What are the best models to run on NVIDIA RTX 4080 SUPER (16GB)?

Based on model rankings and 16 GB usable VRAM: Top Overall: Qwen 3.8 27B (INT4) ~33 tok/s. Top Coding: Qwen3.5-27B (INT4) ~33 tok/s. Optimal Daily Driver: Phi-4 Reasoning Plus (INT4) ~48 tok/s. Maximum Model Size: ERNIE-4.5-VL-28B-A3B (INT4) ~139 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 4080 SUPER (16GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4	33 tok/s	~1.7s
#83	Qwen3.5-27B	27B	INT4	33 tok/s	~1.7s
#100	Phi-4 Reasoning Plus	14B	INT4	48 tok/s	~763 ms
#101	MiMo V2 Flash	15B (309B active)	INT4	3 tok/s	~17.0s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4	134 tok/s	~284 ms
#109	Gemma 4 12B	11.95B	INT4 Q8	54 tok/s 34 tok/s	~654 ms ~775 ms
#113	Qwen3.5-9B	9B	INT4 Q8	90 tok/s 57 tok/s	~493 ms ~499 ms
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	108 tok/s 70 tok/s 35 tok/s	~387 ms ~392 ms ~471 ms
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	32 tok/s 19 tok/s 10 tok/s	~1.5s ~1.6s ~1.6s
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	153 tok/s 106 tok/s 63 tok/s	~230 ms ~233 ms ~239 ms
#136	GPT-OSS 20B	21B (3.6B active)	INT4	66 tok/s	~295 ms
#146	Granite 4.2 8B	8B	INT4 Q8	99 tok/s 63 tok/s	~440 ms ~445 ms
#150	Gemma 4 E4B	8B	INT4 Q8	99 tok/s 63 tok/s	~440 ms ~445 ms
#158	GLM-4V	9B	INT4 Q8	67 tok/s 45 tok/s	~497 ms ~545 ms
#159	Gemma 3 27B	27B	INT4	33 tok/s	~1.7s
#160	Phi-4	14B	INT4 Q8	64 tok/s 35 tok/s	~757 ms ~899 ms
#169	OLMo 3 7B Think	7B	INT4 Q8	80 tok/s 58 tok/s	~390 ms ~395 ms
#172	DeepSeek-R1 14B	14B	INT4 Q8	66 tok/s 35 tok/s	~757 ms ~899 ms
#173	Ministral 3 14B	14B	INT4	48 tok/s	~763 ms
#177	Gemma 3 12B	12B	INT4 Q8	72 tok/s 42 tok/s	~652 ms ~716 ms
#181	Mistral-Small-2501	24B	INT4	39 tok/s	~1.4s
#182	Ministral 3 8B	8B	INT4 Q8	73 tok/s 52 tok/s	~443 ms ~449 ms
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	211 tok/s 162 tok/s 107 tok/s	~125 ms ~127 ms ~130 ms
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	211 tok/s 162 tok/s 107 tok/s	~125 ms ~127 ms ~130 ms
#185	Qwen2.5-14B	14B	INT4 Q8	64 tok/s 35 tok/s	~757 ms ~899 ms
#186	Gemma 2 27B	27B	INT4	33 tok/s	~1.7s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	133 tok/s 89 tok/s 52 tok/s	~287 ms ~291 ms ~299 ms
#189	Qwen3-4B	4B	INT4 Q8 FP16	153 tok/s 106 tok/s 63 tok/s	~230 ms ~233 ms ~239 ms
#190	DeepSeek-R1 8B	8B	INT4 Q8	101 tok/s 64 tok/s	~439 ms ~445 ms
#193	Ministral 3 3B	3B	INT4 Q8 FP16	127 tok/s 101 tok/s 68 tok/s	~180 ms ~182 ms ~186 ms
#195	Qwen3-14B	14B	INT4 Q8	64 tok/s 35 tok/s	~757 ms ~899 ms
#197	Gemma 2 9B	9B	INT4 Q8	90 tok/s 57 tok/s	~493 ms ~499 ms
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	108 tok/s 70 tok/s 35 tok/s	~387 ms ~392 ms ~471 ms
#199	Qwen3-8B	8B	INT4 Q8	99 tok/s 63 tok/s	~440 ms ~445 ms
#200	Gemma 3 4B	4B	INT4 Q8 FP16	153 tok/s 106 tok/s 63 tok/s	~230 ms ~233 ms ~239 ms
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	111 tok/s 71 tok/s 38 tok/s	~387 ms ~392 ms ~437 ms
#202	Llama 3.1 8B	8B	INT4 Q8	99 tok/s 63 tok/s	~440 ms ~445 ms
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	226 tok/s 177 tok/s 119 tok/s	~108 ms ~109 ms ~112 ms
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	315 tok/s 299 tok/s 264 tok/s	~23 ms ~23 ms ~24 ms
#208	Llama 3 8B	8B	INT4 Q8	99 tok/s 63 tok/s	~440 ms ~445 ms
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	303 tok/s 268 tok/s 213 tok/s	~43 ms ~43 ms ~44 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	157 tok/s 110 tok/s 66 tok/s	~219 ms ~222 ms ~228 ms
#212	Ministral-8B-2410	8B	INT4 Q8	99 tok/s 63 tok/s	~440 ms ~445 ms
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	236 tok/s 188 tok/s 130 tok/s	~96 ms ~98 ms ~100 ms
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	178 tok/s 128 tok/s 79 tok/s	~177 ms ~179 ms ~184 ms
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	284 tok/s 244 tok/s 186 tok/s	~55 ms ~56 ms ~57 ms
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	157 tok/s 110 tok/s 66 tok/s	~219 ms ~222 ms ~228 ms
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	127 tok/s 110 tok/s 78 tok/s	~136 ms ~137 ms ~152 ms
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	105 tok/s 68 tok/s 34 tok/s	~403 ms ~408 ms ~490 ms
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	267 tok/s 224 tok/s 165 tok/s	~68 ms ~69 ms ~70 ms
#220	Gemma 2 2B	2B	INT4 Q8 FP16	211 tok/s 162 tok/s 107 tok/s	~125 ms ~127 ms ~130 ms
#221	Phi-3-medium	14B	INT4 Q8	64 tok/s 35 tok/s	~757 ms ~899 ms
#222	OLMo 3 7B Instruct	7B	INT4 Q8	80 tok/s 58 tok/s	~390 ms ~395 ms
#224	Qwen2-7B	7B	INT4 Q8 FP16	108 tok/s 70 tok/s 35 tok/s	~387 ms ~392 ms ~471 ms
#225	Phi-3-small	7B	INT4 Q8 FP16	108 tok/s 70 tok/s 35 tok/s	~387 ms ~392 ms ~471 ms
#226	Gemma 3 1B	1B	INT4 Q8 FP16	267 tok/s 224 tok/s 165 tok/s	~68 ms ~69 ms ~70 ms
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	105 tok/s 68 tok/s 34 tok/s	~403 ms ~408 ms ~490 ms
#229	Gemma 1 7B	7B	INT4 Q8	80 tok/s 58 tok/s	~390 ms ~395 ms
#230	Gemma 1 2B	2B	INT4 Q8 FP16	222 tok/s 168 tok/s 109 tok/s	~125 ms ~126 ms ~130 ms
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	88 tok/s 65 tok/s 38 tok/s	~338 ms ~342 ms ~380 ms
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	88 tok/s 65 tok/s 38 tok/s	~338 ms ~342 ms ~380 ms
#233	ChatGLM-6B	6B	INT4 Q8 FP16	88 tok/s 65 tok/s 38 tok/s	~338 ms ~342 ms ~380 ms
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	313 tok/s 281 tok/s 230 tok/s	~37 ms ~37 ms ~38 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	313 tok/s 281 tok/s 230 tok/s	~37 ms ~37 ms ~38 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	309 tok/s 291 tok/s 256 tok/s	~25 ms ~25 ms ~26 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	309 tok/s 291 tok/s 256 tok/s	~25 ms ~25 ms ~26 ms
-	Falcon-1B	1B	INT4 Q8 FP16	277 tok/s 231 tok/s 169 tok/s	~68 ms ~69 ms ~70 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	267 tok/s 224 tok/s 165 tok/s	~68 ms ~69 ms ~70 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	236 tok/s 188 tok/s 130 tok/s	~96 ms ~98 ms ~100 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	236 tok/s 188 tok/s 130 tok/s	~96 ms ~98 ms ~100 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	226 tok/s 177 tok/s 119 tok/s	~108 ms ~109 ms ~112 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	221 tok/s 172 tok/s 115 tok/s	~113 ms ~115 ms ~118 ms
-	Falcon-3B	3B	INT4 Q8 FP16	186 tok/s 132 tok/s 81 tok/s	~177 ms ~179 ms ~184 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	183 tok/s 159 tok/s 120 tok/s	~86 ms ~87 ms ~89 ms
-	Phi-1	1.3B	INT4 Q8 FP16	183 tok/s 159 tok/s 120 tok/s	~86 ms ~87 ms ~89 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	183 tok/s 159 tok/s 120 tok/s	~86 ms ~87 ms ~89 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	180 tok/s 129 tok/s 80 tok/s	~177 ms ~179 ms ~184 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	178 tok/s 128 tok/s 79 tok/s	~177 ms ~179 ms ~184 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	178 tok/s 128 tok/s 79 tok/s	~177 ms ~179 ms ~184 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	153 tok/s 106 tok/s 63 tok/s	~230 ms ~233 ms ~239 ms
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4	151 tok/s	~233 ms
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4	151 tok/s	~233 ms
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4	139 tok/s	~264 ms
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4	139 tok/s	~264 ms
-	Phi-2	2.7B	INT4 Q8 FP16	133 tok/s 107 tok/s 73 tok/s	~164 ms ~166 ms ~170 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	127 tok/s 101 tok/s 68 tok/s	~180 ms ~182 ms ~186 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	127 tok/s 101 tok/s 68 tok/s	~180 ms ~182 ms ~186 ms
-	Falcon-7B	7B	INT4 Q8 FP16	112 tok/s 71 tok/s 38 tok/s	~387 ms ~392 ms ~437 ms
-	Falcon3-7B	7B	INT4 Q8 FP16	108 tok/s 70 tok/s 35 tok/s	~387 ms ~392 ms ~471 ms
-	Hy-MT2 7B	7B	INT4 Q8 FP16	108 tok/s 70 tok/s 35 tok/s	~387 ms ~392 ms ~471 ms
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	105 tok/s 68 tok/s 34 tok/s	~403 ms ~408 ms ~490 ms
-	LensVLM 9B	9B	INT4 Q8	90 tok/s 57 tok/s	~493 ms ~499 ms
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8	90 tok/s 57 tok/s	~493 ms ~499 ms
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	88 tok/s 65 tok/s 38 tok/s	~338 ms ~342 ms ~380 ms
-	Yi-6B	6B	INT4 Q8 FP16	88 tok/s 65 tok/s 38 tok/s	~338 ms ~342 ms ~380 ms
-	Falcon3-10B	10B	INT4 Q8	84 tok/s 52 tok/s	~545 ms ~553 ms
-	Falcon2-11B	11B	INT4 Q8	81 tok/s 49 tok/s	~598 ms ~606 ms
-	Hunyuan Lite	7B	INT4 Q8	80 tok/s 58 tok/s	~390 ms ~395 ms
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4	80 tok/s	~230 ms
-	OLMo 3 7B Base	7B	INT4 Q8	80 tok/s 58 tok/s	~390 ms ~395 ms
-	SEA-LION-7B	7.1B	INT4 Q8	79 tok/s 57 tok/s	~396 ms ~401 ms
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8	79 tok/s 57 tok/s	~396 ms ~401 ms
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8	73 tok/s 52 tok/s	~443 ms ~449 ms
-	Typhoon-2-8B	8B	INT4 Q8	73 tok/s 52 tok/s	~443 ms ~449 ms
-	GLM-4-9B	9B	INT4 Q8	67 tok/s 45 tok/s	~497 ms ~545 ms
-	GLM-4-9B-Chat	9B	INT4 Q8	67 tok/s 45 tok/s	~497 ms ~545 ms
-	GLM-4-9B-Chat-1M	9B	INT4 Q8	67 tok/s 45 tok/s	~497 ms ~545 ms
-	Yi-9B	9B	INT4 Q8	67 tok/s 45 tok/s	~497 ms ~545 ms
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8	66 tok/s 44 tok/s	~507 ms ~557 ms
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8	66 tok/s 44 tok/s	~507 ms ~557 ms
-	GreenMind-14B-R1	14B	INT4	48 tok/s	~763 ms

NVIDIA RTX 4080 SUPER (16GB)

VRAM Scaling by Context Length

Best Models to Run on NVIDIA RTX 4080 SUPER (16GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions