What are the best models to run on NVIDIA RTX 4080 (16GB)?

Based on model rankings and 16 GB usable VRAM: Top Overall: Qwen 3.8 27B (INT4) ~32 tok/s. Top Coding: Qwen3.5-27B (INT4) ~32 tok/s. Optimal Daily Driver: Phi-4 Reasoning Plus (INT4) ~47 tok/s. Maximum Model Size: ERNIE-4.5-VL-28B-A3B (INT4) ~137 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 NVIDIA RTX 4080 (16GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4	32 tok/s	~1.8s
#83	Qwen3.5-27B	27B	INT4	32 tok/s	~1.8s
#100	Phi-4 Reasoning Plus	14B	INT4	47 tok/s	~815 ms
#101	MiMo V2 Flash	15B (309B active)	INT4	3 tok/s	~18.1s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4	132 tok/s	~303 ms
#109	Gemma 4 12B	11.95B	INT4 Q8	53 tok/s 33 tok/s	~699 ms ~829 ms
#113	Qwen3.5-9B	9B	INT4 Q8	89 tok/s 56 tok/s	~527 ms ~534 ms
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	106 tok/s 68 tok/s 34 tok/s	~414 ms ~419 ms ~503 ms
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	31 tok/s 18 tok/s 10 tok/s	~1.6s ~1.7s ~1.7s
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	151 tok/s 104 tok/s 62 tok/s	~245 ms ~248 ms ~255 ms
#136	GPT-OSS 20B	21B (3.6B active)	INT4	65 tok/s	~314 ms
#146	Granite 4.2 8B	8B	INT4 Q8	97 tok/s 61 tok/s	~470 ms ~476 ms
#150	Gemma 4 E4B	8B	INT4 Q8	97 tok/s 61 tok/s	~470 ms ~476 ms
#158	GLM-4V	9B	INT4 Q8	66 tok/s 44 tok/s	~531 ms ~582 ms
#159	Gemma 3 27B	27B	INT4	32 tok/s	~1.8s
#160	Phi-4	14B	INT4 Q8	63 tok/s 34 tok/s	~810 ms ~961 ms
#169	OLMo 3 7B Think	7B	INT4 Q8	78 tok/s 56 tok/s	~417 ms ~422 ms
#172	DeepSeek-R1 14B	14B	INT4 Q8	64 tok/s 35 tok/s	~809 ms ~960 ms
#173	Ministral 3 14B	14B	INT4	47 tok/s	~815 ms
#177	Gemma 3 12B	12B	INT4 Q8	71 tok/s 41 tok/s	~697 ms ~765 ms
#181	Mistral-Small-2501	24B	INT4	38 tok/s	~1.5s
#182	Ministral 3 8B	8B	INT4 Q8	72 tok/s 51 tok/s	~473 ms ~479 ms
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	209 tok/s 160 tok/s 105 tok/s	~133 ms ~135 ms ~138 ms
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	209 tok/s 160 tok/s 105 tok/s	~133 ms ~135 ms ~138 ms
#185	Qwen2.5-14B	14B	INT4 Q8	63 tok/s 34 tok/s	~810 ms ~961 ms
#186	Gemma 2 27B	27B	INT4	32 tok/s	~1.8s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	131 tok/s 87 tok/s 50 tok/s	~307 ms ~311 ms ~319 ms
#189	Qwen3-4B	4B	INT4 Q8 FP16	151 tok/s 104 tok/s 62 tok/s	~245 ms ~248 ms ~255 ms
#190	DeepSeek-R1 8B	8B	INT4 Q8	99 tok/s 62 tok/s	~469 ms ~475 ms
#193	Ministral 3 3B	3B	INT4 Q8 FP16	125 tok/s 100 tok/s 67 tok/s	~191 ms ~193 ms ~198 ms
#195	Qwen3-14B	14B	INT4 Q8	63 tok/s 34 tok/s	~810 ms ~961 ms
#197	Gemma 2 9B	9B	INT4 Q8	89 tok/s 56 tok/s	~527 ms ~534 ms
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	106 tok/s 68 tok/s 34 tok/s	~414 ms ~419 ms ~503 ms
#199	Qwen3-8B	8B	INT4 Q8	97 tok/s 61 tok/s	~470 ms ~476 ms
#200	Gemma 3 4B	4B	INT4 Q8 FP16	151 tok/s 104 tok/s 62 tok/s	~245 ms ~248 ms ~255 ms
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	109 tok/s 69 tok/s 37 tok/s	~413 ms ~419 ms ~466 ms
#202	Llama 3.1 8B	8B	INT4 Q8	97 tok/s 61 tok/s	~470 ms ~476 ms
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	223 tok/s 174 tok/s 117 tok/s	~115 ms ~116 ms ~119 ms
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	314 tok/s 297 tok/s 262 tok/s	~24 ms ~25 ms ~25 ms
#208	Llama 3 8B	8B	INT4 Q8	97 tok/s 61 tok/s	~470 ms ~476 ms
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	301 tok/s 266 tok/s 211 tok/s	~45 ms ~46 ms ~46 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	155 tok/s 108 tok/s 64 tok/s	~234 ms ~237 ms ~243 ms
#212	Ministral-8B-2410	8B	INT4 Q8	97 tok/s 61 tok/s	~470 ms ~476 ms
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	234 tok/s 186 tok/s 128 tok/s	~103 ms ~104 ms ~106 ms
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	175 tok/s 126 tok/s 78 tok/s	~189 ms ~191 ms ~196 ms
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	282 tok/s 242 tok/s 184 tok/s	~58 ms ~59 ms ~60 ms
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	155 tok/s 108 tok/s 64 tok/s	~234 ms ~237 ms ~243 ms
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	125 tok/s 108 tok/s 76 tok/s	~145 ms ~146 ms ~162 ms
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	103 tok/s 66 tok/s 33 tok/s	~430 ms ~436 ms ~523 ms
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	265 tok/s 222 tok/s 163 tok/s	~72 ms ~73 ms ~75 ms
#220	Gemma 2 2B	2B	INT4 Q8 FP16	209 tok/s 160 tok/s 105 tok/s	~133 ms ~135 ms ~138 ms
#221	Phi-3-medium	14B	INT4 Q8	63 tok/s 34 tok/s	~810 ms ~961 ms
#222	OLMo 3 7B Instruct	7B	INT4 Q8	78 tok/s 56 tok/s	~417 ms ~422 ms
#224	Qwen2-7B	7B	INT4 Q8 FP16	106 tok/s 68 tok/s 34 tok/s	~414 ms ~419 ms ~503 ms
#225	Phi-3-small	7B	INT4 Q8 FP16	106 tok/s 68 tok/s 34 tok/s	~414 ms ~419 ms ~503 ms
#226	Gemma 3 1B	1B	INT4 Q8 FP16	265 tok/s 222 tok/s 163 tok/s	~72 ms ~73 ms ~75 ms
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	103 tok/s 66 tok/s 33 tok/s	~430 ms ~436 ms ~523 ms
#229	Gemma 1 7B	7B	INT4 Q8	78 tok/s 56 tok/s	~417 ms ~422 ms
#230	Gemma 1 2B	2B	INT4 Q8 FP16	219 tok/s 165 tok/s 107 tok/s	~133 ms ~135 ms ~138 ms
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	86 tok/s 63 tok/s 37 tok/s	~361 ms ~365 ms ~406 ms
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	86 tok/s 63 tok/s 37 tok/s	~361 ms ~365 ms ~406 ms
#233	ChatGLM-6B	6B	INT4 Q8 FP16	86 tok/s 63 tok/s 37 tok/s	~361 ms ~365 ms ~406 ms
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	311 tok/s 279 tok/s 227 tok/s	~39 ms ~39 ms ~40 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	311 tok/s 279 tok/s 227 tok/s	~39 ms ~39 ms ~40 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	308 tok/s 290 tok/s 253 tok/s	~26 ms ~26 ms ~27 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	308 tok/s 290 tok/s 253 tok/s	~26 ms ~26 ms ~27 ms
-	Falcon-1B	1B	INT4 Q8 FP16	275 tok/s 228 tok/s 166 tok/s	~72 ms ~73 ms ~75 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	265 tok/s 222 tok/s 163 tok/s	~72 ms ~73 ms ~75 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	234 tok/s 186 tok/s 128 tok/s	~103 ms ~104 ms ~106 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	234 tok/s 186 tok/s 128 tok/s	~103 ms ~104 ms ~106 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	223 tok/s 174 tok/s 117 tok/s	~115 ms ~116 ms ~119 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	219 tok/s 169 tok/s 113 tok/s	~121 ms ~122 ms ~125 ms
-	Falcon-3B	3B	INT4 Q8 FP16	183 tok/s 130 tok/s 79 tok/s	~189 ms ~191 ms ~196 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	181 tok/s 157 tok/s 118 tok/s	~92 ms ~93 ms ~95 ms
-	Phi-1	1.3B	INT4 Q8 FP16	181 tok/s 157 tok/s 118 tok/s	~92 ms ~93 ms ~95 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	181 tok/s 157 tok/s 118 tok/s	~92 ms ~93 ms ~95 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	177 tok/s 127 tok/s 78 tok/s	~189 ms ~191 ms ~196 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	175 tok/s 126 tok/s 78 tok/s	~189 ms ~191 ms ~196 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	175 tok/s 126 tok/s 78 tok/s	~189 ms ~191 ms ~196 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	151 tok/s 104 tok/s 62 tok/s	~245 ms ~248 ms ~255 ms
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4	149 tok/s	~249 ms
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4	149 tok/s	~249 ms
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4	137 tok/s	~282 ms
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4	137 tok/s	~282 ms
-	Phi-2	2.7B	INT4 Q8 FP16	130 tok/s 105 tok/s 72 tok/s	~175 ms ~176 ms ~181 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	125 tok/s 100 tok/s 67 tok/s	~191 ms ~193 ms ~198 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	125 tok/s 100 tok/s 67 tok/s	~191 ms ~193 ms ~198 ms
-	Falcon-7B	7B	INT4 Q8 FP16	110 tok/s 70 tok/s 37 tok/s	~413 ms ~419 ms ~466 ms
-	Falcon3-7B	7B	INT4 Q8 FP16	106 tok/s 68 tok/s 34 tok/s	~414 ms ~419 ms ~503 ms
-	Hy-MT2 7B	7B	INT4 Q8 FP16	106 tok/s 68 tok/s 34 tok/s	~414 ms ~419 ms ~503 ms
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	103 tok/s 66 tok/s 33 tok/s	~430 ms ~436 ms ~523 ms
-	LensVLM 9B	9B	INT4 Q8	89 tok/s 56 tok/s	~527 ms ~534 ms
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8	89 tok/s 56 tok/s	~527 ms ~534 ms
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	86 tok/s 63 tok/s 37 tok/s	~361 ms ~365 ms ~406 ms
-	Yi-6B	6B	INT4 Q8 FP16	86 tok/s 63 tok/s 37 tok/s	~361 ms ~365 ms ~406 ms
-	Falcon3-10B	10B	INT4 Q8	82 tok/s 51 tok/s	~583 ms ~591 ms
-	Falcon2-11B	11B	INT4 Q8	79 tok/s 48 tok/s	~639 ms ~647 ms
-	Hunyuan Lite	7B	INT4 Q8	78 tok/s 56 tok/s	~417 ms ~422 ms
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4	78 tok/s	~245 ms
-	OLMo 3 7B Base	7B	INT4 Q8	78 tok/s 56 tok/s	~417 ms ~422 ms
-	SEA-LION-7B	7.1B	INT4 Q8	78 tok/s 56 tok/s	~423 ms ~428 ms
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8	78 tok/s 56 tok/s	~423 ms ~428 ms
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8	72 tok/s 51 tok/s	~473 ms ~479 ms
-	Typhoon-2-8B	8B	INT4 Q8	72 tok/s 51 tok/s	~473 ms ~479 ms
-	GLM-4-9B	9B	INT4 Q8	66 tok/s 44 tok/s	~531 ms ~582 ms
-	GLM-4-9B-Chat	9B	INT4 Q8	66 tok/s 44 tok/s	~531 ms ~582 ms
-	GLM-4-9B-Chat-1M	9B	INT4 Q8	66 tok/s 44 tok/s	~531 ms ~582 ms
-	Yi-9B	9B	INT4 Q8	66 tok/s 44 tok/s	~531 ms ~582 ms
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8	65 tok/s 43 tok/s	~542 ms ~595 ms
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8	65 tok/s 43 tok/s	~542 ms ~595 ms
-	GreenMind-14B-R1	14B	INT4	47 tok/s	~815 ms

NVIDIA RTX 4080 (16GB)

VRAM Scaling by Context Length

Best Models to Run on NVIDIA RTX 4080 (16GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions