What are the best models to run on NVIDIA RTX 3080 (10GB)?

Based on model rankings and 10 GB usable VRAM: Top Overall: Qwen3.5-9B (INT4) ~93 tok/s. Top Coding: K2 Horizon 7B (INT4) ~111 tok/s. Optimal Daily Driver: GLM-4V (INT4) ~65 tok/s. Maximum Model Size: DeepSeek-R1 14B (INT4) ~64 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 (10GB)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#121	Qwen3.5-9B	9B	INT4	93 tok/s	~849 ms
#127	K2 Horizon 7B	7B	INT4 Q8	111 tok/s 68 tok/s	~666 ms ~728 ms
#136	Qwen3.5-4B	4B	INT4 Q8 FP16	156 tok/s 109 tok/s 62 tok/s	~393 ms ~395 ms ~435 ms
#140	GLM-4V	9B	INT4	65 tok/s	~925 ms
#145	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	33 tok/s 19 tok/s 9 tok/s	~2.7s ~2.7s ~3.2s
#152	Granite 4.2 8B	8B	INT4 Q8	101 tok/s 58 tok/s	~756 ms ~894 ms
#157	Gemma 4 E4B	8B	INT4 Q8	101 tok/s 58 tok/s	~756 ms ~894 ms
#164	Phi-4	14B	INT4	62 tok/s	~1.4s
#176	DeepSeek-R1 14B	14B	INT4	64 tok/s	~1.4s
#180	Gemma 3 12B	12B	INT4	74 tok/s	~1.1s
#185	Ministral 3 8B	8B	INT4	75 tok/s	~760 ms
#187	OLMo 3 7B Think	7B	INT4 Q8	82 tok/s 53 tok/s	~669 ms ~790 ms
#188	Qwen3.5-2B	2B	INT4 Q8 FP16	214 tok/s 165 tok/s 109 tok/s	~212 ms ~213 ms ~216 ms
#189	MiniCPM5-2B	2B	INT4 Q8 FP16	214 tok/s 165 tok/s 109 tok/s	~212 ms ~213 ms ~216 ms
#190	Qwen2.5-14B	14B	INT4	62 tok/s	~1.4s
#193	Gemma 4 E2B	5.1B	INT4 Q8	136 tok/s 91 tok/s	~493 ms ~496 ms
#195	Ministral 3 3B	3B	INT4 Q8 FP16	130 tok/s 104 tok/s 67 tok/s	~304 ms ~306 ms ~336 ms
#196	DeepSeek-R1 8B	8B	INT4 Q8	103 tok/s 62 tok/s	~756 ms ~827 ms
#198	Gemma 2 9B	9B	INT4	93 tok/s	~849 ms
#201	Qwen3-4B	4B	INT4 Q8 FP16	156 tok/s 109 tok/s 62 tok/s	~393 ms ~395 ms ~435 ms
#202	Qwen3-14B	14B	INT4	62 tok/s	~1.4s
#203	Gemma 3 4B	4B	INT4 Q8 FP16	156 tok/s 109 tok/s 62 tok/s	~393 ms ~395 ms ~435 ms
#204	DeepSeek-R1 7B	7B	INT4 Q8	113 tok/s 69 tok/s	~665 ms ~727 ms
#206	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	286 tok/s 247 tok/s 189 tok/s	~90 ms ~90 ms ~92 ms
#207	Llama 3.1 8B	8B	INT4 Q8	101 tok/s 58 tok/s	~756 ms ~894 ms
#207	Qwen3-8B	8B	INT4 Q8	101 tok/s 58 tok/s	~756 ms ~894 ms
#212	Llama 3 8B	8B	INT4 Q8	101 tok/s 58 tok/s	~756 ms ~894 ms
#213	Qwen3-1.7B	1.7B	INT4 Q8 FP16	228 tok/s 180 tok/s 122 tok/s	~182 ms ~183 ms ~186 ms
#214	Gemma 3 270M	0.27B	INT4 Q8 FP16	317 tok/s 300 tok/s 267 tok/s	~34 ms ~35 ms ~35 ms
#216	Qwen3-0.6B	0.6B	INT4 Q8 FP16	305 tok/s 270 tok/s 216 tok/s	~68 ms ~69 ms ~70 ms
#217	Phi-4-Mini	3.8B	INT4 Q8 FP16	160 tok/s 112 tok/s 64 tok/s	~374 ms ~377 ms ~415 ms
#218	Qwen2.5-7B	7B	INT4 Q8	111 tok/s 68 tok/s	~666 ms ~728 ms
#220	Ministral-8B-2410	8B	INT4 Q8	101 tok/s 58 tok/s	~756 ms ~894 ms
#221	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	239 tok/s 191 tok/s 133 tok/s	~162 ms ~163 ms ~165 ms
#222	Llama 3.2 3B	3B	INT4 Q8 FP16	181 tok/s 131 tok/s 82 tok/s	~302 ms ~304 ms ~308 ms
#223	Phi-3-mini	3.8B	INT4 Q8 FP16	160 tok/s 112 tok/s 64 tok/s	~374 ms ~377 ms ~415 ms
#224	Gemma 3n E2B IT	6B (2B active)	INT4 Q8	130 tok/s 107 tok/s	~228 ms ~248 ms
#225	Gemma 2 2B	2B	INT4 Q8 FP16	214 tok/s 165 tok/s 109 tok/s	~212 ms ~213 ms ~216 ms
#226	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8	108 tok/s 66 tok/s	~692 ms ~757 ms
#227	Llama 3.2 1B	1B	INT4 Q8 FP16	270 tok/s 227 tok/s 168 tok/s	~112 ms ~113 ms ~115 ms
#228	Phi-3-medium	14B	INT4	62 tok/s	~1.4s
#230	OLMo 3 7B Instruct	7B	INT4 Q8	82 tok/s 53 tok/s	~669 ms ~790 ms
#231	Qwen2-7B	7B	INT4 Q8	111 tok/s 68 tok/s	~666 ms ~728 ms
#232	Phi-3-small	7B	INT4 Q8	111 tok/s 68 tok/s	~666 ms ~728 ms
#233	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8	108 tok/s 66 tok/s	~692 ms ~757 ms
#234	Gemma 3 1B	1B	INT4 Q8 FP16	270 tok/s 227 tok/s 168 tok/s	~112 ms ~113 ms ~115 ms
#236	Gemma 1 7B	7B	INT4 Q8	82 tok/s 53 tok/s	~669 ms ~790 ms
#237	Gemma 1 2B	2B	INT4 Q8 FP16	225 tok/s 171 tok/s 112 tok/s	~211 ms ~213 ms ~216 ms
#238	ChatGLM3-6B	6B	INT4 Q8	90 tok/s 63 tok/s	~578 ms ~631 ms
#239	ChatGLM2-6B	6B	INT4 Q8	90 tok/s 63 tok/s	~578 ms ~631 ms
#240	ChatGLM-6B	6B	INT4 Q8	90 tok/s 63 tok/s	~578 ms ~631 ms
Unranked Models (Sorted by Speed)
-	DINOv3 ViT-L/16	0.3B	INT4 Q8 FP16	335 tok/s 313 tok/s 273 tok/s	~37 ms ~38 ms ~38 ms
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	314 tok/s 283 tok/s 232 tok/s	~58 ms ~58 ms ~59 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	314 tok/s 283 tok/s 232 tok/s	~58 ms ~58 ms ~59 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	311 tok/s 293 tok/s 258 tok/s	~38 ms ~38 ms ~38 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	311 tok/s 293 tok/s 258 tok/s	~38 ms ~38 ms ~38 ms
-	Liquid D1-Omni 600M	0.6B	INT4 Q8 FP16	305 tok/s 270 tok/s 216 tok/s	~68 ms ~69 ms ~70 ms
-	Falcon-1B	1B	INT4 Q8 FP16	279 tok/s 233 tok/s 171 tok/s	~112 ms ~113 ms ~115 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	270 tok/s 227 tok/s 168 tok/s	~112 ms ~113 ms ~115 ms
-	MiniCPM-V 4.7 1B	1B	INT4 Q8 FP16	270 tok/s 227 tok/s 168 tok/s	~112 ms ~113 ms ~115 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	239 tok/s 191 tok/s 133 tok/s	~162 ms ~163 ms ~165 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	239 tok/s 191 tok/s 133 tok/s	~162 ms ~163 ms ~165 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	228 tok/s 180 tok/s 122 tok/s	~182 ms ~183 ms ~186 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	224 tok/s 175 tok/s 118 tok/s	~191 ms ~192 ms ~195 ms
-	Falcon-3B	3B	INT4 Q8 FP16	189 tok/s 135 tok/s 83 tok/s	~301 ms ~304 ms ~308 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	186 tok/s 162 tok/s 122 tok/s	~143 ms ~144 ms ~146 ms
-	Phi-1	1.3B	INT4 Q8 FP16	186 tok/s 162 tok/s 122 tok/s	~143 ms ~144 ms ~146 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	186 tok/s 162 tok/s 122 tok/s	~143 ms ~144 ms ~146 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	183 tok/s 132 tok/s 82 tok/s	~302 ms ~304 ms ~308 ms
-	Mellum 2.1 12B A2.5B Thinking	12B (2.5B active)	INT4	183 tok/s	~295 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	181 tok/s 131 tok/s 82 tok/s	~302 ms ~304 ms ~308 ms
-	LFM2-VL 3B	3B	INT4 Q8 FP16	181 tok/s 131 tok/s 82 tok/s	~302 ms ~304 ms ~308 ms
-	Liquid D1 3B	3B	INT4 Q8 FP16	181 tok/s 131 tok/s 82 tok/s	~302 ms ~304 ms ~308 ms
-	Liquid D1-VL 3B	3B	INT4 Q8 FP16	181 tok/s 131 tok/s 82 tok/s	~302 ms ~304 ms ~308 ms
-	Liquid LFM2-VL 3B D1	3B	INT4 Q8 FP16	181 tok/s 131 tok/s 82 tok/s	~302 ms ~304 ms ~308 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	181 tok/s 131 tok/s 82 tok/s	~302 ms ~304 ms ~308 ms
-	FrogNano 4B	4B	INT4 Q8 FP16	156 tok/s 109 tok/s 62 tok/s	~393 ms ~395 ms ~435 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	156 tok/s 109 tok/s 62 tok/s	~393 ms ~395 ms ~435 ms
-	EmbeddingGemma 2	2B	INT4 Q8 FP16	151 tok/s 127 tok/s 91 tok/s	~213 ms ~215 ms ~218 ms
-	Phi-2	2.7B	INT4 Q8 FP16	135 tok/s 110 tok/s 75 tok/s	~277 ms ~278 ms ~283 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	130 tok/s 104 tok/s 67 tok/s	~304 ms ~306 ms ~336 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	130 tok/s 104 tok/s 67 tok/s	~304 ms ~306 ms ~336 ms
-	Falcon-7B	7B	INT4 Q8	115 tok/s 70 tok/s	~665 ms ~727 ms
-	Falcon3-7B	7B	INT4 Q8	111 tok/s 68 tok/s	~666 ms ~728 ms
-	Hy-MT2 7B	7B	INT4 Q8	111 tok/s 68 tok/s	~666 ms ~728 ms
-	Mistral-7B-v0.1	7.3B	INT4 Q8	108 tok/s 66 tok/s	~692 ms ~757 ms
-	Youtu Parsing Omni	8B	INT4 Q8	103 tok/s 62 tok/s	~756 ms ~827 ms
-	LensVLM 9B	9B	INT4	93 tok/s	~849 ms
-	MiMo V2.6 Distill Qwen 9B	9B	INT4	93 tok/s	~849 ms
-	ChatGLM3-6B-32K	6B	INT4 Q8	90 tok/s 63 tok/s	~578 ms ~631 ms
-	Yi-6B	6B	INT4 Q8	90 tok/s 63 tok/s	~578 ms ~631 ms
-	Falcon3-10B	10B	INT4	86 tok/s	~940 ms
-	Falcon2-11B	11B	INT4	83 tok/s	~1.0s
-	Hunyuan Lite	7B	INT4 Q8	82 tok/s 53 tok/s	~669 ms ~790 ms
-	OLMo 3 7B Base	7B	INT4 Q8	82 tok/s 53 tok/s	~669 ms ~790 ms
-	SEA-LION-7B	7.1B	INT4 Q8	81 tok/s 53 tok/s	~678 ms ~800 ms
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8	81 tok/s 53 tok/s	~678 ms ~800 ms
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4	75 tok/s	~760 ms
-	Typhoon-2-8B	8B	INT4	75 tok/s	~760 ms
-	GLM-4-9B	9B	INT4	65 tok/s	~925 ms
-	GLM-4-9B-Chat	9B	INT4	65 tok/s	~925 ms
-	GLM-4-9B-Chat-1M	9B	INT4	65 tok/s	~925 ms
-	Yi-9B	9B	INT4	65 tok/s	~925 ms
-	Sahabat-AI-Gemma2-9B	9.2B	INT4	64 tok/s	~945 ms
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4	64 tok/s	~945 ms

NVIDIA RTX 3080 (10GB)

VRAM Scaling by Context Length

Best Models to Run on NVIDIA RTX 3080 (10GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions