What are the best models to run on NVIDIA RTX 4070 Ti SUPER (16GB)?

Based on model rankings and 16 GB usable VRAM: Top Overall: Qwen 3.8 27B (INT4) ~30 tok/s. Top Coding: Qwen3.5-27B (INT4) ~30 tok/s. Optimal Daily Driver: Phi-4 Reasoning Plus (INT4) ~44 tok/s. Maximum Model Size: ERNIE-4.5-VL-28B-A3B (INT4) ~132 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 4070 Ti SUPER (16GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4	30 tok/s	~2.0s
#83	Qwen3.5-27B	27B	INT4	30 tok/s	~2.0s
#100	Phi-4 Reasoning Plus	14B	INT4	44 tok/s	~899 ms
#101	MiMo V2 Flash	15B (309B active)	INT4	3 tok/s	~20.0s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4	126 tok/s	~334 ms
#109	Gemma 4 12B	11.95B	INT4 Q8	50 tok/s 31 tok/s	~771 ms ~913 ms
#113	Qwen3.5-9B	9B	INT4 Q8	84 tok/s 53 tok/s	~581 ms ~588 ms
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	101 tok/s 65 tok/s 33 tok/s	~456 ms ~461 ms ~554 ms
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	29 tok/s 17 tok/s 9 tok/s	~1.8s ~1.8s ~1.9s
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	145 tok/s 99 tok/s 59 tok/s	~270 ms ~273 ms ~280 ms
#136	GPT-OSS 20B	21B (3.6B active)	INT4	61 tok/s	~346 ms
#146	Granite 4.2 8B	8B	INT4 Q8	92 tok/s 58 tok/s	~518 ms ~524 ms
#150	Gemma 4 E4B	8B	INT4 Q8	92 tok/s 58 tok/s	~518 ms ~524 ms
#158	GLM-4V	9B	INT4 Q8	62 tok/s 41 tok/s	~585 ms ~642 ms
#159	Gemma 3 27B	27B	INT4	30 tok/s	~2.0s
#160	Phi-4	14B	INT4 Q8	59 tok/s 32 tok/s	~893 ms ~1.1s
#169	OLMo 3 7B Think	7B	INT4 Q8	74 tok/s 53 tok/s	~459 ms ~465 ms
#172	DeepSeek-R1 14B	14B	INT4 Q8	61 tok/s 33 tok/s	~892 ms ~1.1s
#173	Ministral 3 14B	14B	INT4	44 tok/s	~899 ms
#177	Gemma 3 12B	12B	INT4 Q8	67 tok/s 39 tok/s	~768 ms ~844 ms
#181	Mistral-Small-2501	24B	INT4	36 tok/s	~1.6s
#182	Ministral 3 8B	8B	INT4 Q8	68 tok/s 48 tok/s	~522 ms ~528 ms
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	203 tok/s 154 tok/s 100 tok/s	~146 ms ~148 ms ~152 ms
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	203 tok/s 154 tok/s 100 tok/s	~146 ms ~148 ms ~152 ms
#185	Qwen2.5-14B	14B	INT4 Q8	59 tok/s 32 tok/s	~893 ms ~1.1s
#186	Gemma 2 27B	27B	INT4	30 tok/s	~2.0s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	125 tok/s 83 tok/s 48 tok/s	~338 ms ~342 ms ~351 ms
#189	Qwen3-4B	4B	INT4 Q8 FP16	145 tok/s 99 tok/s 59 tok/s	~270 ms ~273 ms ~280 ms
#190	DeepSeek-R1 8B	8B	INT4 Q8	94 tok/s 59 tok/s	~517 ms ~524 ms
#193	Ministral 3 3B	3B	INT4 Q8 FP16	120 tok/s 95 tok/s 63 tok/s	~210 ms ~213 ms ~218 ms
#195	Qwen3-14B	14B	INT4 Q8	59 tok/s 32 tok/s	~893 ms ~1.1s
#197	Gemma 2 9B	9B	INT4 Q8	84 tok/s 53 tok/s	~581 ms ~588 ms
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	101 tok/s 65 tok/s 33 tok/s	~456 ms ~461 ms ~554 ms
#199	Qwen3-8B	8B	INT4 Q8	92 tok/s 58 tok/s	~518 ms ~524 ms
#200	Gemma 3 4B	4B	INT4 Q8 FP16	145 tok/s 99 tok/s 59 tok/s	~270 ms ~273 ms ~280 ms
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	104 tok/s 66 tok/s 35 tok/s	~456 ms ~461 ms ~513 ms
#202	Llama 3.1 8B	8B	INT4 Q8	92 tok/s 58 tok/s	~518 ms ~524 ms
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	217 tok/s 169 tok/s 112 tok/s	~126 ms ~128 ms ~131 ms
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	311 tok/s 293 tok/s 257 tok/s	~26 ms ~26 ms ~27 ms
#208	Llama 3 8B	8B	INT4 Q8	92 tok/s 58 tok/s	~518 ms ~524 ms
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	298 tok/s 261 tok/s 205 tok/s	~49 ms ~49 ms ~50 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	149 tok/s 103 tok/s 61 tok/s	~258 ms ~261 ms ~267 ms
#212	Ministral-8B-2410	8B	INT4 Q8	92 tok/s 58 tok/s	~518 ms ~524 ms
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	228 tok/s 180 tok/s 122 tok/s	~113 ms ~114 ms ~116 ms
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	169 tok/s 121 tok/s 74 tok/s	~208 ms ~210 ms ~216 ms
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	278 tok/s 237 tok/s 178 tok/s	~64 ms ~64 ms ~66 ms
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	149 tok/s 103 tok/s 61 tok/s	~258 ms ~261 ms ~267 ms
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	120 tok/s 103 tok/s 72 tok/s	~159 ms ~161 ms ~177 ms
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	98 tok/s 63 tok/s 31 tok/s	~474 ms ~480 ms ~576 ms
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	260 tok/s 216 tok/s 157 tok/s	~79 ms ~80 ms ~82 ms
#220	Gemma 2 2B	2B	INT4 Q8 FP16	203 tok/s 154 tok/s 100 tok/s	~146 ms ~148 ms ~152 ms
#221	Phi-3-medium	14B	INT4 Q8	59 tok/s 32 tok/s	~893 ms ~1.1s
#222	OLMo 3 7B Instruct	7B	INT4 Q8	74 tok/s 53 tok/s	~459 ms ~465 ms
#224	Qwen2-7B	7B	INT4 Q8 FP16	101 tok/s 65 tok/s 33 tok/s	~456 ms ~461 ms ~554 ms
#225	Phi-3-small	7B	INT4 Q8 FP16	101 tok/s 65 tok/s 33 tok/s	~456 ms ~461 ms ~554 ms
#226	Gemma 3 1B	1B	INT4 Q8 FP16	260 tok/s 216 tok/s 157 tok/s	~79 ms ~80 ms ~82 ms
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	98 tok/s 63 tok/s 31 tok/s	~474 ms ~480 ms ~576 ms
#229	Gemma 1 7B	7B	INT4 Q8	74 tok/s 53 tok/s	~459 ms ~465 ms
#230	Gemma 1 2B	2B	INT4 Q8 FP16	214 tok/s 159 tok/s 102 tok/s	~146 ms ~148 ms ~151 ms
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	82 tok/s 60 tok/s 35 tok/s	~397 ms ~402 ms ~446 ms
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	82 tok/s 60 tok/s 35 tok/s	~397 ms ~402 ms ~446 ms
#233	ChatGLM-6B	6B	INT4 Q8 FP16	82 tok/s 60 tok/s 35 tok/s	~397 ms ~402 ms ~446 ms
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	308 tok/s 275 tok/s 222 tok/s	~42 ms ~42 ms ~43 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	308 tok/s 275 tok/s 222 tok/s	~42 ms ~42 ms ~43 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	304 tok/s 285 tok/s 248 tok/s	~28 ms ~28 ms ~29 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	304 tok/s 285 tok/s 248 tok/s	~28 ms ~28 ms ~29 ms
-	Falcon-1B	1B	INT4 Q8 FP16	270 tok/s 223 tok/s 160 tok/s	~79 ms ~80 ms ~81 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	260 tok/s 216 tok/s 157 tok/s	~79 ms ~80 ms ~82 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	228 tok/s 180 tok/s 122 tok/s	~113 ms ~114 ms ~116 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	228 tok/s 180 tok/s 122 tok/s	~113 ms ~114 ms ~116 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	217 tok/s 169 tok/s 112 tok/s	~126 ms ~128 ms ~131 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	213 tok/s 164 tok/s 108 tok/s	~133 ms ~134 ms ~137 ms
-	Falcon-3B	3B	INT4 Q8 FP16	177 tok/s 125 tok/s 75 tok/s	~208 ms ~210 ms ~215 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	175 tok/s 151 tok/s 112 tok/s	~100 ms ~101 ms ~104 ms
-	Phi-1	1.3B	INT4 Q8 FP16	175 tok/s 151 tok/s 112 tok/s	~100 ms ~101 ms ~104 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	175 tok/s 151 tok/s 112 tok/s	~100 ms ~101 ms ~104 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	171 tok/s 122 tok/s 74 tok/s	~208 ms ~210 ms ~216 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	169 tok/s 121 tok/s 74 tok/s	~208 ms ~210 ms ~216 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	169 tok/s 121 tok/s 74 tok/s	~208 ms ~210 ms ~216 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	145 tok/s 99 tok/s 59 tok/s	~270 ms ~273 ms ~280 ms
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4	143 tok/s	~274 ms
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4	143 tok/s	~274 ms
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4	132 tok/s	~311 ms
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4	132 tok/s	~311 ms
-	Phi-2	2.7B	INT4 Q8 FP16	125 tok/s 101 tok/s 68 tok/s	~192 ms ~194 ms ~199 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	120 tok/s 95 tok/s 63 tok/s	~210 ms ~213 ms ~218 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	120 tok/s 95 tok/s 63 tok/s	~210 ms ~213 ms ~218 ms
-	Falcon-7B	7B	INT4 Q8 FP16	105 tok/s 66 tok/s 35 tok/s	~456 ms ~461 ms ~513 ms
-	Falcon3-7B	7B	INT4 Q8 FP16	101 tok/s 65 tok/s 33 tok/s	~456 ms ~461 ms ~554 ms
-	Hy-MT2 7B	7B	INT4 Q8 FP16	101 tok/s 65 tok/s 33 tok/s	~456 ms ~461 ms ~554 ms
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	98 tok/s 63 tok/s 31 tok/s	~474 ms ~480 ms ~576 ms
-	LensVLM 9B	9B	INT4 Q8	84 tok/s 53 tok/s	~581 ms ~588 ms
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8	84 tok/s 53 tok/s	~581 ms ~588 ms
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	82 tok/s 60 tok/s 35 tok/s	~397 ms ~402 ms ~446 ms
-	Yi-6B	6B	INT4 Q8 FP16	82 tok/s 60 tok/s 35 tok/s	~397 ms ~402 ms ~446 ms
-	Falcon3-10B	10B	INT4 Q8	78 tok/s 48 tok/s	~643 ms ~651 ms
-	Falcon2-11B	11B	INT4 Q8	75 tok/s 45 tok/s	~704 ms ~713 ms
-	Hunyuan Lite	7B	INT4 Q8	74 tok/s 53 tok/s	~459 ms ~465 ms
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4	74 tok/s	~270 ms
-	OLMo 3 7B Base	7B	INT4 Q8	74 tok/s 53 tok/s	~459 ms ~465 ms
-	SEA-LION-7B	7.1B	INT4 Q8	74 tok/s 53 tok/s	~466 ms ~471 ms
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8	74 tok/s 53 tok/s	~466 ms ~471 ms
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8	68 tok/s 48 tok/s	~522 ms ~528 ms
-	Typhoon-2-8B	8B	INT4 Q8	68 tok/s 48 tok/s	~522 ms ~528 ms
-	GLM-4-9B	9B	INT4 Q8	62 tok/s 41 tok/s	~585 ms ~642 ms
-	GLM-4-9B-Chat	9B	INT4 Q8	62 tok/s 41 tok/s	~585 ms ~642 ms
-	GLM-4-9B-Chat-1M	9B	INT4 Q8	62 tok/s 41 tok/s	~585 ms ~642 ms
-	Yi-9B	9B	INT4 Q8	62 tok/s 41 tok/s	~585 ms ~642 ms
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8	61 tok/s 41 tok/s	~597 ms ~655 ms
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8	61 tok/s 41 tok/s	~597 ms ~655 ms
-	GreenMind-14B-R1	14B	INT4	44 tok/s	~899 ms

NVIDIA RTX 4070 Ti SUPER (16GB)

VRAM Scaling by Context Length

Best Models to Run on NVIDIA RTX 4070 Ti SUPER (16GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions