What are the best models to run on NVIDIA RTX 5070 Ti (16GB)?

Based on model rankings and 16 GB usable VRAM: Top Overall: Qwen 3.8 27B (INT4) ~39 tok/s. Top Coding: Qwen3.5-27B (INT4) ~39 tok/s. Optimal Daily Driver: Phi-4 Reasoning Plus (INT4) ~56 tok/s. Maximum Model Size: ERNIE-4.5-VL-28B-A3B (INT4) ~155 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 5070 Ti (16GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4	39 tok/s	~2.0s
#83	Qwen3.5-27B	27B	INT4	39 tok/s	~2.0s
#100	Phi-4 Reasoning Plus	14B	INT4	56 tok/s	~897 ms
#101	MiMo V2 Flash	15B (309B active)	INT4	4 tok/s	~20.0s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4	151 tok/s	~334 ms
#109	Gemma 4 12B	11.95B	INT4 Q8	64 tok/s 40 tok/s	~769 ms ~909 ms
#113	Qwen3.5-9B	9B	INT4 Q8	104 tok/s 67 tok/s	~580 ms ~586 ms
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	124 tok/s 82 tok/s 42 tok/s	~456 ms ~460 ms ~549 ms
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	38 tok/s 23 tok/s 12 tok/s	~1.8s ~1.8s ~1.9s
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	171 tok/s 122 tok/s 74 tok/s	~270 ms ~272 ms ~277 ms
#136	GPT-OSS 20B	21B (3.6B active)	INT4	77 tok/s	~344 ms
#146	Granite 4.2 8B	8B	INT4 Q8	113 tok/s 74 tok/s	~517 ms ~522 ms
#150	Gemma 4 E4B	8B	INT4 Q8	113 tok/s 74 tok/s	~517 ms ~522 ms
#158	GLM-4V	9B	INT4 Q8	78 tok/s 53 tok/s	~584 ms ~638 ms
#159	Gemma 3 27B	27B	INT4	39 tok/s	~2.0s
#160	Phi-4	14B	INT4 Q8	75 tok/s 42 tok/s	~892 ms ~1.1s
#169	OLMo 3 7B Think	7B	INT4 Q8	93 tok/s 68 tok/s	~458 ms ~462 ms
#172	DeepSeek-R1 14B	14B	INT4 Q8	77 tok/s 42 tok/s	~892 ms ~1.1s
#173	Ministral 3 14B	14B	INT4	56 tok/s	~897 ms
#177	Gemma 3 12B	12B	INT4 Q8	85 tok/s 50 tok/s	~768 ms ~841 ms
#181	Mistral-Small-2501	24B	INT4	46 tok/s	~1.6s
#182	Ministral 3 8B	8B	INT4 Q8	85 tok/s 62 tok/s	~520 ms ~525 ms
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	229 tok/s 180 tok/s 122 tok/s	~146 ms ~148 ms ~150 ms
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	229 tok/s 180 tok/s 122 tok/s	~146 ms ~148 ms ~150 ms
#185	Qwen2.5-14B	14B	INT4 Q8	75 tok/s 42 tok/s	~892 ms ~1.1s
#186	Gemma 2 27B	27B	INT4	39 tok/s	~2.0s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	150 tok/s 103 tok/s 61 tok/s	~338 ms ~341 ms ~348 ms
#189	Qwen3-4B	4B	INT4 Q8 FP16	171 tok/s 122 tok/s 74 tok/s	~270 ms ~272 ms ~277 ms
#190	DeepSeek-R1 8B	8B	INT4 Q8	116 tok/s 75 tok/s	~517 ms ~522 ms
#193	Ministral 3 3B	3B	INT4 Q8 FP16	144 tok/s 117 tok/s 80 tok/s	~210 ms ~211 ms ~215 ms
#195	Qwen3-14B	14B	INT4 Q8	75 tok/s 42 tok/s	~892 ms ~1.1s
#197	Gemma 2 9B	9B	INT4 Q8	104 tok/s 67 tok/s	~580 ms ~586 ms
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	124 tok/s 82 tok/s 42 tok/s	~456 ms ~460 ms ~549 ms
#199	Qwen3-8B	8B	INT4 Q8	113 tok/s 74 tok/s	~517 ms ~522 ms
#200	Gemma 3 4B	4B	INT4 Q8 FP16	171 tok/s 122 tok/s 74 tok/s	~270 ms ~272 ms ~277 ms
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	126 tok/s 83 tok/s 45 tok/s	~455 ms ~460 ms ~508 ms
#202	Llama 3.1 8B	8B	INT4 Q8	113 tok/s 74 tok/s	~517 ms ~522 ms
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	242 tok/s 195 tok/s 136 tok/s	~126 ms ~127 ms ~129 ms
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	324 tok/s 309 tok/s 279 tok/s	~26 ms ~26 ms ~26 ms
#208	Llama 3 8B	8B	INT4 Q8	113 tok/s 74 tok/s	~517 ms ~522 ms
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	313 tok/s 282 tok/s 231 tok/s	~49 ms ~49 ms ~50 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	175 tok/s 126 tok/s 77 tok/s	~257 ms ~260 ms ~265 ms
#212	Ministral-8B-2410	8B	INT4 Q8	113 tok/s 74 tok/s	~517 ms ~522 ms
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	253 tok/s 206 tok/s 147 tok/s	~113 ms ~113 ms ~115 ms
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	196 tok/s 145 tok/s 92 tok/s	~208 ms ~210 ms ~214 ms
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	297 tok/s 260 tok/s 204 tok/s	~64 ms ~64 ms ~65 ms
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	175 tok/s 126 tok/s 77 tok/s	~257 ms ~260 ms ~265 ms
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	144 tok/s 126 tok/s 90 tok/s	~158 ms ~159 ms ~175 ms
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	121 tok/s 79 tok/s 40 tok/s	~474 ms ~478 ms ~571 ms
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	281 tok/s 241 tok/s 183 tok/s	~79 ms ~80 ms ~81 ms
#220	Gemma 2 2B	2B	INT4 Q8 FP16	229 tok/s 180 tok/s 122 tok/s	~146 ms ~148 ms ~150 ms
#221	Phi-3-medium	14B	INT4 Q8	75 tok/s 42 tok/s	~892 ms ~1.1s
#222	OLMo 3 7B Instruct	7B	INT4 Q8	93 tok/s 68 tok/s	~458 ms ~462 ms
#224	Qwen2-7B	7B	INT4 Q8 FP16	124 tok/s 82 tok/s 42 tok/s	~456 ms ~460 ms ~549 ms
#225	Phi-3-small	7B	INT4 Q8 FP16	124 tok/s 82 tok/s 42 tok/s	~456 ms ~460 ms ~549 ms
#226	Gemma 3 1B	1B	INT4 Q8 FP16	281 tok/s 241 tok/s 183 tok/s	~79 ms ~80 ms ~81 ms
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	121 tok/s 79 tok/s 40 tok/s	~474 ms ~478 ms ~571 ms
#229	Gemma 1 7B	7B	INT4 Q8	93 tok/s 68 tok/s	~458 ms ~462 ms
#230	Gemma 1 2B	2B	INT4 Q8 FP16	239 tok/s 186 tok/s 125 tok/s	~146 ms ~147 ms ~150 ms
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	102 tok/s 76 tok/s 45 tok/s	~396 ms ~400 ms ~441 ms
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	102 tok/s 76 tok/s 45 tok/s	~396 ms ~400 ms ~441 ms
#233	ChatGLM-6B	6B	INT4 Q8 FP16	102 tok/s 76 tok/s 45 tok/s	~396 ms ~400 ms ~441 ms
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	322 tok/s 294 tok/s 246 tok/s	~42 ms ~42 ms ~43 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	322 tok/s 294 tok/s 246 tok/s	~42 ms ~42 ms ~43 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	318 tok/s 303 tok/s 270 tok/s	~28 ms ~28 ms ~29 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	318 tok/s 303 tok/s 270 tok/s	~28 ms ~28 ms ~29 ms
-	Falcon-1B	1B	INT4 Q8 FP16	290 tok/s 247 tok/s 187 tok/s	~79 ms ~79 ms ~81 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	281 tok/s 241 tok/s 183 tok/s	~79 ms ~80 ms ~81 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	253 tok/s 206 tok/s 147 tok/s	~113 ms ~113 ms ~115 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	253 tok/s 206 tok/s 147 tok/s	~113 ms ~113 ms ~115 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	242 tok/s 195 tok/s 136 tok/s	~126 ms ~127 ms ~129 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	238 tok/s 190 tok/s 131 tok/s	~133 ms ~134 ms ~136 ms
-	Falcon-3B	3B	INT4 Q8 FP16	204 tok/s 149 tok/s 94 tok/s	~208 ms ~209 ms ~213 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	201 tok/s 177 tok/s 136 tok/s	~100 ms ~101 ms ~102 ms
-	Phi-1	1.3B	INT4 Q8 FP16	201 tok/s 177 tok/s 136 tok/s	~100 ms ~101 ms ~102 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	201 tok/s 177 tok/s 136 tok/s	~100 ms ~101 ms ~102 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	198 tok/s 146 tok/s 93 tok/s	~208 ms ~210 ms ~214 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	196 tok/s 145 tok/s 92 tok/s	~208 ms ~210 ms ~214 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	196 tok/s 145 tok/s 92 tok/s	~208 ms ~210 ms ~214 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	171 tok/s 122 tok/s 74 tok/s	~270 ms ~272 ms ~277 ms
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4	168 tok/s	~274 ms
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4	168 tok/s	~274 ms
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4	155 tok/s	~310 ms
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4	155 tok/s	~310 ms
-	Phi-2	2.7B	INT4 Q8 FP16	150 tok/s 123 tok/s 86 tok/s	~191 ms ~193 ms ~196 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	144 tok/s 117 tok/s 80 tok/s	~210 ms ~211 ms ~215 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	144 tok/s 117 tok/s 80 tok/s	~210 ms ~211 ms ~215 ms
-	Falcon-7B	7B	INT4 Q8 FP16	128 tok/s 84 tok/s 45 tok/s	~455 ms ~460 ms ~508 ms
-	Falcon3-7B	7B	INT4 Q8 FP16	124 tok/s 82 tok/s 42 tok/s	~456 ms ~460 ms ~549 ms
-	Hy-MT2 7B	7B	INT4 Q8 FP16	124 tok/s 82 tok/s 42 tok/s	~456 ms ~460 ms ~549 ms
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	121 tok/s 79 tok/s 40 tok/s	~474 ms ~478 ms ~571 ms
-	LensVLM 9B	9B	INT4 Q8	104 tok/s 67 tok/s	~580 ms ~586 ms
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8	104 tok/s 67 tok/s	~580 ms ~586 ms
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	102 tok/s 76 tok/s 45 tok/s	~396 ms ~400 ms ~441 ms
-	Yi-6B	6B	INT4 Q8 FP16	102 tok/s 76 tok/s 45 tok/s	~396 ms ~400 ms ~441 ms
-	Falcon3-10B	10B	INT4 Q8	97 tok/s 61 tok/s	~642 ms ~648 ms
-	Falcon2-11B	11B	INT4 Q8	94 tok/s 58 tok/s	~704 ms ~711 ms
-	Hunyuan Lite	7B	INT4 Q8	93 tok/s 68 tok/s	~458 ms ~462 ms
-	OLMo 3 7B Base	7B	INT4 Q8	93 tok/s 68 tok/s	~458 ms ~462 ms
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4	92 tok/s	~268 ms
-	SEA-LION-7B	7.1B	INT4 Q8	92 tok/s 67 tok/s	~464 ms ~468 ms
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8	92 tok/s 67 tok/s	~464 ms ~468 ms
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8	85 tok/s 62 tok/s	~520 ms ~525 ms
-	Typhoon-2-8B	8B	INT4 Q8	85 tok/s 62 tok/s	~520 ms ~525 ms
-	GLM-4-9B	9B	INT4 Q8	78 tok/s 53 tok/s	~584 ms ~638 ms
-	GLM-4-9B-Chat	9B	INT4 Q8	78 tok/s 53 tok/s	~584 ms ~638 ms
-	GLM-4-9B-Chat-1M	9B	INT4 Q8	78 tok/s 53 tok/s	~584 ms ~638 ms
-	Yi-9B	9B	INT4 Q8	78 tok/s 53 tok/s	~584 ms ~638 ms
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8	77 tok/s 52 tok/s	~596 ms ~652 ms
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8	77 tok/s 52 tok/s	~596 ms ~652 ms
-	GreenMind-14B-R1	14B	INT4	56 tok/s	~897 ms

NVIDIA RTX 5070 Ti (16GB)

VRAM Scaling by Context Length

Best Models to Run on NVIDIA RTX 5070 Ti (16GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions