What are the best models to run on NVIDIA RTX 3090 Ti (24GB)?

Based on model rankings and 24 GB usable VRAM: Top Overall: Qwen 3.8 27B (INT4) ~48 tok/s. Top Coding: Qwen3.5-27B (INT4) ~48 tok/s. Optimal Daily Driver: Phi-4 Reasoning Plus (INT4) ~62 tok/s. Maximum Model Size: Falcon-40B (INT4) ~34 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 3090 Ti (24GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4	48 tok/s	~1.9s
#48	Agnes 3.0 Flash	33B	INT4	41 tok/s	~2.3s
#72	K2 Horizon MoVA 36B A4B	36B (4B active)	INT4	153 tok/s	~398 ms
#83	Qwen3.5-27B	27B	INT4	48 tok/s	~1.9s
#92	Sarvam-30B	32B (2.4B active)	INT4	194 tok/s	~258 ms
#94	Qwen3.6 35B A3B	35B (3B active)	INT4	171 tok/s	~325 ms
#96	Muse Glimmer 30B	30B	INT4	44 tok/s	~2.1s
#100	Phi-4 Reasoning Plus	14B	INT4 Q8	62 tok/s 43 tok/s	~981 ms ~988 ms
#101	MiMo V2 Flash	15B (309B active)	INT4 Q8	5 tok/s 3 tok/s	~20.2s ~22.1s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4	169 tok/s	~337 ms
#109	Gemma 4 12B	11.95B	INT4 Q8	70 tok/s 49 tok/s	~841 ms ~847 ms
#113	Qwen3.5-9B	9B	INT4 Q8 FP16	113 tok/s 74 tok/s 39 tok/s	~635 ms ~640 ms ~705 ms
#114	Qwen3.5-35B-A3B	35B (3B active)	INT4	171 tok/s	~325 ms
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	134 tok/s 89 tok/s 52 tok/s	~498 ms ~502 ms ~510 ms
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	42 tok/s 25 tok/s 13 tok/s	~2.0s ~2.0s ~2.0s
#129	Gemma 4 31B	30.7B	INT4	43 tok/s	~2.1s
#131	Nemotron 3.5 Lightning	30B (3B active)	INT4	81 tok/s	~326 ms
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	182 tok/s 131 tok/s 81 tok/s	~295 ms ~297 ms ~302 ms
#133	Qwen3-30B-A3B	30B (3B active)	INT4	182 tok/s	~294 ms
#136	GPT-OSS 20B	21B (3.6B active)	INT4	93 tok/s	~320 ms
#146	Granite 4.2 8B	8B	INT4 Q8 FP16	123 tok/s 81 tok/s 46 tok/s	~566 ms ~570 ms ~580 ms
#150	Gemma 4 E4B	8B	INT4 Q8 FP16	123 tok/s 81 tok/s 46 tok/s	~566 ms ~570 ms ~580 ms
#151	Nemotron 3.5 Lightning 30B A3B	30B (3B active)	INT4	182 tok/s	~294 ms
#153	Qwen3-32B	32B	INT4	42 tok/s	~2.2s
#156	GLM-4	32B	INT4	30 tok/s	~2.6s
#158	GLM-4V	9B	INT4 Q8 FP16	86 tok/s 62 tok/s 36 tok/s	~638 ms ~642 ms ~708 ms
#159	Gemma 3 27B	27B	INT4	48 tok/s	~1.9s
#160	Phi-4	14B	INT4 Q8	82 tok/s 51 tok/s	~977 ms ~984 ms
#162	DeepSeek-R1 32B	32B	INT4	43 tok/s	~2.2s
#164	Qwen2.5-32B	32B	INT4	42 tok/s	~2.2s
#168	Magistral Small	24B	INT4	41 tok/s	~1.7s
#169	OLMo 3 7B Think	7B	INT4 Q8 FP16	101 tok/s 75 tok/s 46 tok/s	~501 ms ~504 ms ~512 ms
#170	OLMo 3.1 32B Instruct	32B	INT4	30 tok/s	~2.6s
#172	DeepSeek-R1 14B	14B	INT4 Q8	85 tok/s 52 tok/s	~976 ms ~984 ms
#173	Ministral 3 14B	14B	INT4 Q8	62 tok/s 43 tok/s	~981 ms ~988 ms
#177	Gemma 3 12B	12B	INT4 Q8 FP16	92 tok/s 58 tok/s 29 tok/s	~840 ms ~847 ms ~1.0s
#180	OLMo 3 32B Base	32B	INT4	30 tok/s	~2.6s
#181	Mistral-Small-2501	24B	INT4	53 tok/s	~1.7s
#182	Ministral 3 8B	8B	INT4 Q8 FP16	93 tok/s 68 tok/s 39 tok/s	~569 ms ~573 ms ~631 ms
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	239 tok/s 191 tok/s 132 tok/s	~160 ms ~161 ms ~163 ms
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	239 tok/s 191 tok/s 132 tok/s	~160 ms ~161 ms ~163 ms
#185	Qwen2.5-14B	14B	INT4 Q8	82 tok/s 51 tok/s	~977 ms ~984 ms
#186	Gemma 2 27B	27B	INT4	48 tok/s	~1.9s
#187	OLMo 3.1 32B Think	32B	INT4	30 tok/s	~2.6s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	161 tok/s 112 tok/s 67 tok/s	~369 ms ~372 ms ~378 ms
#189	Qwen3-4B	4B	INT4 Q8 FP16	182 tok/s 131 tok/s 81 tok/s	~295 ms ~297 ms ~302 ms
#190	DeepSeek-R1 8B	8B	INT4 Q8 FP16	125 tok/s 82 tok/s 46 tok/s	~566 ms ~570 ms ~580 ms
#193	Ministral 3 3B	3B	INT4 Q8 FP16	154 tok/s 126 tok/s 88 tok/s	~229 ms ~230 ms ~234 ms
#195	Qwen3-14B	14B	INT4 Q8	82 tok/s 51 tok/s	~977 ms ~984 ms
#197	Gemma 2 9B	9B	INT4 Q8 FP16	113 tok/s 74 tok/s 39 tok/s	~635 ms ~640 ms ~705 ms
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	134 tok/s 89 tok/s 52 tok/s	~498 ms ~502 ms ~510 ms
#199	Qwen3-8B	8B	INT4 Q8 FP16	123 tok/s 81 tok/s 46 tok/s	~566 ms ~570 ms ~580 ms
#200	Gemma 3 4B	4B	INT4 Q8 FP16	182 tok/s 131 tok/s 81 tok/s	~295 ms ~297 ms ~302 ms
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	136 tok/s 91 tok/s 52 tok/s	~498 ms ~502 ms ~510 ms
#202	Llama 3.1 8B	8B	INT4 Q8 FP16	123 tok/s 81 tok/s 46 tok/s	~566 ms ~570 ms ~580 ms
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	252 tok/s 206 tok/s 146 tok/s	~138 ms ~138 ms ~140 ms
#205	Command R	35B	INT4	28 tok/s	~2.8s
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	328 tok/s 315 tok/s 287 tok/s	~28 ms ~28 ms ~28 ms
#208	Llama 3 8B	8B	INT4 Q8 FP16	123 tok/s 81 tok/s 46 tok/s	~566 ms ~570 ms ~580 ms
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	318 tok/s 289 tok/s 241 tok/s	~53 ms ~53 ms ~54 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	186 tok/s 136 tok/s 85 tok/s	~281 ms ~283 ms ~288 ms
#212	Ministral-8B-2410	8B	INT4 Q8 FP16	123 tok/s 81 tok/s 46 tok/s	~566 ms ~570 ms ~580 ms
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	262 tok/s 217 tok/s 157 tok/s	~122 ms ~123 ms ~125 ms
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	207 tok/s 156 tok/s 101 tok/s	~227 ms ~229 ms ~232 ms
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	303 tok/s 269 tok/s 215 tok/s	~69 ms ~69 ms ~70 ms
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	186 tok/s 136 tok/s 85 tok/s	~281 ms ~283 ms ~288 ms
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	154 tok/s 136 tok/s 103 tok/s	~172 ms ~173 ms ~175 ms
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	130 tok/s 87 tok/s 50 tok/s	~518 ms ~522 ms ~531 ms
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	289 tok/s 251 tok/s 194 tok/s	~86 ms ~86 ms ~88 ms
#220	Gemma 2 2B	2B	INT4 Q8 FP16	239 tok/s 191 tok/s 132 tok/s	~160 ms ~161 ms ~163 ms
#221	Phi-3-medium	14B	INT4 Q8	82 tok/s 51 tok/s	~977 ms ~984 ms
#222	OLMo 3 7B Instruct	7B	INT4 Q8 FP16	101 tok/s 75 tok/s 46 tok/s	~501 ms ~504 ms ~512 ms
#223	Yi-34B	34B	INT4	28 tok/s	~2.7s
#224	Qwen2-7B	7B	INT4 Q8 FP16	134 tok/s 89 tok/s 52 tok/s	~498 ms ~502 ms ~510 ms
#225	Phi-3-small	7B	INT4 Q8 FP16	134 tok/s 89 tok/s 52 tok/s	~498 ms ~502 ms ~510 ms
#226	Gemma 3 1B	1B	INT4 Q8 FP16	289 tok/s 251 tok/s 194 tok/s	~86 ms ~86 ms ~88 ms
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	130 tok/s 87 tok/s 50 tok/s	~518 ms ~522 ms ~531 ms
#229	Gemma 1 7B	7B	INT4 Q8 FP16	101 tok/s 75 tok/s 46 tok/s	~501 ms ~504 ms ~512 ms
#230	Gemma 1 2B	2B	INT4 Q8 FP16	249 tok/s 197 tok/s 135 tok/s	~159 ms ~161 ms ~163 ms
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	111 tok/s 83 tok/s 53 tok/s	~433 ms ~436 ms ~443 ms
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	111 tok/s 83 tok/s 53 tok/s	~433 ms ~436 ms ~443 ms
#233	ChatGLM-6B	6B	INT4 Q8 FP16	111 tok/s 83 tok/s 53 tok/s	~433 ms ~436 ms ~443 ms
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	326 tok/s 301 tok/s 256 tok/s	~45 ms ~45 ms ~46 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	326 tok/s 301 tok/s 256 tok/s	~45 ms ~45 ms ~46 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	323 tok/s 309 tok/s 279 tok/s	~30 ms ~30 ms ~31 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	323 tok/s 309 tok/s 279 tok/s	~30 ms ~30 ms ~31 ms
-	Falcon-1B	1B	INT4 Q8 FP16	297 tok/s 257 tok/s 198 tok/s	~86 ms ~86 ms ~87 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	289 tok/s 251 tok/s 194 tok/s	~86 ms ~86 ms ~88 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	262 tok/s 217 tok/s 157 tok/s	~122 ms ~123 ms ~125 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	262 tok/s 217 tok/s 157 tok/s	~122 ms ~123 ms ~125 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	252 tok/s 206 tok/s 146 tok/s	~138 ms ~138 ms ~140 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	248 tok/s 201 tok/s 141 tok/s	~144 ms ~145 ms ~148 ms
-	Falcon-3B	3B	INT4 Q8 FP16	215 tok/s 160 tok/s 103 tok/s	~227 ms ~228 ms ~232 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	212 tok/s 188 tok/s 146 tok/s	~109 ms ~109 ms ~111 ms
-	Phi-1	1.3B	INT4 Q8 FP16	212 tok/s 188 tok/s 146 tok/s	~109 ms ~109 ms ~111 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	212 tok/s 188 tok/s 146 tok/s	~109 ms ~109 ms ~111 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	209 tok/s 157 tok/s 101 tok/s	~227 ms ~229 ms ~232 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	207 tok/s 156 tok/s 101 tok/s	~227 ms ~229 ms ~232 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	207 tok/s 156 tok/s 101 tok/s	~227 ms ~229 ms ~232 ms
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4 Q8	188 tok/s 131 tok/s	~276 ms ~325 ms
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4 Q8	188 tok/s 131 tok/s	~276 ms ~325 ms
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4	183 tok/s	~290 ms
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4	183 tok/s	~290 ms
-	Hy-MT2 30B A3B	30B (3B active)	INT4	182 tok/s	~294 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	182 tok/s 131 tok/s 81 tok/s	~295 ms ~297 ms ~302 ms
-	Phi-2	2.7B	INT4 Q8 FP16	160 tok/s 133 tok/s 94 tok/s	~208 ms ~210 ms ~213 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	154 tok/s 126 tok/s 88 tok/s	~229 ms ~230 ms ~234 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	154 tok/s 126 tok/s 88 tok/s	~229 ms ~230 ms ~234 ms
-	Falcon-7B	7B	INT4 Q8 FP16	138 tok/s 91 tok/s 52 tok/s	~498 ms ~502 ms ~510 ms
-	Falcon3-7B	7B	INT4 Q8 FP16	134 tok/s 89 tok/s 52 tok/s	~498 ms ~502 ms ~510 ms
-	Hy-MT2 7B	7B	INT4 Q8 FP16	134 tok/s 89 tok/s 52 tok/s	~498 ms ~502 ms ~510 ms
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	130 tok/s 87 tok/s 50 tok/s	~518 ms ~522 ms ~531 ms
-	LensVLM 9B	9B	INT4 Q8 FP16	113 tok/s 74 tok/s 39 tok/s	~635 ms ~640 ms ~705 ms
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8 FP16	113 tok/s 74 tok/s 39 tok/s	~635 ms ~640 ms ~705 ms
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	111 tok/s 83 tok/s 53 tok/s	~433 ms ~436 ms ~443 ms
-	Yi-6B	6B	INT4 Q8 FP16	111 tok/s 83 tok/s 53 tok/s	~433 ms ~436 ms ~443 ms
-	Falcon3-10B	10B	INT4 Q8 FP16	106 tok/s 68 tok/s 36 tok/s	~703 ms ~708 ms ~781 ms
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4 Q8	106 tok/s 89 tok/s	~269 ms ~293 ms
-	Falcon2-11B	11B	INT4 Q8 FP16	102 tok/s 64 tok/s 32 tok/s	~771 ms ~776 ms ~925 ms
-	Hunyuan Lite	7B	INT4 Q8 FP16	101 tok/s 75 tok/s 46 tok/s	~501 ms ~504 ms ~512 ms
-	OLMo 3 7B Base	7B	INT4 Q8 FP16	101 tok/s 75 tok/s 46 tok/s	~501 ms ~504 ms ~512 ms
-	SEA-LION-7B	7.1B	INT4 Q8 FP16	101 tok/s 74 tok/s 46 tok/s	~507 ms ~511 ms ~519 ms
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8 FP16	101 tok/s 74 tok/s 46 tok/s	~507 ms ~511 ms ~519 ms
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8 FP16	93 tok/s 68 tok/s 39 tok/s	~569 ms ~573 ms ~631 ms
-	Typhoon-2-8B	8B	INT4 Q8 FP16	93 tok/s 68 tok/s 39 tok/s	~569 ms ~573 ms ~631 ms
-	GLM-4-9B	9B	INT4 Q8 FP16	86 tok/s 62 tok/s 36 tok/s	~638 ms ~642 ms ~708 ms
-	GLM-4-9B-Chat	9B	INT4 Q8 FP16	86 tok/s 62 tok/s 36 tok/s	~638 ms ~642 ms ~708 ms
-	GLM-4-9B-Chat-1M	9B	INT4 Q8 FP16	86 tok/s 62 tok/s 36 tok/s	~638 ms ~642 ms ~708 ms
-	Yi-9B	9B	INT4 Q8 FP16	86 tok/s 62 tok/s 36 tok/s	~638 ms ~642 ms ~708 ms
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8 FP16	85 tok/s 61 tok/s 35 tok/s	~651 ms ~656 ms ~723 ms
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8 FP16	85 tok/s 61 tok/s 35 tok/s	~651 ms ~656 ms ~723 ms
-	GreenMind-14B-R1	14B	INT4 Q8	62 tok/s 43 tok/s	~981 ms ~988 ms
-	Falcon-40B	40B	INT4	34 tok/s	~3.0s

NVIDIA RTX 3090 Ti (24GB)

VRAM Scaling by Context Length

Best Models to Run on NVIDIA RTX 3090 Ti (24GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions