What are the best models to run on NVIDIA A30 (24GB)?

Based on model rankings and 24 GB usable VRAM: Top Overall: Qwen 3.8 27B (INT4) ~45 tok/s. Top Coding: Qwen3.5-27B (INT4) ~45 tok/s. Optimal Daily Driver: Phi-4 Reasoning Plus (INT4) ~59 tok/s. Maximum Model Size: Falcon-40B (INT4) ~31 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 A30 (24GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4	45 tok/s	~2.0s
#48	Agnes 3.0 Flash	33B	INT4	38 tok/s	~2.5s
#72	K2 Horizon MoVA 36B A4B	36B (4B active)	INT4	172 tok/s	~435 ms
#83	Qwen3.5-27B	27B	INT4	45 tok/s	~2.0s
#92	Sarvam-30B	32B (2.4B active)	INT4	232 tok/s	~280 ms
#94	Qwen3.6 35B A3B	35B (3B active)	INT4	199 tok/s	~354 ms
#96	Muse Glimmer 30B	30B	INT4	41 tok/s	~2.3s
#100	Phi-4 Reasoning Plus	14B	INT4 Q8	59 tok/s 40 tok/s	~1.1s ~1.1s
#101	MiMo V2 Flash	15B (309B active)	INT4 Q8	4 tok/s 2 tok/s	~22.2s ~24.4s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4	193 tok/s	~366 ms
#109	Gemma 4 12B	11.95B	INT4 Q8	68 tok/s 46 tok/s	~922 ms ~928 ms
#113	Qwen3.5-9B	9B	INT4 Q8 FP16	117 tok/s 72 tok/s 36 tok/s	~695 ms ~700 ms ~773 ms
#114	Qwen3.5-35B-A3B	35B (3B active)	INT4	199 tok/s	~354 ms
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	143 tok/s 89 tok/s 49 tok/s	~544 ms ~548 ms ~558 ms
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	39 tok/s 23 tok/s 12 tok/s	~2.2s ~2.2s ~2.2s
#129	Gemma 4 31B	30.7B	INT4	40 tok/s	~2.3s
#131	Nemotron 3.5 Lightning	30B (3B active)	INT4	80 tok/s	~355 ms
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	212 tok/s 140 tok/s 80 tok/s	~320 ms ~322 ms ~328 ms
#133	Qwen3-30B-A3B	30B (3B active)	INT4	212 tok/s	~320 ms
#136	GPT-OSS 20B	21B (3.6B active)	INT4	94 tok/s	~348 ms
#146	Granite 4.2 8B	8B	INT4 Q8 FP16	129 tok/s 79 tok/s 43 tok/s	~619 ms ~624 ms ~634 ms
#150	Gemma 4 E4B	8B	INT4 Q8 FP16	129 tok/s 79 tok/s 43 tok/s	~619 ms ~624 ms ~634 ms
#151	Nemotron 3.5 Lightning 30B A3B	30B (3B active)	INT4	212 tok/s	~320 ms
#153	Qwen3-32B	32B	INT4	39 tok/s	~2.4s
#156	GLM-4	32B	INT4	27 tok/s	~2.8s
#158	GLM-4V	9B	INT4 Q8 FP16	85 tok/s 59 tok/s 33 tok/s	~698 ms ~703 ms ~776 ms
#159	Gemma 3 27B	27B	INT4	45 tok/s	~2.0s
#160	Phi-4	14B	INT4 Q8	81 tok/s 48 tok/s	~1.1s ~1.1s
#162	DeepSeek-R1 32B	32B	INT4	40 tok/s	~2.4s
#164	Qwen2.5-32B	32B	INT4	39 tok/s	~2.4s
#168	Magistral Small	24B	INT4	38 tok/s	~1.8s
#169	OLMo 3 7B Think	7B	INT4 Q8 FP16	103 tok/s 73 tok/s 43 tok/s	~547 ms ~551 ms ~560 ms
#170	OLMo 3.1 32B Instruct	32B	INT4	27 tok/s	~2.8s
#172	DeepSeek-R1 14B	14B	INT4 Q8	84 tok/s 49 tok/s	~1.1s ~1.1s
#173	Ministral 3 14B	14B	INT4 Q8	59 tok/s 40 tok/s	~1.1s ~1.1s
#177	Gemma 3 12B	12B	INT4 Q8 FP16	92 tok/s 55 tok/s 26 tok/s	~921 ms ~928 ms ~1.1s
#180	OLMo 3 32B Base	32B	INT4	27 tok/s	~2.8s
#181	Mistral-Small-2501	24B	INT4	50 tok/s	~1.8s
#182	Ministral 3 8B	8B	INT4 Q8 FP16	94 tok/s 65 tok/s 37 tok/s	~622 ms ~626 ms ~691 ms
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	312 tok/s 227 tok/s 141 tok/s	~171 ms ~172 ms ~175 ms
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	312 tok/s 227 tok/s 141 tok/s	~171 ms ~172 ms ~175 ms
#185	Qwen2.5-14B	14B	INT4 Q8	81 tok/s 48 tok/s	~1.1s ~1.1s
#186	Gemma 2 27B	27B	INT4	45 tok/s	~2.0s
#187	OLMo 3.1 32B Think	32B	INT4	27 tok/s	~2.8s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	180 tok/s 116 tok/s 65 tok/s	~402 ms ~405 ms ~412 ms
#189	Qwen3-4B	4B	INT4 Q8 FP16	212 tok/s 140 tok/s 80 tok/s	~320 ms ~322 ms ~328 ms
#190	DeepSeek-R1 8B	8B	INT4 Q8 FP16	132 tok/s 80 tok/s 43 tok/s	~619 ms ~624 ms ~634 ms
#193	Ministral 3 3B	3B	INT4 Q8 FP16	171 tok/s 133 tok/s 87 tok/s	~247 ms ~249 ms ~253 ms
#195	Qwen3-14B	14B	INT4 Q8	81 tok/s 48 tok/s	~1.1s ~1.1s
#197	Gemma 2 9B	9B	INT4 Q8 FP16	117 tok/s 72 tok/s 36 tok/s	~695 ms ~700 ms ~773 ms
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	143 tok/s 89 tok/s 49 tok/s	~544 ms ~548 ms ~558 ms
#199	Qwen3-8B	8B	INT4 Q8 FP16	129 tok/s 79 tok/s 43 tok/s	~619 ms ~624 ms ~634 ms
#200	Gemma 3 4B	4B	INT4 Q8 FP16	212 tok/s 140 tok/s 80 tok/s	~320 ms ~322 ms ~328 ms
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	147 tok/s 90 tok/s 49 tok/s	~544 ms ~548 ms ~558 ms
#202	Llama 3.1 8B	8B	INT4 Q8 FP16	129 tok/s 79 tok/s 43 tok/s	~619 ms ~624 ms ~634 ms
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	339 tok/s 252 tok/s 160 tok/s	~147 ms ~148 ms ~150 ms
#205	Command R	35B	INT4	25 tok/s	~3.1s
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	528 tok/s 489 tok/s 416 tok/s	~26 ms ~26 ms ~26 ms
#208	Llama 3 8B	8B	INT4 Q8 FP16	129 tok/s 79 tok/s 43 tok/s	~619 ms ~624 ms ~634 ms
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	499 tok/s 423 tok/s 315 tok/s	~53 ms ~54 ms ~54 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	219 tok/s 146 tok/s 84 tok/s	~305 ms ~307 ms ~312 ms
#212	Ministral-8B-2410	8B	INT4 Q8 FP16	129 tok/s 79 tok/s 43 tok/s	~619 ms ~624 ms ~634 ms
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	359 tok/s 271 tok/s 176 tok/s	~130 ms ~131 ms ~133 ms
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	253 tok/s 173 tok/s 102 tok/s	~245 ms ~247 ms ~251 ms
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	457 tok/s 375 tok/s 267 tok/s	~71 ms ~72 ms ~73 ms
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	219 tok/s 146 tok/s 84 tok/s	~305 ms ~307 ms ~312 ms
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	171 tok/s 146 tok/s 105 tok/s	~185 ms ~186 ms ~189 ms
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	139 tok/s 86 tok/s 47 tok/s	~566 ms ~571 ms ~580 ms
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	421 tok/s 336 tok/s 232 tok/s	~90 ms ~90 ms ~92 ms
#220	Gemma 2 2B	2B	INT4 Q8 FP16	312 tok/s 227 tok/s 141 tok/s	~171 ms ~172 ms ~175 ms
#221	Phi-3-medium	14B	INT4 Q8	81 tok/s 48 tok/s	~1.1s ~1.1s
#222	OLMo 3 7B Instruct	7B	INT4 Q8 FP16	103 tok/s 73 tok/s 43 tok/s	~547 ms ~551 ms ~560 ms
#223	Yi-34B	34B	INT4	26 tok/s	~3.0s
#224	Qwen2-7B	7B	INT4 Q8 FP16	143 tok/s 89 tok/s 49 tok/s	~544 ms ~548 ms ~558 ms
#225	Phi-3-small	7B	INT4 Q8 FP16	143 tok/s 89 tok/s 49 tok/s	~544 ms ~548 ms ~558 ms
#226	Gemma 3 1B	1B	INT4 Q8 FP16	421 tok/s 336 tok/s 232 tok/s	~90 ms ~90 ms ~92 ms
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	139 tok/s 86 tok/s 47 tok/s	~566 ms ~571 ms ~580 ms
#229	Gemma 1 7B	7B	INT4 Q8 FP16	103 tok/s 73 tok/s 43 tok/s	~547 ms ~551 ms ~560 ms
#230	Gemma 1 2B	2B	INT4 Q8 FP16	332 tok/s 236 tok/s 145 tok/s	~171 ms ~172 ms ~175 ms
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	114 tok/s 82 tok/s 49 tok/s	~472 ms ~475 ms ~483 ms
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	114 tok/s 82 tok/s 49 tok/s	~472 ms ~475 ms ~483 ms
#233	ChatGLM-6B	6B	INT4 Q8 FP16	114 tok/s 82 tok/s 49 tok/s	~472 ms ~475 ms ~483 ms
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	522 tok/s 451 tok/s 346 tok/s	~45 ms ~45 ms ~46 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	522 tok/s 451 tok/s 346 tok/s	~45 ms ~45 ms ~46 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	513 tok/s 473 tok/s 397 tok/s	~28 ms ~28 ms ~29 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	513 tok/s 473 tok/s 397 tok/s	~28 ms ~28 ms ~29 ms
-	Falcon-1B	1B	INT4 Q8 FP16	442 tok/s 348 tok/s 237 tok/s	~90 ms ~90 ms ~92 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	421 tok/s 336 tok/s 232 tok/s	~90 ms ~90 ms ~92 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	359 tok/s 271 tok/s 176 tok/s	~130 ms ~131 ms ~133 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	359 tok/s 271 tok/s 176 tok/s	~130 ms ~131 ms ~133 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	339 tok/s 252 tok/s 160 tok/s	~147 ms ~148 ms ~150 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	330 tok/s 243 tok/s 153 tok/s	~154 ms ~155 ms ~158 ms
-	Falcon-3B	3B	INT4 Q8 FP16	267 tok/s 179 tok/s 104 tok/s	~245 ms ~247 ms ~251 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	262 tok/s 222 tok/s 160 tok/s	~115 ms ~115 ms ~117 ms
-	Phi-1	1.3B	INT4 Q8 FP16	262 tok/s 222 tok/s 160 tok/s	~115 ms ~115 ms ~117 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	262 tok/s 222 tok/s 160 tok/s	~115 ms ~115 ms ~117 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	256 tok/s 175 tok/s 103 tok/s	~245 ms ~247 ms ~251 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	253 tok/s 173 tok/s 102 tok/s	~245 ms ~247 ms ~251 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	253 tok/s 173 tok/s 102 tok/s	~245 ms ~247 ms ~251 ms
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4 Q8	222 tok/s 143 tok/s	~299 ms ~354 ms
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4 Q8	222 tok/s 143 tok/s	~299 ms ~354 ms
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4	215 tok/s	~315 ms
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4	215 tok/s	~315 ms
-	Hy-MT2 30B A3B	30B (3B active)	INT4	212 tok/s	~320 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	212 tok/s 140 tok/s 80 tok/s	~320 ms ~322 ms ~328 ms
-	Phi-2	2.7B	INT4 Q8 FP16	180 tok/s 142 tok/s 94 tok/s	~225 ms ~226 ms ~230 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	171 tok/s 133 tok/s 87 tok/s	~247 ms ~249 ms ~253 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	171 tok/s 133 tok/s 87 tok/s	~247 ms ~249 ms ~253 ms
-	Falcon-7B	7B	INT4 Q8 FP16	149 tok/s 91 tok/s 49 tok/s	~544 ms ~548 ms ~558 ms
-	Falcon3-7B	7B	INT4 Q8 FP16	143 tok/s 89 tok/s 49 tok/s	~544 ms ~548 ms ~558 ms
-	Hy-MT2 7B	7B	INT4 Q8 FP16	143 tok/s 89 tok/s 49 tok/s	~544 ms ~548 ms ~558 ms
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	139 tok/s 86 tok/s 47 tok/s	~566 ms ~571 ms ~580 ms
-	LensVLM 9B	9B	INT4 Q8 FP16	117 tok/s 72 tok/s 36 tok/s	~695 ms ~700 ms ~773 ms
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8 FP16	117 tok/s 72 tok/s 36 tok/s	~695 ms ~700 ms ~773 ms
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	114 tok/s 82 tok/s 49 tok/s	~472 ms ~475 ms ~483 ms
-	Yi-6B	6B	INT4 Q8 FP16	114 tok/s 82 tok/s 49 tok/s	~472 ms ~475 ms ~483 ms
-	Falcon3-10B	10B	INT4 Q8 FP16	108 tok/s 65 tok/s 33 tok/s	~770 ms ~776 ms ~857 ms
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4 Q8	108 tok/s 89 tok/s	~292 ms ~319 ms
-	Falcon2-11B	11B	INT4 Q8 FP16	104 tok/s 61 tok/s 29 tok/s	~844 ms ~851 ms ~1.0s
-	Hunyuan Lite	7B	INT4 Q8 FP16	103 tok/s 73 tok/s 43 tok/s	~547 ms ~551 ms ~560 ms
-	OLMo 3 7B Base	7B	INT4 Q8 FP16	103 tok/s 73 tok/s 43 tok/s	~547 ms ~551 ms ~560 ms
-	SEA-LION-7B	7.1B	INT4 Q8 FP16	102 tok/s 72 tok/s 43 tok/s	~554 ms ~558 ms ~568 ms
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8 FP16	102 tok/s 72 tok/s 43 tok/s	~554 ms ~558 ms ~568 ms
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8 FP16	94 tok/s 65 tok/s 37 tok/s	~622 ms ~626 ms ~691 ms
-	Typhoon-2-8B	8B	INT4 Q8 FP16	94 tok/s 65 tok/s 37 tok/s	~622 ms ~626 ms ~691 ms
-	GLM-4-9B	9B	INT4 Q8 FP16	85 tok/s 59 tok/s 33 tok/s	~698 ms ~703 ms ~776 ms
-	GLM-4-9B-Chat	9B	INT4 Q8 FP16	85 tok/s 59 tok/s 33 tok/s	~698 ms ~703 ms ~776 ms
-	GLM-4-9B-Chat-1M	9B	INT4 Q8 FP16	85 tok/s 59 tok/s 33 tok/s	~698 ms ~703 ms ~776 ms
-	Yi-9B	9B	INT4 Q8 FP16	85 tok/s 59 tok/s 33 tok/s	~698 ms ~703 ms ~776 ms
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8 FP16	84 tok/s 58 tok/s 32 tok/s	~713 ms ~718 ms ~793 ms
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8 FP16	84 tok/s 58 tok/s 32 tok/s	~713 ms ~718 ms ~793 ms
-	GreenMind-14B-R1	14B	INT4 Q8	59 tok/s 40 tok/s	~1.1s ~1.1s
-	Falcon-40B	40B	INT4	31 tok/s	~3.2s

NVIDIA A30 (24GB)

VRAM Scaling by Context Length

Best Models to Run on NVIDIA A30 (24GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions