What are the best models to run on Radeon PRO W7800 (32GB)?

Based on model rankings and 32 GB usable VRAM: Top Overall: Qwen 3.8 27B (INT4) ~29 tok/s. Top Coding: Qwen3.5-27B (INT4) ~29 tok/s. Optimal Daily Driver: Phi-4 Reasoning Plus (INT4) ~38 tok/s. Maximum Model Size: Kimi Linear 48B A3B Instruct (INT4) ~41 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 Radeon PRO W7800 (32GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4 Q8	29 tok/s 16 tok/s	~3.3s ~3.6s
#48	Agnes 3.0 Flash	33B	INT4	24 tok/s	~4.0s
#72	K2 Horizon MoVA 36B A4B	36B (4B active)	INT4	113 tok/s	~645 ms
#83	Qwen3.5-27B	27B	INT4 Q8	29 tok/s 16 tok/s	~3.3s ~3.6s
#92	Sarvam-30B	32B (2.4B active)	INT4	143 tok/s	~451 ms
#94	Qwen3.6 35B A3B	35B (3B active)	INT4	130 tok/s	~526 ms
#96	Muse Glimmer 30B	30B	INT4 Q8	27 tok/s 14 tok/s	~3.6s ~4.3s
#100	Phi-4 Reasoning Plus	14B	INT4 Q8 FP16	38 tok/s 26 tok/s 13 tok/s	~1.7s ~1.7s ~2.1s
#101	MiMo V2 Flash	15B (309B active)	INT4 Q8	3 tok/s 2 tok/s	~35.8s ~36.1s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4 Q8	120 tok/s 81 tok/s	~591 ms ~645 ms
#109	Gemma 4 12B	11.95B	INT4 Q8 FP16	44 tok/s 30 tok/s 16 tok/s	~1.5s ~1.5s ~1.6s
#113	Qwen3.5-9B	9B	INT4 Q8 FP16	75 tok/s 46 tok/s 25 tok/s	~1.1s ~1.1s ~1.1s
#114	Qwen3.5-35B-A3B	35B (3B active)	INT4	130 tok/s	~526 ms
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	90 tok/s 57 tok/s 31 tok/s	~877 ms ~884 ms ~898 ms
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	25 tok/s 15 tok/s 8 tok/s	~3.5s ~3.5s ~3.6s
#129	Gemma 4 31B	30.7B	INT4 Q8	26 tok/s 13 tok/s	~3.7s ~4.4s
#131	Nemotron 3.5 Lightning	30B (3B active)	INT4	54 tok/s	~527 ms
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	132 tok/s 89 tok/s 51 tok/s	~516 ms ~520 ms ~528 ms
#133	Qwen3-30B-A3B	30B (3B active)	INT4 Q8	132 tok/s 88 tok/s	~516 ms ~608 ms
#136	GPT-OSS 20B	21B (3.6B active)	INT4 Q8	60 tok/s 49 tok/s	~561 ms ~612 ms
#146	Granite 4.2 8B	8B	INT4 Q8 FP16	82 tok/s 51 tok/s 28 tok/s	~997 ms ~1.0s ~1.0s
#150	Gemma 4 E4B	8B	INT4 Q8 FP16	82 tok/s 51 tok/s 28 tok/s	~997 ms ~1.0s ~1.0s
#151	Nemotron 3.5 Lightning 30B A3B	30B (3B active)	INT4 Q8	132 tok/s 88 tok/s	~516 ms ~608 ms
#153	Qwen3-32B	32B	INT4	25 tok/s	~3.9s
#155	Mistral Large 3	41B (675B active)	INT4	1 tok/s	~91.8s
#156	GLM-4	32B	INT4	20 tok/s	~3.9s
#158	GLM-4V	9B	INT4 Q8 FP16	55 tok/s 38 tok/s 22 tok/s	~1.1s ~1.1s ~1.2s
#159	Gemma 3 27B	27B	INT4 Q8	29 tok/s 16 tok/s	~3.3s ~3.6s
#160	Phi-4	14B	INT4 Q8 FP16	52 tok/s 31 tok/s 16 tok/s	~1.7s ~1.7s ~1.9s
#162	DeepSeek-R1 32B	32B	INT4 Q8	26 tok/s 13 tok/s	~3.9s ~4.6s
#164	Qwen2.5-32B	32B	INT4	25 tok/s	~3.9s
#168	Magistral Small	24B	INT4 Q8	25 tok/s 15 tok/s	~2.9s ~3.2s
#169	OLMo 3 7B Think	7B	INT4 Q8 FP16	66 tok/s 47 tok/s 28 tok/s	~881 ms ~887 ms ~902 ms
#170	OLMo 3.1 32B Instruct	32B	INT4	20 tok/s	~3.9s
#172	DeepSeek-R1 14B	14B	INT4 Q8 FP16	54 tok/s 32 tok/s 16 tok/s	~1.7s ~1.7s ~1.9s
#173	Ministral 3 14B	14B	INT4 Q8 FP16	38 tok/s 26 tok/s 13 tok/s	~1.7s ~1.7s ~2.1s
#177	Gemma 3 12B	12B	INT4 Q8 FP16	59 tok/s 36 tok/s 19 tok/s	~1.5s ~1.5s ~1.5s
#180	OLMo 3 32B Base	32B	INT4	20 tok/s	~3.9s
#181	Mistral-Small-2501	24B	INT4 Q8	33 tok/s 18 tok/s	~2.9s ~3.2s
#182	Ministral 3 8B	8B	INT4 Q8 FP16	60 tok/s 42 tok/s 25 tok/s	~1.0s ~1.0s ~1.0s
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	189 tok/s 140 tok/s 89 tok/s	~277 ms ~279 ms ~283 ms
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	189 tok/s 140 tok/s 89 tok/s	~277 ms ~279 ms ~283 ms
#185	Qwen2.5-14B	14B	INT4 Q8 FP16	52 tok/s 31 tok/s 16 tok/s	~1.7s ~1.7s ~1.9s
#186	Gemma 2 27B	27B	INT4 Q8	29 tok/s 16 tok/s	~3.3s ~3.6s
#187	OLMo 3.1 32B Think	32B	INT4	20 tok/s	~3.9s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	113 tok/s 74 tok/s 42 tok/s	~649 ms ~653 ms ~664 ms
#189	Qwen3-4B	4B	INT4 Q8 FP16	132 tok/s 89 tok/s 51 tok/s	~516 ms ~520 ms ~528 ms
#190	DeepSeek-R1 8B	8B	INT4 Q8 FP16	84 tok/s 52 tok/s 28 tok/s	~997 ms ~1.0s ~1.0s
#193	Ministral 3 3B	3B	INT4 Q8 FP16	107 tok/s 84 tok/s 56 tok/s	~399 ms ~402 ms ~408 ms
#195	Qwen3-14B	14B	INT4 Q8 FP16	52 tok/s 31 tok/s 16 tok/s	~1.7s ~1.7s ~1.9s
#196	Kimi Linear 48B A3B Instruct	48B (3B active)	INT4	41 tok/s	~668 ms
#197	Gemma 2 9B	9B	INT4 Q8 FP16	75 tok/s 46 tok/s 25 tok/s	~1.1s ~1.1s ~1.1s
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	90 tok/s 57 tok/s 31 tok/s	~877 ms ~884 ms ~898 ms
#199	Qwen3-8B	8B	INT4 Q8 FP16	82 tok/s 51 tok/s 28 tok/s	~997 ms ~1.0s ~1.0s
#200	Gemma 3 4B	4B	INT4 Q8 FP16	132 tok/s 89 tok/s 51 tok/s	~516 ms ~520 ms ~528 ms
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	93 tok/s 58 tok/s 32 tok/s	~877 ms ~883 ms ~898 ms
#202	Llama 3.1 8B	8B	INT4 Q8 FP16	82 tok/s 51 tok/s 28 tok/s	~997 ms ~1.0s ~1.0s
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	203 tok/s 154 tok/s 101 tok/s	~238 ms ~239 ms ~243 ms
#205	Command R	35B	INT4	18 tok/s	~4.2s
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	303 tok/s 283 tok/s 245 tok/s	~43 ms ~43 ms ~44 ms
#208	Llama 3 8B	8B	INT4 Q8 FP16	82 tok/s 51 tok/s 28 tok/s	~997 ms ~1.0s ~1.0s
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	288 tok/s 249 tok/s 191 tok/s	~88 ms ~88 ms ~89 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	136 tok/s 92 tok/s 54 tok/s	~492 ms ~496 ms ~504 ms
#211	Mixtral-8x7B-v0.1	46.7B (7B active)	INT4	76 tok/s	~1.1s
#212	Ministral-8B-2410	8B	INT4 Q8 FP16	82 tok/s 51 tok/s 28 tok/s	~997 ms ~1.0s ~1.0s
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	215 tok/s 166 tok/s 110 tok/s	~211 ms ~212 ms ~216 ms
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	155 tok/s 109 tok/s 65 tok/s	~396 ms ~399 ms ~405 ms
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	267 tok/s 223 tok/s 164 tok/s	~116 ms ~117 ms ~119 ms
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	136 tok/s 92 tok/s 54 tok/s	~492 ms ~496 ms ~504 ms
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	107 tok/s 92 tok/s 67 tok/s	~299 ms ~301 ms ~305 ms
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	88 tok/s 55 tok/s 30 tok/s	~912 ms ~919 ms ~934 ms
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	248 tok/s 202 tok/s 143 tok/s	~146 ms ~147 ms ~149 ms
#220	Gemma 2 2B	2B	INT4 Q8 FP16	189 tok/s 140 tok/s 89 tok/s	~277 ms ~279 ms ~283 ms
#221	Phi-3-medium	14B	INT4 Q8 FP16	52 tok/s 31 tok/s 16 tok/s	~1.7s ~1.7s ~1.9s
#222	OLMo 3 7B Instruct	7B	INT4 Q8 FP16	66 tok/s 47 tok/s 28 tok/s	~881 ms ~887 ms ~902 ms
#223	Yi-34B	34B	INT4	19 tok/s	~4.1s
#224	Qwen2-7B	7B	INT4 Q8 FP16	90 tok/s 57 tok/s 31 tok/s	~877 ms ~884 ms ~898 ms
#225	Phi-3-small	7B	INT4 Q8 FP16	90 tok/s 57 tok/s 31 tok/s	~877 ms ~884 ms ~898 ms
#226	Gemma 3 1B	1B	INT4 Q8 FP16	248 tok/s 202 tok/s 143 tok/s	~146 ms ~147 ms ~149 ms
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	88 tok/s 55 tok/s 30 tok/s	~912 ms ~919 ms ~934 ms
#229	Gemma 1 7B	7B	INT4 Q8 FP16	66 tok/s 47 tok/s 28 tok/s	~881 ms ~887 ms ~902 ms
#230	Gemma 1 2B	2B	INT4 Q8 FP16	200 tok/s 146 tok/s 91 tok/s	~277 ms ~279 ms ~283 ms
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	73 tok/s 53 tok/s 32 tok/s	~761 ms ~766 ms ~778 ms
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	73 tok/s 53 tok/s 32 tok/s	~761 ms ~766 ms ~778 ms
#233	ChatGLM-6B	6B	INT4 Q8 FP16	73 tok/s 53 tok/s 32 tok/s	~761 ms ~766 ms ~778 ms
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	300 tok/s 263 tok/s 208 tok/s	~74 ms ~74 ms ~75 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	300 tok/s 263 tok/s 208 tok/s	~74 ms ~74 ms ~75 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	295 tok/s 275 tok/s 235 tok/s	~47 ms ~47 ms ~48 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	295 tok/s 275 tok/s 235 tok/s	~47 ms ~47 ms ~48 ms
-	Falcon-1B	1B	INT4 Q8 FP16	259 tok/s 209 tok/s 146 tok/s	~146 ms ~147 ms ~149 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	248 tok/s 202 tok/s 143 tok/s	~146 ms ~147 ms ~149 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	215 tok/s 166 tok/s 110 tok/s	~211 ms ~212 ms ~216 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	215 tok/s 166 tok/s 110 tok/s	~211 ms ~212 ms ~216 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	203 tok/s 154 tok/s 101 tok/s	~238 ms ~239 ms ~243 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	199 tok/s 150 tok/s 97 tok/s	~250 ms ~252 ms ~255 ms
-	Falcon-3B	3B	INT4 Q8 FP16	163 tok/s 112 tok/s 66 tok/s	~396 ms ~399 ms ~405 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	161 tok/s 138 tok/s 101 tok/s	~186 ms ~188 ms ~190 ms
-	Phi-1	1.3B	INT4 Q8 FP16	161 tok/s 138 tok/s 101 tok/s	~186 ms ~188 ms ~190 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	161 tok/s 138 tok/s 101 tok/s	~186 ms ~188 ms ~190 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	157 tok/s 109 tok/s 66 tok/s	~396 ms ~399 ms ~405 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	155 tok/s 109 tok/s 65 tok/s	~396 ms ~399 ms ~405 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	155 tok/s 109 tok/s 65 tok/s	~396 ms ~399 ms ~405 ms
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4 Q8	137 tok/s 100 tok/s	~483 ms ~486 ms
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4 Q8	137 tok/s 100 tok/s	~483 ms ~486 ms
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4 Q8	133 tok/s 93 tok/s	~508 ms ~554 ms
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4 Q8	133 tok/s 93 tok/s	~508 ms ~554 ms
-	Hy-MT2 30B A3B	30B (3B active)	INT4 Q8	132 tok/s 88 tok/s	~516 ms ~608 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	132 tok/s 89 tok/s 51 tok/s	~516 ms ~520 ms ~528 ms
-	Phi-2	2.7B	INT4 Q8 FP16	113 tok/s 90 tok/s 60 tok/s	~363 ms ~365 ms ~371 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	107 tok/s 84 tok/s 56 tok/s	~399 ms ~402 ms ~408 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	107 tok/s 84 tok/s 56 tok/s	~399 ms ~402 ms ~408 ms
-	Falcon-7B	7B	INT4 Q8 FP16	94 tok/s 58 tok/s 32 tok/s	~877 ms ~883 ms ~897 ms
-	Falcon3-7B	7B	INT4 Q8 FP16	90 tok/s 57 tok/s 31 tok/s	~877 ms ~884 ms ~898 ms
-	Hy-MT2 7B	7B	INT4 Q8 FP16	90 tok/s 57 tok/s 31 tok/s	~877 ms ~884 ms ~898 ms
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	88 tok/s 55 tok/s 30 tok/s	~912 ms ~919 ms ~934 ms
-	LensVLM 9B	9B	INT4 Q8 FP16	75 tok/s 46 tok/s 25 tok/s	~1.1s ~1.1s ~1.1s
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8 FP16	75 tok/s 46 tok/s 25 tok/s	~1.1s ~1.1s ~1.1s
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	73 tok/s 53 tok/s 32 tok/s	~761 ms ~766 ms ~778 ms
-	Yi-6B	6B	INT4 Q8 FP16	73 tok/s 53 tok/s 32 tok/s	~761 ms ~766 ms ~778 ms
-	Falcon3-10B	10B	INT4 Q8 FP16	69 tok/s 42 tok/s 23 tok/s	~1.2s ~1.2s ~1.3s
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4 Q8	69 tok/s 60 tok/s	~471 ms ~473 ms
-	Falcon2-11B	11B	INT4 Q8 FP16	66 tok/s 39 tok/s 21 tok/s	~1.4s ~1.4s ~1.4s
-	Hunyuan Lite	7B	INT4 Q8 FP16	66 tok/s 47 tok/s 28 tok/s	~881 ms ~887 ms ~902 ms
-	OLMo 3 7B Base	7B	INT4 Q8 FP16	66 tok/s 47 tok/s 28 tok/s	~881 ms ~887 ms ~902 ms
-	SEA-LION-7B	7.1B	INT4 Q8 FP16	65 tok/s 46 tok/s 28 tok/s	~893 ms ~899 ms ~914 ms
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8 FP16	65 tok/s 46 tok/s 28 tok/s	~893 ms ~899 ms ~914 ms
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8 FP16	60 tok/s 42 tok/s 25 tok/s	~1.0s ~1.0s ~1.0s
-	Typhoon-2-8B	8B	INT4 Q8 FP16	60 tok/s 42 tok/s 25 tok/s	~1.0s ~1.0s ~1.0s
-	GLM-4-9B	9B	INT4 Q8 FP16	55 tok/s 38 tok/s 22 tok/s	~1.1s ~1.1s ~1.2s
-	GLM-4-9B-Chat	9B	INT4 Q8 FP16	55 tok/s 38 tok/s 22 tok/s	~1.1s ~1.1s ~1.2s
-	GLM-4-9B-Chat-1M	9B	INT4 Q8 FP16	55 tok/s 38 tok/s 22 tok/s	~1.1s ~1.1s ~1.2s
-	Yi-9B	9B	INT4 Q8 FP16	55 tok/s 38 tok/s 22 tok/s	~1.1s ~1.1s ~1.2s
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8 FP16	54 tok/s 37 tok/s 22 tok/s	~1.1s ~1.2s ~1.2s
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8 FP16	54 tok/s 37 tok/s 22 tok/s	~1.1s ~1.2s ~1.2s
-	GreenMind-14B-R1	14B	INT4 Q8 FP16	38 tok/s 26 tok/s 13 tok/s	~1.7s ~1.7s ~2.1s
-	Falcon-40B	40B	INT4	21 tok/s	~4.8s

Radeon PRO W7800 (32GB)

VRAM Scaling by Context Length

Best Models to Run on Radeon PRO W7800 (32GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions