What are the best models to run on NVIDIA RTX 3080 (12GB)?

Based on model rankings and 12 GB usable VRAM: Top Overall: Phi-4 Reasoning Plus (INT4) ~52 tok/s. Top Coding: K2 Horizon 7B (INT4) ~125 tok/s. Optimal Daily Driver: Qwen3.5-9B (INT4) ~106 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 NVIDIA RTX 3080 (12GB)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#102	Phi-4 Reasoning Plus	14B	INT4	52 tok/s	~1.5s
#121	Qwen3.5-9B	9B	INT4 Q8	106 tok/s 65 tok/s	~824 ms ~900 ms
#127	K2 Horizon 7B	7B	INT4 Q8	125 tok/s 83 tok/s	~646 ms ~650 ms
#130	Gemma 4 12B	11.95B	INT4	61 tok/s	~1.2s
#136	Qwen3.5-4B	4B	INT4 Q8 FP16	173 tok/s 123 tok/s 75 tok/s	~381 ms ~384 ms ~389 ms
#140	GLM-4V	9B	INT4	80 tok/s	~827 ms
#145	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	39 tok/s 23 tok/s 12 tok/s	~2.6s ~2.6s ~2.9s
#152	Granite 4.2 8B	8B	INT4 Q8	115 tok/s 71 tok/s	~734 ms ~802 ms
#157	Gemma 4 E4B	8B	INT4 Q8	115 tok/s 71 tok/s	~734 ms ~802 ms
#164	Phi-4	14B	INT4	76 tok/s	~1.3s
#176	DeepSeek-R1 14B	14B	INT4	78 tok/s	~1.3s
#177	Ministral 3 14B	14B	INT4	52 tok/s	~1.5s
#180	Gemma 3 12B	12B	INT4	86 tok/s	~1.1s
#185	Ministral 3 8B	8B	INT4 Q8	87 tok/s 56 tok/s	~737 ms ~870 ms
#187	OLMo 3 7B Think	7B	INT4 Q8	94 tok/s 66 tok/s	~649 ms ~708 ms
#188	Qwen3.5-2B	2B	INT4 Q8 FP16	231 tok/s 182 tok/s 124 tok/s	~205 ms ~207 ms ~209 ms
#189	MiniCPM5-2B	2B	INT4 Q8 FP16	231 tok/s 182 tok/s 124 tok/s	~205 ms ~207 ms ~209 ms
#190	Qwen2.5-14B	14B	INT4	76 tok/s	~1.3s
#193	Gemma 4 E2B	5.1B	INT4 Q8 FP16	152 tok/s 104 tok/s 56 tok/s	~478 ms ~481 ms ~572 ms
#195	Ministral 3 3B	3B	INT4 Q8 FP16	145 tok/s 118 tok/s 81 tok/s	~295 ms ~296 ms ~300 ms
#196	DeepSeek-R1 8B	8B	INT4 Q8	117 tok/s 76 tok/s	~734 ms ~739 ms
#198	Gemma 2 9B	9B	INT4 Q8	106 tok/s 65 tok/s	~824 ms ~900 ms
#201	Qwen3-4B	4B	INT4 Q8 FP16	173 tok/s 123 tok/s 75 tok/s	~381 ms ~384 ms ~389 ms
#202	Qwen3-14B	14B	INT4	76 tok/s	~1.3s
#203	Gemma 3 4B	4B	INT4 Q8 FP16	173 tok/s 123 tok/s 75 tok/s	~381 ms ~384 ms ~389 ms
#204	DeepSeek-R1 7B	7B	INT4 Q8	128 tok/s 84 tok/s	~646 ms ~650 ms
#206	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	298 tok/s 262 tok/s 206 tok/s	~88 ms ~88 ms ~89 ms
#207	Llama 3.1 8B	8B	INT4 Q8	115 tok/s 71 tok/s	~734 ms ~802 ms
#207	Qwen3-8B	8B	INT4 Q8	115 tok/s 71 tok/s	~734 ms ~802 ms
#212	Llama 3 8B	8B	INT4 Q8	115 tok/s 71 tok/s	~734 ms ~802 ms
#213	Qwen3-1.7B	1.7B	INT4 Q8 FP16	244 tok/s 197 tok/s 137 tok/s	~177 ms ~178 ms ~180 ms
#214	Gemma 3 270M	0.27B	INT4 Q8 FP16	325 tok/s 310 tok/s 280 tok/s	~34 ms ~34 ms ~34 ms
#216	Qwen3-0.6B	0.6B	INT4 Q8 FP16	314 tok/s 283 tok/s 232 tok/s	~66 ms ~67 ms ~68 ms
#217	Phi-4-Mini	3.8B	INT4 Q8 FP16	177 tok/s 127 tok/s 78 tok/s	~364 ms ~366 ms ~371 ms
#218	Qwen2.5-7B	7B	INT4 Q8	125 tok/s 83 tok/s	~646 ms ~650 ms
#220	Ministral-8B-2410	8B	INT4 Q8	115 tok/s 71 tok/s	~734 ms ~802 ms
#221	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	254 tok/s 208 tok/s 148 tok/s	~157 ms ~158 ms ~160 ms
#222	Llama 3.2 3B	3B	INT4 Q8 FP16	198 tok/s 147 tok/s 94 tok/s	~293 ms ~295 ms ~299 ms
#223	Phi-3-mini	3.8B	INT4 Q8 FP16	177 tok/s 127 tok/s 78 tok/s	~364 ms ~366 ms ~371 ms
#224	Gemma 3n E2B IT	6B (2B active)	INT4 Q8	145 tok/s 128 tok/s	~221 ms ~222 ms
#225	Gemma 2 2B	2B	INT4 Q8 FP16	231 tok/s 182 tok/s 124 tok/s	~205 ms ~207 ms ~209 ms
#226	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8	122 tok/s 80 tok/s	~672 ms ~676 ms
#227	Llama 3.2 1B	1B	INT4 Q8 FP16	282 tok/s 243 tok/s 185 tok/s	~109 ms ~110 ms ~111 ms
#228	Phi-3-medium	14B	INT4	76 tok/s	~1.3s
#230	OLMo 3 7B Instruct	7B	INT4 Q8	94 tok/s 66 tok/s	~649 ms ~708 ms
#231	Qwen2-7B	7B	INT4 Q8	125 tok/s 83 tok/s	~646 ms ~650 ms
#232	Phi-3-small	7B	INT4 Q8	125 tok/s 83 tok/s	~646 ms ~650 ms
#233	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8	122 tok/s 80 tok/s	~672 ms ~676 ms
#234	Gemma 3 1B	1B	INT4 Q8 FP16	282 tok/s 243 tok/s 185 tok/s	~109 ms ~110 ms ~111 ms
#236	Gemma 1 7B	7B	INT4 Q8	94 tok/s 66 tok/s	~649 ms ~708 ms
#237	Gemma 1 2B	2B	INT4 Q8 FP16	240 tok/s 188 tok/s 126 tok/s	~205 ms ~206 ms ~209 ms
#238	ChatGLM3-6B	6B	INT4 Q8	103 tok/s 77 tok/s	~560 ms ~563 ms
#239	ChatGLM2-6B	6B	INT4 Q8	103 tok/s 77 tok/s	~560 ms ~563 ms
#240	ChatGLM-6B	6B	INT4 Q8	103 tok/s 77 tok/s	~560 ms ~563 ms
Unranked Models (Sorted by Speed)
-	DINOv3 ViT-L/16	0.3B	INT4 Q8 FP16	340 tok/s 321 tok/s 286 tok/s	~36 ms ~37 ms ~37 ms
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	322 tok/s 295 tok/s 248 tok/s	~56 ms ~57 ms ~57 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	322 tok/s 295 tok/s 248 tok/s	~56 ms ~57 ms ~57 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	319 tok/s 304 tok/s 272 tok/s	~37 ms ~37 ms ~37 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	319 tok/s 304 tok/s 272 tok/s	~37 ms ~37 ms ~37 ms
-	Liquid D1-Omni 600M	0.6B	INT4 Q8 FP16	314 tok/s 283 tok/s 232 tok/s	~66 ms ~67 ms ~68 ms
-	Falcon-1B	1B	INT4 Q8 FP16	291 tok/s 249 tok/s 188 tok/s	~109 ms ~110 ms ~111 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	282 tok/s 243 tok/s 185 tok/s	~109 ms ~110 ms ~111 ms
-	MiniCPM-V 4.7 1B	1B	INT4 Q8 FP16	282 tok/s 243 tok/s 185 tok/s	~109 ms ~110 ms ~111 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	254 tok/s 208 tok/s 148 tok/s	~157 ms ~158 ms ~160 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	254 tok/s 208 tok/s 148 tok/s	~157 ms ~158 ms ~160 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	244 tok/s 197 tok/s 137 tok/s	~177 ms ~178 ms ~180 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	240 tok/s 192 tok/s 133 tok/s	~186 ms ~187 ms ~189 ms
-	Falcon-3B	3B	INT4 Q8 FP16	206 tok/s 151 tok/s 95 tok/s	~293 ms ~294 ms ~298 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	203 tok/s 179 tok/s 138 tok/s	~139 ms ~139 ms ~141 ms
-	Phi-1	1.3B	INT4 Q8 FP16	203 tok/s 179 tok/s 138 tok/s	~139 ms ~139 ms ~141 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	203 tok/s 179 tok/s 138 tok/s	~139 ms ~139 ms ~141 ms
-	Mellum 2.1 12B A2.5B Thinking	12B (2.5B active)	INT4	200 tok/s	~286 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	199 tok/s 148 tok/s 94 tok/s	~293 ms ~295 ms ~298 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	198 tok/s 147 tok/s 94 tok/s	~293 ms ~295 ms ~299 ms
-	LFM2-VL 3B	3B	INT4 Q8 FP16	198 tok/s 147 tok/s 94 tok/s	~293 ms ~295 ms ~299 ms
-	Liquid D1 3B	3B	INT4 Q8 FP16	198 tok/s 147 tok/s 94 tok/s	~293 ms ~295 ms ~299 ms
-	Liquid D1-VL 3B	3B	INT4 Q8 FP16	198 tok/s 147 tok/s 94 tok/s	~293 ms ~295 ms ~299 ms
-	Liquid LFM2-VL 3B D1	3B	INT4 Q8 FP16	198 tok/s 147 tok/s 94 tok/s	~293 ms ~295 ms ~299 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	198 tok/s 147 tok/s 94 tok/s	~293 ms ~295 ms ~299 ms
-	FrogNano 4B	4B	INT4 Q8 FP16	173 tok/s 123 tok/s 75 tok/s	~381 ms ~384 ms ~389 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	173 tok/s 123 tok/s 75 tok/s	~381 ms ~384 ms ~389 ms
-	EmbeddingGemma 2	2B	INT4 Q8 FP16	167 tok/s 143 tok/s 104 tok/s	~207 ms ~208 ms ~211 ms
-	Phi-2	2.7B	INT4 Q8 FP16	151 tok/s 124 tok/s 87 tok/s	~268 ms ~270 ms ~273 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	145 tok/s 118 tok/s 81 tok/s	~295 ms ~296 ms ~300 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	145 tok/s 118 tok/s 81 tok/s	~295 ms ~296 ms ~300 ms
-	Falcon-7B	7B	INT4 Q8	130 tok/s 85 tok/s	~646 ms ~650 ms
-	Falcon3-7B	7B	INT4 Q8	125 tok/s 83 tok/s	~646 ms ~650 ms
-	Hy-MT2 7B	7B	INT4 Q8	125 tok/s 83 tok/s	~646 ms ~650 ms
-	Mistral-7B-v0.1	7.3B	INT4 Q8	122 tok/s 80 tok/s	~672 ms ~676 ms
-	Youtu Parsing Omni	8B	INT4 Q8	117 tok/s 76 tok/s	~734 ms ~739 ms
-	LensVLM 9B	9B	INT4 Q8	106 tok/s 65 tok/s	~824 ms ~900 ms
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8	106 tok/s 65 tok/s	~824 ms ~900 ms
-	ChatGLM3-6B-32K	6B	INT4 Q8	103 tok/s 77 tok/s	~560 ms ~563 ms
-	Yi-6B	6B	INT4 Q8	103 tok/s 77 tok/s	~560 ms ~563 ms
-	Falcon3-10B	10B	INT4 Q8	98 tok/s 56 tok/s	~912 ms ~1.1s
-	Falcon2-11B	11B	INT4 Q8	95 tok/s 53 tok/s	~1.0s ~1.2s
-	Hunyuan Lite	7B	INT4 Q8	94 tok/s 66 tok/s	~649 ms ~708 ms
-	OLMo 3 7B Base	7B	INT4 Q8	94 tok/s 66 tok/s	~649 ms ~708 ms
-	SEA-LION-7B	7.1B	INT4 Q8	93 tok/s 65 tok/s	~657 ms ~717 ms
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8	93 tok/s 65 tok/s	~657 ms ~717 ms
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8	87 tok/s 56 tok/s	~737 ms ~870 ms
-	Typhoon-2-8B	8B	INT4 Q8	87 tok/s 56 tok/s	~737 ms ~870 ms
-	GLM-4-9B	9B	INT4	80 tok/s	~827 ms
-	GLM-4-9B-Chat	9B	INT4	80 tok/s	~827 ms
-	GLM-4-9B-Chat-1M	9B	INT4	80 tok/s	~827 ms
-	Yi-9B	9B	INT4	80 tok/s	~827 ms
-	Sahabat-AI-Gemma2-9B	9.2B	INT4	78 tok/s	~844 ms
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4	78 tok/s	~844 ms
-	GreenMind-14B-R1	14B	INT4	52 tok/s	~1.5s

NVIDIA RTX 3080 (12GB)

VRAM Scaling by Context Length

Best Models to Run on NVIDIA RTX 3080 (12GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions