FP16	235 tok/s 187 tok/s 128 tok/s	~205 ms ~206 ms ~208 ms
#221	Phi-3-medium	14B	INT4 Q8	79 tok/s 49 tok/s	~1.3s ~1.3s
#222	OLMo 3 7B Instruct	7B	INT4 Q8 FP16	98 tok/s 72 tok/s 44 tok/s	~646 ms ~650 ms ~658 ms
#223	Yi-34B	34B	INT4	27 tok/s	~3.5s
#224	Qwen2-7B	7B	INT4 Q8 FP16	130 tok/s 86 tok/s 50 tok/s	~644 ms ~648 ms ~656 ms
#225	Phi-3-small	7B	INT4 Q8 FP16	130 tok/s 86 tok/s 50 tok/s	~644 ms ~648 ms ~656 ms
#226	Gemma 3 1B	1B	INT4 Q8 FP16	286 tok/s 247 tok/s 190 tok/s	~109 ms ~110 ms ~111 ms
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	126 tok/s 83 tok/s 48 tok/s	~670 ms ~674 ms ~683 ms
#229	Gemma 1 7B	7B	INT4 Q8 FP16	98 tok/s 72 tok/s 44 tok/s	~646 ms ~650 ms ~658 ms
#230	Gemma 1 2B	2B	INT4 Q8 FP16	245 tok/s 192 tok/s 131 tok/s	~205 ms ~206 ms ~208 ms
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	107 tok/s 80 tok/s 50 tok/s	~558 ms ~561 ms ~569 ms
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	107 tok/s 80 tok/s 50 tok/s	~558 ms ~561 ms ~569 ms
#233	ChatGLM-6B	6B	INT4 Q8 FP16	107 tok/s 80 tok/s 50 tok/s	~558 ms ~561 ms ~569 ms
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	324 tok/s 298 tok/s 252 tok/s	~56 ms ~56 ms ~57 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	324 tok/s 298 tok/s 252 tok/s	~56 ms ~56 ms ~57 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	321 tok/s 306 tok/s 275 tok/s	~37 ms ~37 ms ~37 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	321 tok/s 306 tok/s 275 tok/s	~37 ms ~37 ms ~37 ms
-	Falcon-1B	1B	INT4 Q8 FP16	294 tok/s 253 tok/s 193 tok/s	~109 ms ~110 ms ~111 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	286 tok/s 247 tok/s 190 tok/s	~109 ms ~110 ms ~111 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	258 tok/s 213 tok/s 153 tok/s	~157 ms ~157 ms ~159 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	258 tok/s 213 tok/s 153 tok/s	~157 ms ~157 ms ~159 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	248 tok/s 201 tok/s 142 tok/s	~176 ms ~177 ms ~179 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	244 tok/s 196 tok/s 137 tok/s	~185 ms ~186 ms ~188 ms
-	Falcon-3B	3B	INT4 Q8 FP16	210 tok/s 156 tok/s 99 tok/s	~292 ms ~293 ms ~297 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	208 tok/s 184 tok/s 142 tok/s	~138 ms ~139 ms ~140 ms
-	Phi-1	1.3B	INT4 Q8 FP16	208 tok/s 184 tok/s 142 tok/s	~138 ms ~139 ms ~140 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	208 tok/s 184 tok/s 142 tok/s	~138 ms ~139 ms ~140 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	204 tok/s 152 tok/s 98 tok/s	~292 ms ~294 ms ~297 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	202 tok/s 151 tok/s 97 tok/s	~292 ms ~294 ms ~297 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	202 tok/s 151 tok/s 97 tok/s	~292 ms ~294 ms ~297 ms
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4 Q8	184 tok/s 127 tok/s	~355 ms ~418 ms
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4 Q8	184 tok/s 127 tok/s	~355 ms ~418 ms
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4	179 tok/s	~373 ms
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4	179 tok/s	~373 ms
-	Hy-MT2 30B A3B	30B (3B active)	INT4	177 tok/s	~379 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	177 tok/s 127 tok/s 78 tok/s	~380 ms ~382 ms ~387 ms
-	Phi-2	2.7B	INT4 Q8 FP16	156 tok/s 129 tok/s 90 tok/s	~267 ms ~269 ms ~272 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	150 tok/s 122 tok/s 84 tok/s	~294 ms ~295 ms ~299 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	150 tok/s 122 tok/s 84 tok/s	~294 ms ~295 ms ~299 ms
-	Falcon-7B	7B	INT4 Q8 FP16	134 tok/s 88 tok/s 50 tok/s	~643 ms ~647 ms ~656 ms
-	Falcon3-7B	7B	INT4 Q8 FP16	130 tok/s 86 tok/s 50 tok/s	~644 ms ~648 ms ~656 ms
-	Hy-MT2 7B	7B	INT4 Q8 FP16	130 tok/s 86 tok/s 50 tok/s	~644 ms ~648 ms ~656 ms
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	126 tok/s 83 tok/s 48 tok/s	~670 ms ~674 ms ~683 ms
-	LensVLM 9B	9B	INT4 Q8 FP16	110 tok/s 71 tok/s 38 tok/s	~821 ms ~826 ms ~908 ms
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8 FP16	110 tok/s 71 tok/s 38 tok/s	~821 ms ~826 ms ~908 ms
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	107 tok/s 80 tok/s 50 tok/s	~558 ms ~561 ms ~569 ms
-	Yi-6B	6B	INT4 Q8 FP16	107 tok/s 80 tok/s 50 tok/s	~558 ms ~561 ms ~569 ms
-	Falcon3-10B	10B	INT4 Q8 FP16	102 tok/s 65 tok/s 34 tok/s	~909 ms ~914 ms ~1.0s
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4 Q8	102 tok/s 86 tok/s	~346 ms ~376 ms
-	Falcon2-11B	11B	INT4 Q8 FP16	99 tok/s 61 tok/s 30 tok/s	~997 ms ~1.0s ~1.2s
-	Hunyuan Lite	7B	INT4 Q8 FP16	98 tok/s 72 tok/s 44 tok/s	~646 ms ~650 ms ~658 ms
-	OLMo 3 7B Base	7B	INT4 Q8 FP16	98 tok/s 72 tok/s 44 tok/s	~646 ms ~650 ms ~658 ms
-	SEA-LION-7B	7.1B	INT4 Q8 FP16	97 tok/s 71 tok/s 44 tok/s	~655 ms ~659 ms ~667 ms
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8 FP16	97 tok/s 71 tok/s 44 tok/s	~655 ms ~659 ms ~667 ms
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8 FP16	90 tok/s 65 tok/s 38 tok/s	~734 ms ~738 ms ~812 ms
-	Typhoon-2-8B	8B	INT4 Q8 FP16	90 tok/s 65 tok/s 38 tok/s	~734 ms ~738 ms ~812 ms
-	GLM-4-9B	9B	INT4 Q8 FP16	83 tok/s 59 tok/s 34 tok/s	~824 ms ~829 ms ~911 ms
-	GLM-4-9B-Chat	9B	INT4 Q8 FP16	83 tok/s 59 tok/s 34 tok/s	~824 ms ~829 ms ~911 ms
-	GLM-4-9B-Chat-1M	9B	INT4 Q8 FP16	83 tok/s 59 tok/s 34 tok/s	~824 ms ~829 ms ~911 ms
-	Yi-9B	9B	INT4 Q8 FP16	83 tok/s 59 tok/s 34 tok/s	~824 ms ~829 ms ~911 ms
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8 FP16	82 tok/s 58 tok/s 33 tok/s	~841 ms ~846 ms ~930 ms
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8 FP16	82 tok/s 58 tok/s 33 tok/s	~841 ms ~846 ms ~930 ms
-	GreenMind-14B-R1	14B	INT4 Q8	60 tok/s 41 tok/s	~1.3s ~1.3s
-	Falcon-40B	40B	INT4	32 tok/s	~3.8s

AMD RX 7900 XTX (24GB)

VRAM Scaling by Context Length

Best Models to Run on AMD RX 7900 XTX (24GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions