What are the best models to run on Radeon PRO W6600 (8GB)?

Based on model rankings and 8 GB usable VRAM: Top Overall: Qwen3.5-9B (INT4) ~32 tok/s. Top Coding: K2 Horizon 7B (INT4) ~40 tok/s. Optimal Daily Driver: Granite 4.2 8B (INT4) ~36 tok/s. Maximum Model Size: HunyuanVideo 1.5 720P T2V (INT4) ~21 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 LLMs to Run on Radeon PRO W6600 (8GB)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#125	Qwen3.5-9B	9B	INT4	32 tok/s	~5.7s
#128	K2 Horizon 7B	7B	INT4	40 tok/s	~4.4s
#140	Qwen3.5-4B	4B	INT4 Q8	62 tok/s 39 tok/s	~2.6s ~2.6s
#150	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8	10 tok/s 6 tok/s	~17.9s ~19.5s
#158	Granite 4.2 8B	8B	INT4	36 tok/s	~5.1s
#159	Gemma 4 E4B	8B	INT4	36 tok/s	~5.1s
#178	OLMo 3 7B Think	7B	INT4	27 tok/s	~4.8s
#186	Gemma 3 12B	12B	INT4	22 tok/s	~8.9s
#189	Ministral 3 8B	8B	INT4	23 tok/s	~6.0s
#191	MiniCPM5-2B	2B	INT4 Q8 FP16	98 tok/s 67 tok/s 39 tok/s	~1.4s ~1.4s ~1.4s
#192	Qwen3.5-2B	2B	INT4 Q8 FP16	98 tok/s 67 tok/s 39 tok/s	~1.4s ~1.4s ~1.4s
#200	Gemma 4 E2B	5.1B	INT4 Q8	52 tok/s 30 tok/s	~3.3s ~3.6s
#202	DeepSeek-R1 8B	8B	INT4	37 tok/s	~5.1s
#204	Qwen3-4B	4B	INT4 Q8	62 tok/s 39 tok/s	~2.6s ~2.6s
#205	Ministral 3 3B	3B	INT4 Q8	49 tok/s 37 tok/s	~2.0s ~2.0s
#207	Gemma 2 9B	9B	INT4	32 tok/s	~5.7s
#208	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	162 tok/s 124 tok/s 81 tok/s	~564 ms ~566 ms ~570 ms
#210	Gemma 3 4B	4B	INT4 Q8	62 tok/s 39 tok/s	~2.6s ~2.6s
#211	Qwen3-8B	8B	INT4	36 tok/s	~5.1s
#213	DeepSeek-R1 7B	7B	INT4 Q8	41 tok/s 22 tok/s	~4.4s ~5.2s
#214	Qwen3-1.7B	1.7B	INT4 Q8 FP16	109 tok/s 76 tok/s 45 tok/s	~1.2s ~1.2s ~1.2s
#215	Llama 3.1 8B	8B	INT4	36 tok/s	~5.1s
#216	Gemma 3 270M	0.27B	INT4 Q8 FP16	198 tok/s 177 tok/s 142 tok/s	~189 ms ~190 ms ~191 ms
#218	Qwen3-0.6B	0.6B	INT4 Q8 FP16	182 tok/s 145 tok/s 100 tok/s	~418 ms ~419 ms ~422 ms
#219	Llama 3 8B	8B	INT4	36 tok/s	~5.1s
#221	Phi-4-Mini	3.8B	INT4 Q8	65 tok/s 41 tok/s	~2.5s ~2.5s
#223	Qwen2.5-7B	7B	INT4	40 tok/s	~4.4s
#224	Ministral-8B-2410	8B	INT4	36 tok/s	~5.1s
#225	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	117 tok/s 83 tok/s 50 tok/s	~1.0s ~1.1s ~1.1s
#226	Llama 3.2 3B	3B	INT4 Q8 FP16	76 tok/s 49 tok/s 26 tok/s	~2.0s ~2.0s ~2.2s
#228	Phi-3-mini	3.8B	INT4 Q8	65 tok/s 41 tok/s	~2.5s ~2.5s
#229	Gemma 3n E2B IT	6B (2B active)	INT4	46 tok/s	~1.6s
#230	Mistral-7B-Instruct-v0.1	7.3B	INT4	39 tok/s	~4.6s
#231	Llama 3.2 1B	1B	INT4 Q8 FP16	144 tok/s 108 tok/s 69 tok/s	~716 ms ~719 ms ~724 ms
#232	Gemma 2 2B	2B	INT4 Q8 FP16	98 tok/s 67 tok/s 39 tok/s	~1.4s ~1.4s ~1.4s
#234	OLMo 3 7B Instruct	7B	INT4	27 tok/s	~4.8s
#236	Qwen2-7B	7B	INT4	40 tok/s	~4.4s
#237	Phi-3-small	7B	INT4	40 tok/s	~4.4s
#238	Gemma 3 1B	1B	INT4 Q8 FP16	144 tok/s 108 tok/s 69 tok/s	~716 ms ~719 ms ~724 ms
#239	Mistral-7B-Instruct-v0.2	7.3B	INT4	39 tok/s	~4.6s
#241	Gemma 1 7B	7B	INT4	27 tok/s	~4.8s
#242	Gemma 1 2B	2B	INT4 Q8 FP16	106 tok/s 70 tok/s 40 tok/s	~1.4s ~1.4s ~1.4s
#243	ChatGLM3-6B	6B	INT4	30 tok/s	~4.2s
#244	ChatGLM2-6B	6B	INT4	30 tok/s	~4.2s
#245	ChatGLM-6B	6B	INT4	30 tok/s	~4.2s
Unranked Models (Sorted by Speed)
-	DINOv3 ViT-L/16	0.3B	INT4 Q8 FP16	224 tok/s 193 tok/s 148 tok/s	~210 ms ~210 ms ~212 ms
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	195 tok/s 158 tok/s 112 tok/s	~347 ms ~348 ms ~351 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	195 tok/s 158 tok/s 112 tok/s	~347 ms ~348 ms ~351 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	190 tok/s 169 tok/s 134 tok/s	~210 ms ~211 ms ~213 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	190 tok/s 169 tok/s 134 tok/s	~210 ms ~211 ms ~213 ms
-	Liquid D1-Omni 600M	0.6B	INT4 Q8 FP16	182 tok/s 145 tok/s 100 tok/s	~418 ms ~419 ms ~422 ms
-	PixArt-α XL 2 1024-MS	0.61B	INT4 Q8 FP16	181 tok/s 144 tok/s 98 tok/s	~424 ms ~426 ms ~429 ms
-	PixArt-Σ XL 2 1024-MS	0.61B	INT4 Q8 FP16	181 tok/s 144 tok/s 98 tok/s	~424 ms ~426 ms ~429 ms
-	Nemotron 3.5 Parse	0.63B	INT4 Q8 FP16	179 tok/s 141 tok/s 96 tok/s	~441 ms ~442 ms ~446 ms
-	SD 2.1 Turbo	0.86B	INT4 Q8 FP16	156 tok/s 119 tok/s 77 tok/s	~608 ms ~610 ms ~615 ms
-	Stable Diffusion 1.4	0.86B	INT4 Q8 FP16	156 tok/s 119 tok/s 77 tok/s	~608 ms ~610 ms ~615 ms
-	Stable Diffusion 2.1	0.86B	INT4 Q8 FP16	156 tok/s 119 tok/s 77 tok/s	~608 ms ~610 ms ~615 ms
-	Falcon-1B	1B	INT4 Q8 FP16	154 tok/s 113 tok/s 71 tok/s	~716 ms ~718 ms ~724 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	144 tok/s 108 tok/s 69 tok/s	~716 ms ~719 ms ~724 ms
-	MiniCPM-V 4.7 1B	1B	INT4 Q8 FP16	144 tok/s 108 tok/s 69 tok/s	~716 ms ~719 ms ~724 ms
-	Kandinsky 2.2 Decoder	1.22B	INT4 Q8 FP16	131 tok/s 95 tok/s 59 tok/s	~863 ms ~866 ms ~872 ms
-	Wan 2.1 T2V 1.3B	1.3B	INT4 Q8 FP16	127 tok/s 91 tok/s 56 tok/s	~912 ms ~915 ms ~922 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	117 tok/s 83 tok/s 50 tok/s	~1.0s ~1.1s ~1.1s
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	117 tok/s 83 tok/s 50 tok/s	~1.0s ~1.1s ~1.1s
-	SVD Image-to-Video XT	1.5B	INT4 Q8 FP16	117 tok/s 83 tok/s 50 tok/s	~1.0s ~1.1s ~1.1s
-	SVD Image-to-Video XT 1.1	1.5B	INT4 Q8 FP16	117 tok/s 83 tok/s 50 tok/s	~1.0s ~1.1s ~1.1s
-	LFM3 1.7B	1.7B	INT4 Q8 FP16	109 tok/s 76 tok/s 45 tok/s	~1.2s ~1.2s ~1.2s
-	LFM3 1.7B Base	1.7B	INT4 Q8 FP16	109 tok/s 76 tok/s 45 tok/s	~1.2s ~1.2s ~1.2s
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	109 tok/s 76 tok/s 45 tok/s	~1.2s ~1.2s ~1.2s
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	105 tok/s 73 tok/s 43 tok/s	~1.2s ~1.3s ~1.3s
-	Moondream 3.1	9B (2B active)	INT4	99 tok/s	~1.6s
-	CogVideoX 2B	2B	INT4 Q8 FP16	98 tok/s 67 tok/s 39 tok/s	~1.4s ~1.4s ~1.4s
-	LFM3 Audio 2B	2B	INT4 Q8 FP16	98 tok/s 67 tok/s 39 tok/s	~1.4s ~1.4s ~1.4s
-	LFM3 Omni 2B	2B	INT4 Q8 FP16	98 tok/s 67 tok/s 39 tok/s	~1.4s ~1.4s ~1.4s
-	LFM3 VL 2B	2B	INT4 Q8 FP16	98 tok/s 67 tok/s 39 tok/s	~1.4s ~1.4s ~1.4s
-	Liquid D1 2B	2B	INT4 Q8 FP16	98 tok/s 67 tok/s 39 tok/s	~1.4s ~1.4s ~1.4s
-	LTX-Video 0.9.1	2B	INT4 Q8 FP16	98 tok/s 67 tok/s 39 tok/s	~1.4s ~1.4s ~1.4s
-	Stable Diffusion 3 Medium	2B	INT4 Q8 FP16	98 tok/s 67 tok/s 39 tok/s	~1.4s ~1.4s ~1.4s
-	Kandinsky 3.0	2.1B	INT4 Q8 FP16	96 tok/s 65 tok/s 38 tok/s	~1.4s ~1.4s ~1.5s
-	LTX-2	2.2B	INT4 Q8 FP16	93 tok/s 63 tok/s 36 tok/s	~1.5s ~1.5s ~1.5s
-	SDXL 1.0 Refiner	2.31B	INT4 Q8 FP16	90 tok/s 60 tok/s 35 tok/s	~1.6s ~1.6s ~1.6s
-	Stable Diffusion 3.5 Medium	2.5B	INT4 Q8 FP16	86 tok/s 57 tok/s 33 tok/s	~1.7s ~1.7s ~1.7s
-	SDXL 1.0 Base	2.56B	INT4 Q8 FP16	85 tok/s 56 tok/s 30 tok/s	~1.7s ~1.7s ~1.9s
-	SDXL Turbo	2.56B	INT4 Q8 FP16	85 tok/s 56 tok/s 30 tok/s	~1.7s ~1.7s ~1.9s
-	Falcon-3B	3B	INT4 Q8 FP16	81 tok/s 51 tok/s 27 tok/s	~2.0s ~2.0s ~2.2s
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	80 tok/s 66 tok/s 45 tok/s	~917 ms ~920 ms ~927 ms
-	Phi-1	1.3B	INT4 Q8 FP16	80 tok/s 66 tok/s 45 tok/s	~917 ms ~920 ms ~927 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	80 tok/s 66 tok/s 45 tok/s	~917 ms ~920 ms ~927 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	77 tok/s 50 tok/s 27 tok/s	~2.0s ~2.0s ~2.2s
-	Falcon3-3B	3B	INT4 Q8 FP16	76 tok/s 49 tok/s 26 tok/s	~2.0s ~2.0s ~2.2s
-	LFM2-VL 3B	3B	INT4 Q8 FP16	76 tok/s 49 tok/s 26 tok/s	~2.0s ~2.0s ~2.2s
-	Liquid D1 3B	3B	INT4 Q8 FP16	76 tok/s 49 tok/s 26 tok/s	~2.0s ~2.0s ~2.2s
-	Liquid D1-VL 3B	3B	INT4 Q8 FP16	76 tok/s 49 tok/s 26 tok/s	~2.0s ~2.0s ~2.2s
-	Liquid LFM2-VL 3B D1	3B	INT4 Q8 FP16	76 tok/s 49 tok/s 26 tok/s	~2.0s ~2.0s ~2.2s
-	Qwen2.5-3B	3B	INT4 Q8 FP16	76 tok/s 49 tok/s 26 tok/s	~2.0s ~2.0s ~2.2s
-	Mellum 2.1 12B A2.5B Thinking	12B (2.5B active)	INT4	70 tok/s	~2.3s
-	FLUX.2 Klein 4B	4B	INT4 Q8	62 tok/s 39 tok/s	~2.6s ~2.6s
-	FrogNano 4B	4B	INT4 Q8	62 tok/s 39 tok/s	~2.6s ~2.6s
-	Mage-VL Cowork	4B	INT4 Q8	62 tok/s 39 tok/s	~2.6s ~2.6s
-	Spark X2.5 4B	4B	INT4 Q8	62 tok/s 39 tok/s	~2.6s ~2.6s
-	EmbeddingGemma 2	2B	INT4 Q8 FP16	59 tok/s 47 tok/s 30 tok/s	~1.4s ~1.4s ~1.5s
-	CogVideoX 1.5 5B	5B	INT4 Q8	52 tok/s 31 tok/s	~3.2s ~3.5s
-	CogVideoX 5B	5B	INT4 Q8	52 tok/s 31 tok/s	~3.2s ~3.5s
-	Phi-2	2.7B	INT4 Q8 FP16	52 tok/s 40 tok/s 23 tok/s	~1.8s ~1.8s ~2.2s
-	Wan 2.2 TI2V 5B	5B	INT4 Q8	52 tok/s 31 tok/s	~3.2s ~3.5s
-	MaLLaM-3B	3B	INT4 Q8	49 tok/s 37 tok/s	~2.0s ~2.0s
-	SmolLM3 3B	3B	INT4 Q8	49 tok/s 37 tok/s	~2.0s ~2.0s
-	Falcon-7B	7B	INT4 Q8	42 tok/s 22 tok/s	~4.4s ~5.2s
-	AuraFlow v0.2	6.8B	INT4 Q8	41 tok/s 22 tok/s	~4.3s ~5.1s
-	AuraFlow v0.3	6.8B	INT4 Q8	41 tok/s 22 tok/s	~4.3s ~5.1s
-	Falcon3-7B	7B	INT4	40 tok/s	~4.4s
-	Hy-MT2 7B	7B	INT4	40 tok/s	~4.4s
-	Mistral-7B-v0.1	7.3B	INT4	39 tok/s	~4.6s
-	Youtu Parsing Omni	8B	INT4	37 tok/s	~5.1s
-	Stable Diffusion 3.5 Large	8.1B	INT4	35 tok/s	~5.1s
-	Stable Diffusion 3.5 Large Turbo	8.1B	INT4	35 tok/s	~5.1s
-	FLUX.2 Klein 9B	9B	INT4	32 tok/s	~5.7s
-	LensVLM 9B	9B	INT4	32 tok/s	~5.7s
-	MiMo V2.6 Distill Qwen 9B	9B	INT4	32 tok/s	~5.7s
-	ChatGLM3-6B-32K	6B	INT4	30 tok/s	~4.2s
-	Yi-6B	6B	INT4	30 tok/s	~4.2s
-	Falcon3-10B	10B	INT4	28 tok/s	~6.8s
-	Mochi 1 Preview	10B	INT4	28 tok/s	~6.8s
-	Falcon2-11B	11B	INT4	27 tok/s	~7.5s
-	Hunyuan Lite	7B	INT4	27 tok/s	~4.8s
-	OLMo 3 7B Base	7B	INT4	27 tok/s	~4.8s
-	SEA-LION-7B	7.1B	INT4	26 tok/s	~4.9s
-	SEA-LION-7B-Instruct	7.1B	INT4	26 tok/s	~4.9s
-	FLUX.1 Canny Dev	11.9B	INT4	23 tok/s	~8.8s
-	FLUX.1 Depth Dev	11.9B	INT4	23 tok/s	~8.8s
-	FLUX.1 Dev	11.9B	INT4	23 tok/s	~8.8s
-	FLUX.1 Fill Dev	11.9B	INT4	23 tok/s	~8.8s
-	FLUX.1 Schnell	11.9B	INT4	23 tok/s	~8.8s
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4	23 tok/s	~6.0s
-	Typhoon-2-8B	8B	INT4	23 tok/s	~6.0s
-	HunyuanVideo 1.5 720P T2V	13B	INT4	21 tok/s	~9.6s

Radeon PRO W6600 (8GB)

VRAM Scaling by Context Length

Best Models to Run on Radeon PRO W6600 (8GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions