What are the best models to run on NVIDIA RTX 4070 SUPER (12GB)?

Based on model rankings and 12 GB usable VRAM: Top Overall: Phi-4 Reasoning Plus (INT4) ~31 tok/s. Top Coding: K2 Horizon 7B (INT4) ~82 tok/s. Optimal Daily Driver: Qwen3.5-9B (INT4) ~67 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 4070 SUPER (12GB)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#102	Phi-4 Reasoning Plus	14B	INT4	31 tok/s	~1.3s
#121	Qwen3.5-9B	9B	INT4 Q8	67 tok/s 39 tok/s	~720 ms ~791 ms
#127	K2 Horizon 7B	7B	INT4 Q8	82 tok/s 51 tok/s	~565 ms ~572 ms
#130	Gemma 4 12B	11.95B	INT4	37 tok/s	~1.0s
#136	Qwen3.5-4B	4B	INT4 Q8 FP16	121 tok/s 80 tok/s 46 tok/s	~334 ms ~338 ms ~347 ms
#140	GLM-4V	9B	INT4	49 tok/s	~725 ms
#145	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	22 tok/s 13 tok/s 6 tok/s	~2.3s ~2.3s ~2.6s
#152	Granite 4.2 8B	8B	INT4 Q8	74 tok/s 43 tok/s	~642 ms ~705 ms
#157	Gemma 4 E4B	8B	INT4 Q8	74 tok/s 43 tok/s	~642 ms ~705 ms
#164	Phi-4	14B	INT4	46 tok/s	~1.1s
#176	DeepSeek-R1 14B	14B	INT4	48 tok/s	~1.1s
#177	Ministral 3 14B	14B	INT4	31 tok/s	~1.3s
#180	Gemma 3 12B	12B	INT4	53 tok/s	~953 ms
#185	Ministral 3 8B	8B	INT4 Q8	53 tok/s 34 tok/s	~647 ms ~767 ms
#187	OLMo 3 7B Think	7B	INT4 Q8	59 tok/s 40 tok/s	~569 ms ~625 ms
#188	Qwen3.5-2B	2B	INT4 Q8 FP16	177 tok/s 129 tok/s 80 tok/s	~180 ms ~182 ms ~187 ms
#189	MiniCPM5-2B	2B	INT4 Q8 FP16	177 tok/s 129 tok/s 80 tok/s	~180 ms ~182 ms ~187 ms
#190	Qwen2.5-14B	14B	INT4	46 tok/s	~1.1s
#193	Gemma 4 E2B	5.1B	INT4 Q8 FP16	103 tok/s 66 tok/s 33 tok/s	~418 ms ~424 ms ~509 ms
#195	Ministral 3 3B	3B	INT4 Q8 FP16	98 tok/s 76 tok/s 50 tok/s	~260 ms ~263 ms ~270 ms
#196	DeepSeek-R1 8B	8B	INT4 Q8	76 tok/s 46 tok/s	~641 ms ~650 ms
#198	Gemma 2 9B	9B	INT4 Q8	67 tok/s 39 tok/s	~720 ms ~791 ms
#201	Qwen3-4B	4B	INT4 Q8 FP16	121 tok/s 80 tok/s 46 tok/s	~334 ms ~338 ms ~347 ms
#202	Qwen3-14B	14B	INT4	46 tok/s	~1.1s
#203	Gemma 3 4B	4B	INT4 Q8 FP16	121 tok/s 80 tok/s 46 tok/s	~334 ms ~338 ms ~347 ms
#204	DeepSeek-R1 7B	7B	INT4 Q8	84 tok/s 52 tok/s	~564 ms ~572 ms
#206	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	257 tok/s 211 tok/s 152 tok/s	~77 ms ~78 ms ~80 ms
#207	Llama 3.1 8B	8B	INT4 Q8	74 tok/s 43 tok/s	~642 ms ~705 ms
#207	Qwen3-8B	8B	INT4 Q8	74 tok/s 43 tok/s	~642 ms ~705 ms
#212	Llama 3 8B	8B	INT4 Q8	74 tok/s 43 tok/s	~642 ms ~705 ms
#213	Qwen3-1.7B	1.7B	INT4 Q8 FP16	191 tok/s 143 tok/s 91 tok/s	~155 ms ~157 ms ~161 ms
#214	Gemma 3 270M	0.27B	INT4 Q8 FP16	295 tok/s 274 tok/s 234 tok/s	~30 ms ~31 ms ~31 ms
#216	Qwen3-0.6B	0.6B	INT4 Q8 FP16	279 tok/s 238 tok/s 178 tok/s	~59 ms ~60 ms ~61 ms
#217	Phi-4-Mini	3.8B	INT4 Q8 FP16	125 tok/s 83 tok/s 48 tok/s	~318 ms ~322 ms ~331 ms
#218	Qwen2.5-7B	7B	INT4 Q8	82 tok/s 51 tok/s	~565 ms ~572 ms
#220	Ministral-8B-2410	8B	INT4 Q8	74 tok/s 43 tok/s	~642 ms ~705 ms
#221	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	203 tok/s 154 tok/s 100 tok/s	~138 ms ~140 ms ~143 ms
#222	Llama 3.2 3B	3B	INT4 Q8 FP16	143 tok/s 99 tok/s 58 tok/s	~257 ms ~260 ms ~267 ms
#223	Phi-3-mini	3.8B	INT4 Q8 FP16	125 tok/s 83 tok/s 48 tok/s	~318 ms ~322 ms ~331 ms
#224	Gemma 3n E2B IT	6B (2B active)	INT4 Q8	97 tok/s 83 tok/s	~196 ms ~198 ms
#225	Gemma 2 2B	2B	INT4 Q8 FP16	177 tok/s 129 tok/s 80 tok/s	~180 ms ~182 ms ~187 ms
#226	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8	79 tok/s 49 tok/s	~587 ms ~595 ms
#227	Llama 3.2 1B	1B	INT4 Q8 FP16	237 tok/s 190 tok/s 132 tok/s	~96 ms ~97 ms ~100 ms
#228	Phi-3-medium	14B	INT4	46 tok/s	~1.1s
#230	OLMo 3 7B Instruct	7B	INT4 Q8	59 tok/s 40 tok/s	~569 ms ~625 ms
#231	Qwen2-7B	7B	INT4 Q8	82 tok/s 51 tok/s	~565 ms ~572 ms
#232	Phi-3-small	7B	INT4 Q8	82 tok/s 51 tok/s	~565 ms ~572 ms
#233	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8	79 tok/s 49 tok/s	~587 ms ~595 ms
#234	Gemma 3 1B	1B	INT4 Q8 FP16	237 tok/s 190 tok/s 132 tok/s	~96 ms ~97 ms ~100 ms
#236	Gemma 1 7B	7B	INT4 Q8	59 tok/s 40 tok/s	~569 ms ~625 ms
#237	Gemma 1 2B	2B	INT4 Q8 FP16	187 tok/s 134 tok/s 82 tok/s	~180 ms ~182 ms ~187 ms
#238	ChatGLM3-6B	6B	INT4 Q8	65 tok/s 47 tok/s	~492 ms ~498 ms
#239	ChatGLM2-6B	6B	INT4 Q8	65 tok/s 47 tok/s	~492 ms ~498 ms
#240	ChatGLM-6B	6B	INT4 Q8	65 tok/s 47 tok/s	~492 ms ~498 ms
Unranked Models (Sorted by Speed)
-	DINOv3 ViT-L/16	0.3B	INT4 Q8 FP16	319 tok/s 290 tok/s 241 tok/s	~33 ms ~33 ms ~34 ms
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	292 tok/s 253 tok/s 195 tok/s	~50 ms ~51 ms ~52 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	292 tok/s 253 tok/s 195 tok/s	~50 ms ~51 ms ~52 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	287 tok/s 265 tok/s 224 tok/s	~33 ms ~33 ms ~34 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	287 tok/s 265 tok/s 224 tok/s	~33 ms ~33 ms ~34 ms
-	Liquid D1-Omni 600M	0.6B	INT4 Q8 FP16	279 tok/s 238 tok/s 178 tok/s	~59 ms ~60 ms ~61 ms
-	Falcon-1B	1B	INT4 Q8 FP16	248 tok/s 196 tok/s 135 tok/s	~96 ms ~97 ms ~100 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	237 tok/s 190 tok/s 132 tok/s	~96 ms ~97 ms ~100 ms
-	MiniCPM-V 4.7 1B	1B	INT4 Q8 FP16	237 tok/s 190 tok/s 132 tok/s	~96 ms ~97 ms ~100 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	203 tok/s 154 tok/s 100 tok/s	~138 ms ~140 ms ~143 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	203 tok/s 154 tok/s 100 tok/s	~138 ms ~140 ms ~143 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	191 tok/s 143 tok/s 91 tok/s	~155 ms ~157 ms ~161 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	186 tok/s 138 tok/s 87 tok/s	~163 ms ~165 ms ~169 ms
-	Falcon-3B	3B	INT4 Q8 FP16	151 tok/s 102 tok/s 60 tok/s	~256 ms ~259 ms ~266 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	149 tok/s 126 tok/s 91 tok/s	~123 ms ~124 ms ~127 ms
-	Phi-1	1.3B	INT4 Q8 FP16	149 tok/s 126 tok/s 91 tok/s	~123 ms ~124 ms ~127 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	149 tok/s 126 tok/s 91 tok/s	~123 ms ~124 ms ~127 ms
-	Mellum 2.1 12B A2.5B Thinking	12B (2.5B active)	INT4	146 tok/s	~251 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	145 tok/s 99 tok/s 59 tok/s	~257 ms ~260 ms ~267 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	143 tok/s 99 tok/s 58 tok/s	~257 ms ~260 ms ~267 ms
-	LFM2-VL 3B	3B	INT4 Q8 FP16	143 tok/s 99 tok/s 58 tok/s	~257 ms ~260 ms ~267 ms
-	Liquid D1 3B	3B	INT4 Q8 FP16	143 tok/s 99 tok/s 58 tok/s	~257 ms ~260 ms ~267 ms
-	Liquid D1-VL 3B	3B	INT4 Q8 FP16	143 tok/s 99 tok/s 58 tok/s	~257 ms ~260 ms ~267 ms
-	Liquid LFM2-VL 3B D1	3B	INT4 Q8 FP16	143 tok/s 99 tok/s 58 tok/s	~257 ms ~260 ms ~267 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	143 tok/s 99 tok/s 58 tok/s	~257 ms ~260 ms ~267 ms
-	FrogNano 4B	4B	INT4 Q8 FP16	121 tok/s 80 tok/s 46 tok/s	~334 ms ~338 ms ~347 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	121 tok/s 80 tok/s 46 tok/s	~334 ms ~338 ms ~347 ms
-	EmbeddingGemma 2	2B	INT4 Q8 FP16	116 tok/s 95 tok/s 66 tok/s	~183 ms ~185 ms ~190 ms
-	Phi-2	2.7B	INT4 Q8 FP16	102 tok/s 81 tok/s 54 tok/s	~237 ms ~239 ms ~246 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	98 tok/s 76 tok/s 50 tok/s	~260 ms ~263 ms ~270 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	98 tok/s 76 tok/s 50 tok/s	~260 ms ~263 ms ~270 ms
-	Falcon-7B	7B	INT4 Q8	85 tok/s 52 tok/s	~564 ms ~572 ms
-	Falcon3-7B	7B	INT4 Q8	82 tok/s 51 tok/s	~565 ms ~572 ms
-	Hy-MT2 7B	7B	INT4 Q8	82 tok/s 51 tok/s	~565 ms ~572 ms
-	Mistral-7B-v0.1	7.3B	INT4 Q8	79 tok/s 49 tok/s	~587 ms ~595 ms
-	Youtu Parsing Omni	8B	INT4 Q8	76 tok/s 46 tok/s	~641 ms ~650 ms
-	LensVLM 9B	9B	INT4 Q8	67 tok/s 39 tok/s	~720 ms ~791 ms
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8	67 tok/s 39 tok/s	~720 ms ~791 ms
-	ChatGLM3-6B-32K	6B	INT4 Q8	65 tok/s 47 tok/s	~492 ms ~498 ms
-	Yi-6B	6B	INT4 Q8	65 tok/s 47 tok/s	~492 ms ~498 ms
-	Falcon3-10B	10B	INT4 Q8	62 tok/s 34 tok/s	~797 ms ~946 ms
-	Falcon2-11B	11B	INT4 Q8	59 tok/s 32 tok/s	~873 ms ~1.0s
-	Hunyuan Lite	7B	INT4 Q8	59 tok/s 40 tok/s	~569 ms ~625 ms
-	OLMo 3 7B Base	7B	INT4 Q8	59 tok/s 40 tok/s	~569 ms ~625 ms
-	SEA-LION-7B	7.1B	INT4 Q8	58 tok/s 39 tok/s	~577 ms ~633 ms
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8	58 tok/s 39 tok/s	~577 ms ~633 ms
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8	53 tok/s 34 tok/s	~647 ms ~767 ms
-	Typhoon-2-8B	8B	INT4 Q8	53 tok/s 34 tok/s	~647 ms ~767 ms
-	GLM-4-9B	9B	INT4	49 tok/s	~725 ms
-	GLM-4-9B-Chat	9B	INT4	49 tok/s	~725 ms
-	GLM-4-9B-Chat-1M	9B	INT4	49 tok/s	~725 ms
-	Yi-9B	9B	INT4	49 tok/s	~725 ms
-	Sahabat-AI-Gemma2-9B	9.2B	INT4	48 tok/s	~741 ms
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4	48 tok/s	~741 ms
-	GreenMind-14B-R1	14B	INT4	31 tok/s	~1.3s

NVIDIA RTX 4070 SUPER (12GB)

VRAM Scaling by Context Length

Best Models to Run on NVIDIA RTX 4070 SUPER (12GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions