What are the best models to run on NVIDIA RTX 2000 Ada Generation (16GB)?

Based on model rankings and 16 GB usable VRAM: Top Overall: Qwen 3.8 27B (INT4) ~11 tok/s. Top Coding: Qwen3.5-27B (INT4) ~11 tok/s. Optimal Daily Driver: Phi-4 Reasoning Plus (INT4) ~16 tok/s. Maximum Model Size: ERNIE-4.5-VL-28B-A3B (INT4) ~60 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 2000 Ada Generation (16GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4	11 tok/s	~7.2s
#83	Qwen3.5-27B	27B	INT4	11 tok/s	~7.2s
#100	Phi-4 Reasoning Plus	14B	INT4	16 tok/s	~3.3s
#101	MiMo V2 Flash	15B (309B active)	INT4	1 tok/s	~73.0s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4	55 tok/s	~1.2s
#109	Gemma 4 12B	11.95B	INT4 Q8	18 tok/s 11 tok/s	~2.8s ~3.3s
#113	Qwen3.5-9B	9B	INT4 Q8	33 tok/s 19 tok/s	~2.1s ~2.1s
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	41 tok/s 24 tok/s 12 tok/s	~1.6s ~1.7s ~2.0s
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	10 tok/s 6 tok/s 3 tok/s	~6.6s ~6.7s ~6.8s
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	65 tok/s 40 tok/s 22 tok/s	~963 ms ~973 ms ~994 ms
#136	GPT-OSS 20B	21B (3.6B active)	INT4	23 tok/s	~1.2s
#146	Granite 4.2 8B	8B	INT4 Q8	37 tok/s 22 tok/s	~1.9s ~1.9s
#150	Gemma 4 E4B	8B	INT4 Q8	37 tok/s 22 tok/s	~1.9s ~1.9s
#158	GLM-4V	9B	INT4 Q8	23 tok/s 15 tok/s	~2.1s ~2.3s
#159	Gemma 3 27B	27B	INT4	11 tok/s	~7.2s
#160	Phi-4	14B	INT4 Q8	22 tok/s 11 tok/s	~3.2s ~3.8s
#169	OLMo 3 7B Think	7B	INT4 Q8	29 tok/s 20 tok/s	~1.7s ~1.7s
#172	DeepSeek-R1 14B	14B	INT4 Q8	23 tok/s 12 tok/s	~3.2s ~3.8s
#173	Ministral 3 14B	14B	INT4	16 tok/s	~3.3s
#177	Gemma 3 12B	12B	INT4 Q8	26 tok/s 14 tok/s	~2.8s ~3.0s
#181	Mistral-Small-2501	24B	INT4	13 tok/s	~5.9s
#182	Ministral 3 8B	8B	INT4 Q8	26 tok/s 18 tok/s	~1.9s ~1.9s
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	107 tok/s 71 tok/s 41 tok/s	~514 ms ~518 ms ~529 ms
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	107 tok/s 71 tok/s 41 tok/s	~514 ms ~518 ms ~529 ms
#185	Qwen2.5-14B	14B	INT4 Q8	22 tok/s 11 tok/s	~3.2s ~3.8s
#186	Gemma 2 27B	27B	INT4	11 tok/s	~7.2s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	54 tok/s 33 tok/s 17 tok/s	~1.2s ~1.2s ~1.3s
#189	Qwen3-4B	4B	INT4 Q8 FP16	65 tok/s 40 tok/s 22 tok/s	~963 ms ~973 ms ~994 ms
#190	DeepSeek-R1 8B	8B	INT4 Q8	38 tok/s 22 tok/s	~1.9s ~1.9s
#193	Ministral 3 3B	3B	INT4 Q8 FP16	51 tok/s 38 tok/s 24 tok/s	~745 ms ~751 ms ~767 ms
#195	Qwen3-14B	14B	INT4 Q8	22 tok/s 11 tok/s	~3.2s ~3.8s
#197	Gemma 2 9B	9B	INT4 Q8	33 tok/s 19 tok/s	~2.1s ~2.1s
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	41 tok/s 24 tok/s 12 tok/s	~1.6s ~1.7s ~2.0s
#199	Qwen3-8B	8B	INT4 Q8	37 tok/s 22 tok/s	~1.9s ~1.9s
#200	Gemma 3 4B	4B	INT4 Q8 FP16	65 tok/s 40 tok/s 22 tok/s	~963 ms ~973 ms ~994 ms
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	42 tok/s 25 tok/s 12 tok/s	~1.6s ~1.7s ~1.8s
#202	Llama 3.1 8B	8B	INT4 Q8	37 tok/s 22 tok/s	~1.9s ~1.9s
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	119 tok/s 81 tok/s 47 tok/s	~440 ms ~444 ms ~453 ms
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	235 tok/s 207 tok/s 160 tok/s	~74 ms ~75 ms ~76 ms
#208	Llama 3 8B	8B	INT4 Q8	37 tok/s 22 tok/s	~1.9s ~1.9s
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	214 tok/s 164 tok/s 108 tok/s	~158 ms ~159 ms ~163 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	68 tok/s 42 tok/s 23 tok/s	~918 ms ~927 ms ~947 ms
#212	Ministral-8B-2410	8B	INT4 Q8	37 tok/s 22 tok/s	~1.9s ~1.9s
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	129 tok/s 89 tok/s 52 tok/s	~390 ms ~393 ms ~401 ms
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	81 tok/s 51 tok/s 28 tok/s	~738 ms ~745 ms ~760 ms
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	186 tok/s 137 tok/s 87 tok/s	~212 ms ~214 ms ~218 ms
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	68 tok/s 42 tok/s 23 tok/s	~918 ms ~927 ms ~947 ms
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	51 tok/s 42 tok/s 28 tok/s	~558 ms ~562 ms ~620 ms
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	40 tok/s 24 tok/s 11 tok/s	~1.7s ~1.7s ~2.1s
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	163 tok/s 118 tok/s 73 tok/s	~268 ms ~270 ms ~275 ms
#220	Gemma 2 2B	2B	INT4 Q8 FP16	107 tok/s 71 tok/s 41 tok/s	~514 ms ~518 ms ~529 ms
#221	Phi-3-medium	14B	INT4 Q8	22 tok/s 11 tok/s	~3.2s ~3.8s
#222	OLMo 3 7B Instruct	7B	INT4 Q8	29 tok/s 20 tok/s	~1.7s ~1.7s
#224	Qwen2-7B	7B	INT4 Q8 FP16	41 tok/s 24 tok/s 12 tok/s	~1.6s ~1.7s ~2.0s
#225	Phi-3-small	7B	INT4 Q8 FP16	41 tok/s 24 tok/s 12 tok/s	~1.6s ~1.7s ~2.0s
#226	Gemma 3 1B	1B	INT4 Q8 FP16	163 tok/s 118 tok/s 73 tok/s	~268 ms ~270 ms ~275 ms
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	40 tok/s 24 tok/s 11 tok/s	~1.7s ~1.7s ~2.1s
#229	Gemma 1 7B	7B	INT4 Q8	29 tok/s 20 tok/s	~1.7s ~1.7s
#230	Gemma 1 2B	2B	INT4 Q8 FP16	116 tok/s 75 tok/s 42 tok/s	~513 ms ~518 ms ~528 ms
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	32 tok/s 22 tok/s 12 tok/s	~1.4s ~1.4s ~1.6s
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	32 tok/s 22 tok/s 12 tok/s	~1.4s ~1.4s ~1.6s
#233	ChatGLM-6B	6B	INT4 Q8 FP16	32 tok/s 22 tok/s 12 tok/s	~1.4s ~1.4s ~1.6s
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	231 tok/s 181 tok/s 123 tok/s	~132 ms ~133 ms ~136 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	231 tok/s 181 tok/s 123 tok/s	~132 ms ~133 ms ~136 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	224 tok/s 196 tok/s 150 tok/s	~82 ms ~83 ms ~84 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	224 tok/s 196 tok/s 150 tok/s	~82 ms ~83 ms ~84 ms
-	Falcon-1B	1B	INT4 Q8 FP16	176 tok/s 124 tok/s 75 tok/s	~267 ms ~270 ms ~275 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	163 tok/s 118 tok/s 73 tok/s	~268 ms ~270 ms ~275 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	129 tok/s 89 tok/s 52 tok/s	~390 ms ~393 ms ~401 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	129 tok/s 89 tok/s 52 tok/s	~390 ms ~393 ms ~401 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	119 tok/s 81 tok/s 47 tok/s	~440 ms ~444 ms ~453 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	115 tok/s 77 tok/s 45 tok/s	~463 ms ~467 ms ~477 ms
-	Falcon-3B	3B	INT4 Q8 FP16	87 tok/s 54 tok/s 29 tok/s	~737 ms ~744 ms ~760 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	85 tok/s 69 tok/s 47 tok/s	~345 ms ~348 ms ~354 ms
-	Phi-1	1.3B	INT4 Q8 FP16	85 tok/s 69 tok/s 47 tok/s	~345 ms ~348 ms ~354 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	85 tok/s 69 tok/s 47 tok/s	~345 ms ~348 ms ~354 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	82 tok/s 52 tok/s 29 tok/s	~737 ms ~744 ms ~760 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	81 tok/s 51 tok/s 28 tok/s	~738 ms ~745 ms ~760 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	81 tok/s 51 tok/s 28 tok/s	~738 ms ~745 ms ~760 ms
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4	66 tok/s	~978 ms
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4	66 tok/s	~978 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	65 tok/s 40 tok/s 22 tok/s	~963 ms ~973 ms ~994 ms
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4	60 tok/s	~1.1s
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4	60 tok/s	~1.1s
-	Phi-2	2.7B	INT4 Q8 FP16	54 tok/s 41 tok/s 26 tok/s	~678 ms ~683 ms ~698 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	51 tok/s 38 tok/s 24 tok/s	~745 ms ~751 ms ~767 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	51 tok/s 38 tok/s 24 tok/s	~745 ms ~751 ms ~767 ms
-	Falcon-7B	7B	INT4 Q8 FP16	43 tok/s 25 tok/s 12 tok/s	~1.6s ~1.7s ~1.8s
-	Falcon3-7B	7B	INT4 Q8 FP16	41 tok/s 24 tok/s 12 tok/s	~1.6s ~1.7s ~2.0s
-	Hy-MT2 7B	7B	INT4 Q8 FP16	41 tok/s 24 tok/s 12 tok/s	~1.6s ~1.7s ~2.0s
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	40 tok/s 24 tok/s 11 tok/s	~1.7s ~1.7s ~2.1s
-	LensVLM 9B	9B	INT4 Q8	33 tok/s 19 tok/s	~2.1s ~2.1s
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8	33 tok/s 19 tok/s	~2.1s ~2.1s
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	32 tok/s 22 tok/s 12 tok/s	~1.4s ~1.4s ~1.6s
-	Yi-6B	6B	INT4 Q8 FP16	32 tok/s 22 tok/s 12 tok/s	~1.4s ~1.4s ~1.6s
-	Falcon3-10B	10B	INT4 Q8	30 tok/s 18 tok/s	~2.3s ~2.3s
-	Falcon2-11B	11B	INT4 Q8	29 tok/s 16 tok/s	~2.5s ~2.6s
-	Hunyuan Lite	7B	INT4 Q8	29 tok/s 20 tok/s	~1.7s ~1.7s
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4	29 tok/s	~959 ms
-	OLMo 3 7B Base	7B	INT4 Q8	29 tok/s 20 tok/s	~1.7s ~1.7s
-	SEA-LION-7B	7.1B	INT4 Q8	28 tok/s 19 tok/s	~1.7s ~1.7s
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8	28 tok/s 19 tok/s	~1.7s ~1.7s
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8	26 tok/s 18 tok/s	~1.9s ~1.9s
-	Typhoon-2-8B	8B	INT4 Q8	26 tok/s 18 tok/s	~1.9s ~1.9s
-	GLM-4-9B	9B	INT4 Q8	23 tok/s 15 tok/s	~2.1s ~2.3s
-	GLM-4-9B-Chat	9B	INT4 Q8	23 tok/s 15 tok/s	~2.1s ~2.3s
-	GLM-4-9B-Chat-1M	9B	INT4 Q8	23 tok/s 15 tok/s	~2.1s ~2.3s
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8	23 tok/s 15 tok/s	~2.2s ~2.4s
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8	23 tok/s 15 tok/s	~2.2s ~2.4s
-	Yi-9B	9B	INT4 Q8	23 tok/s 15 tok/s	~2.1s ~2.3s
-	GreenMind-14B-R1	14B	INT4	16 tok/s	~3.3s

NVIDIA RTX 2000 Ada Generation (16GB)

VRAM Scaling by Context Length

Best Models to Run on NVIDIA RTX 2000 Ada Generation (16GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions