What are the best models to run on AMD RX 6600 XT (8GB)?

Based on model rankings and 8 GB usable VRAM: Top Overall: Qwen3.5-9B (INT4) ~37 tok/s. Top Coding: K2 Horizon 7B (INT4) ~46 tok/s. Optimal Daily Driver: Granite 4.2 8B (INT4) ~41 tok/s. Maximum Model Size: Mellum 2.1 12B A2.5B Thinking (INT4) ~82 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 AMD RX 6600 XT (8GB)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#121	Qwen3.5-9B	9B	INT4	37 tok/s	~4.7s
#127	K2 Horizon 7B	7B	INT4	46 tok/s	~3.7s
#136	Qwen3.5-4B	4B	INT4 Q8	73 tok/s 45 tok/s	~2.2s ~2.2s
#145	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8	12 tok/s 6 tok/s	~14.9s ~16.3s
#152	Granite 4.2 8B	8B	INT4	41 tok/s	~4.2s
#157	Gemma 4 E4B	8B	INT4	41 tok/s	~4.2s
#180	Gemma 3 12B	12B	INT4	26 tok/s	~7.4s
#185	Ministral 3 8B	8B	INT4	26 tok/s	~5.0s
#187	OLMo 3 7B Think	7B	INT4	31 tok/s	~4.0s
#188	Qwen3.5-2B	2B	INT4 Q8 FP16	117 tok/s 79 tok/s 46 tok/s	~1.2s ~1.2s ~1.2s
#189	MiniCPM5-2B	2B	INT4 Q8 FP16	117 tok/s 79 tok/s 46 tok/s	~1.2s ~1.2s ~1.2s
#193	Gemma 4 E2B	5.1B	INT4 Q8	60 tok/s 35 tok/s	~2.7s ~3.0s
#195	Ministral 3 3B	3B	INT4 Q8	57 tok/s 43 tok/s	~1.7s ~1.7s
#196	DeepSeek-R1 8B	8B	INT4	43 tok/s	~4.2s
#198	Gemma 2 9B	9B	INT4	37 tok/s	~4.7s
#201	Qwen3-4B	4B	INT4 Q8	73 tok/s 45 tok/s	~2.2s ~2.2s
#203	Gemma 3 4B	4B	INT4 Q8	73 tok/s 45 tok/s	~2.2s ~2.2s
#204	DeepSeek-R1 7B	7B	INT4 Q8	48 tok/s 25 tok/s	~3.7s ~4.4s
#206	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	198 tok/s 149 tok/s 96 tok/s	~470 ms ~472 ms ~475 ms
#207	Llama 3.1 8B	8B	INT4	41 tok/s	~4.2s
#207	Qwen3-8B	8B	INT4	41 tok/s	~4.2s
#212	Llama 3 8B	8B	INT4	41 tok/s	~4.2s
#213	Qwen3-1.7B	1.7B	INT4 Q8 FP16	130 tok/s 89 tok/s 53 tok/s	~987 ms ~990 ms ~998 ms
#214	Gemma 3 270M	0.27B	INT4 Q8 FP16	246 tok/s 219 tok/s 172 tok/s	~158 ms ~158 ms ~159 ms
#216	Qwen3-0.6B	0.6B	INT4 Q8 FP16	226 tok/s 176 tok/s 119 tok/s	~348 ms ~349 ms ~352 ms
#217	Phi-4-Mini	3.8B	INT4 Q8	76 tok/s 47 tok/s	~2.1s ~2.1s
#218	Qwen2.5-7B	7B	INT4	46 tok/s	~3.7s
#220	Ministral-8B-2410	8B	INT4	41 tok/s	~4.2s
#221	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	141 tok/s 98 tok/s 59 tok/s	~873 ms ~876 ms ~883 ms
#222	Llama 3.2 3B	3B	INT4 Q8 FP16	90 tok/s 58 tok/s 31 tok/s	~1.7s ~1.7s ~1.8s
#223	Phi-3-mini	3.8B	INT4 Q8	76 tok/s 47 tok/s	~2.1s ~2.1s
#224	Gemma 3n E2B IT	6B (2B active)	INT4	54 tok/s	~1.3s
#225	Gemma 2 2B	2B	INT4 Q8 FP16	117 tok/s 79 tok/s 46 tok/s	~1.2s ~1.2s ~1.2s
#226	Mistral-7B-Instruct-v0.1	7.3B	INT4	45 tok/s	~3.9s
#227	Llama 3.2 1B	1B	INT4 Q8 FP16	175 tok/s 129 tok/s 81 tok/s	~597 ms ~599 ms ~604 ms
#230	OLMo 3 7B Instruct	7B	INT4	31 tok/s	~4.0s
#231	Qwen2-7B	7B	INT4	46 tok/s	~3.7s
#232	Phi-3-small	7B	INT4	46 tok/s	~3.7s
#233	Mistral-7B-Instruct-v0.2	7.3B	INT4	45 tok/s	~3.9s
#234	Gemma 3 1B	1B	INT4 Q8 FP16	175 tok/s 129 tok/s 81 tok/s	~597 ms ~599 ms ~604 ms
#236	Gemma 1 7B	7B	INT4	31 tok/s	~4.0s
#237	Gemma 1 2B	2B	INT4 Q8 FP16	127 tok/s 83 tok/s 47 tok/s	~1.2s ~1.2s ~1.2s
#238	ChatGLM3-6B	6B	INT4	34 tok/s	~3.5s
#239	ChatGLM2-6B	6B	INT4	34 tok/s	~3.5s
#240	ChatGLM-6B	6B	INT4	34 tok/s	~3.5s
Unranked Models (Sorted by Speed)
-	DINOv3 ViT-L/16	0.3B	INT4 Q8 FP16	282 tok/s 240 tok/s 180 tok/s	~174 ms ~175 ms ~176 ms
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	242 tok/s 194 tok/s 134 tok/s	~289 ms ~290 ms ~292 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	242 tok/s 194 tok/s 134 tok/s	~289 ms ~290 ms ~292 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	236 tok/s 208 tok/s 162 tok/s	~175 ms ~176 ms ~177 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	236 tok/s 208 tok/s 162 tok/s	~175 ms ~176 ms ~177 ms
-	Liquid D1-Omni 600M	0.6B	INT4 Q8 FP16	226 tok/s 176 tok/s 119 tok/s	~348 ms ~349 ms ~352 ms
-	Falcon-1B	1B	INT4 Q8 FP16	188 tok/s 135 tok/s 83 tok/s	~597 ms ~599 ms ~603 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	175 tok/s 129 tok/s 81 tok/s	~597 ms ~599 ms ~604 ms
-	MiniCPM-V 4.7 1B	1B	INT4 Q8 FP16	175 tok/s 129 tok/s 81 tok/s	~597 ms ~599 ms ~604 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	141 tok/s 98 tok/s 59 tok/s	~873 ms ~876 ms ~883 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	141 tok/s 98 tok/s 59 tok/s	~873 ms ~876 ms ~883 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	130 tok/s 89 tok/s 53 tok/s	~987 ms ~990 ms ~998 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	126 tok/s 86 tok/s 50 tok/s	~1.0s ~1.0s ~1.1s
-	Falcon-3B	3B	INT4 Q8 FP16	96 tok/s 60 tok/s 31 tok/s	~1.7s ~1.7s ~1.8s
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	94 tok/s 77 tok/s 53 tok/s	~765 ms ~767 ms ~773 ms
-	Phi-1	1.3B	INT4 Q8 FP16	94 tok/s 77 tok/s 53 tok/s	~765 ms ~767 ms ~773 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	94 tok/s 77 tok/s 53 tok/s	~765 ms ~767 ms ~773 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	91 tok/s 58 tok/s 31 tok/s	~1.7s ~1.7s ~1.8s
-	Falcon3-3B	3B	INT4 Q8 FP16	90 tok/s 58 tok/s 31 tok/s	~1.7s ~1.7s ~1.8s
-	LFM2-VL 3B	3B	INT4 Q8 FP16	90 tok/s 58 tok/s 31 tok/s	~1.7s ~1.7s ~1.8s
-	Liquid D1 3B	3B	INT4 Q8 FP16	90 tok/s 58 tok/s 31 tok/s	~1.7s ~1.7s ~1.8s
-	Liquid D1-VL 3B	3B	INT4 Q8 FP16	90 tok/s 58 tok/s 31 tok/s	~1.7s ~1.7s ~1.8s
-	Liquid LFM2-VL 3B D1	3B	INT4 Q8 FP16	90 tok/s 58 tok/s 31 tok/s	~1.7s ~1.7s ~1.8s
-	Qwen2.5-3B	3B	INT4 Q8 FP16	90 tok/s 58 tok/s 31 tok/s	~1.7s ~1.7s ~1.8s
-	Mellum 2.1 12B A2.5B Thinking	12B (2.5B active)	INT4	82 tok/s	~1.9s
-	FrogNano 4B	4B	INT4 Q8	73 tok/s 45 tok/s	~2.2s ~2.2s
-	Spark X2.5 4B	4B	INT4 Q8	73 tok/s 45 tok/s	~2.2s ~2.2s
-	EmbeddingGemma 2	2B	INT4 Q8 FP16	69 tok/s 55 tok/s 35 tok/s	~1.2s ~1.2s ~1.3s
-	Phi-2	2.7B	INT4 Q8 FP16	60 tok/s 46 tok/s 26 tok/s	~1.5s ~1.5s ~1.8s
-	MaLLaM-3B	3B	INT4 Q8	57 tok/s 43 tok/s	~1.7s ~1.7s
-	SmolLM3 3B	3B	INT4 Q8	57 tok/s 43 tok/s	~1.7s ~1.7s
-	Falcon-7B	7B	INT4 Q8	49 tok/s 26 tok/s	~3.7s ~4.4s
-	Falcon3-7B	7B	INT4	46 tok/s	~3.7s
-	Hy-MT2 7B	7B	INT4	46 tok/s	~3.7s
-	Mistral-7B-v0.1	7.3B	INT4	45 tok/s	~3.9s
-	Youtu Parsing Omni	8B	INT4	43 tok/s	~4.2s
-	LensVLM 9B	9B	INT4	37 tok/s	~4.7s
-	MiMo V2.6 Distill Qwen 9B	9B	INT4	37 tok/s	~4.7s
-	ChatGLM3-6B-32K	6B	INT4	34 tok/s	~3.5s
-	Yi-6B	6B	INT4	34 tok/s	~3.5s
-	Falcon3-10B	10B	INT4	32 tok/s	~5.7s
-	Falcon2-11B	11B	INT4	31 tok/s	~6.3s
-	Hunyuan Lite	7B	INT4	31 tok/s	~4.0s
-	OLMo 3 7B Base	7B	INT4	31 tok/s	~4.0s
-	SEA-LION-7B	7.1B	INT4	30 tok/s	~4.1s
-	SEA-LION-7B-Instruct	7.1B	INT4	30 tok/s	~4.1s
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4	26 tok/s	~5.0s
-	Typhoon-2-8B	8B	INT4	26 tok/s	~5.0s

AMD RX 6600 XT (8GB)

VRAM Scaling by Context Length

Best Models to Run on AMD RX 6600 XT (8GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions