What are the best models to run on NVIDIA RTX PRO 4000 Blackwell SFF (24GB)?

Based on model rankings and 24 GB usable VRAM: Top Overall: Qwen 3.8 27B (INT4) ~22 tok/s. Top Coding: Qwen3.5-27B (INT4) ~22 tok/s. Optimal Daily Driver: Phi-4 Reasoning Plus (INT4) ~30 tok/s. Maximum Model Size: Falcon-40B (INT4) ~15 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 PRO 4000 Blackwell SFF (24GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4	22 tok/s	~3.1s
#48	Agnes 3.0 Flash	33B	INT4	19 tok/s	~3.7s
#72	K2 Horizon MoVA 36B A4B	36B (4B active)	INT4	87 tok/s	~659 ms
#83	Qwen3.5-27B	27B	INT4	22 tok/s	~3.1s
#92	Sarvam-30B	32B (2.4B active)	INT4	119 tok/s	~424 ms
#94	Qwen3.6 35B A3B	35B (3B active)	INT4	102 tok/s	~537 ms
#96	Muse Glimmer 30B	30B	INT4	20 tok/s	~3.4s
#100	Phi-4 Reasoning Plus	14B	INT4 Q8	30 tok/s 20 tok/s	~1.6s ~1.6s
#101	MiMo V2 Flash	15B (309B active)	INT4 Q8	2 tok/s 1 tok/s	~33.6s ~36.9s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4	98 tok/s	~555 ms
#109	Gemma 4 12B	11.95B	INT4 Q8	34 tok/s 23 tok/s	~1.4s ~1.4s
#113	Qwen3.5-9B	9B	INT4 Q8 FP16	59 tok/s 36 tok/s 18 tok/s	~1.1s ~1.1s ~1.2s
#114	Qwen3.5-35B-A3B	35B (3B active)	INT4	102 tok/s	~537 ms
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	72 tok/s 44 tok/s 24 tok/s	~825 ms ~833 ms ~852 ms
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	19 tok/s 11 tok/s 6 tok/s	~3.3s ~3.3s ~3.4s
#129	Gemma 4 31B	30.7B	INT4	20 tok/s	~3.5s
#131	Nemotron 3.5 Lightning	30B (3B active)	INT4	40 tok/s	~542 ms
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	108 tok/s 71 tok/s 40 tok/s	~486 ms ~491 ms ~502 ms
#133	Qwen3-30B-A3B	30B (3B active)	INT4	108 tok/s	~485 ms
#136	GPT-OSS 20B	21B (3.6B active)	INT4	47 tok/s	~531 ms
#146	Granite 4.2 8B	8B	INT4 Q8 FP16	65 tok/s 40 tok/s 21 tok/s	~937 ms ~947 ms ~969 ms
#150	Gemma 4 E4B	8B	INT4 Q8 FP16	65 tok/s 40 tok/s 21 tok/s	~937 ms ~947 ms ~969 ms
#151	Nemotron 3.5 Lightning 30B A3B	30B (3B active)	INT4	108 tok/s	~485 ms
#153	Qwen3-32B	32B	INT4	19 tok/s	~3.6s
#156	GLM-4	32B	INT4	13 tok/s	~4.3s
#158	GLM-4V	9B	INT4 Q8 FP16	42 tok/s 29 tok/s 16 tok/s	~1.1s ~1.1s ~1.2s
#159	Gemma 3 27B	27B	INT4	22 tok/s	~3.1s
#160	Phi-4	14B	INT4 Q8	40 tok/s 24 tok/s	~1.6s ~1.6s
#162	DeepSeek-R1 32B	32B	INT4	20 tok/s	~3.6s
#164	Qwen2.5-32B	32B	INT4	19 tok/s	~3.6s
#168	Magistral Small	24B	INT4	19 tok/s	~2.8s
#169	OLMo 3 7B Think	7B	INT4 Q8 FP16	52 tok/s 36 tok/s 21 tok/s	~830 ms ~838 ms ~857 ms
#170	OLMo 3.1 32B Instruct	32B	INT4	13 tok/s	~4.3s
#172	DeepSeek-R1 14B	14B	INT4 Q8	42 tok/s 24 tok/s	~1.6s ~1.6s
#173	Ministral 3 14B	14B	INT4 Q8	30 tok/s 20 tok/s	~1.6s ~1.6s
#177	Gemma 3 12B	12B	INT4 Q8 FP16	46 tok/s 27 tok/s 13 tok/s	~1.4s ~1.4s ~1.7s
#180	OLMo 3 32B Base	32B	INT4	13 tok/s	~4.3s
#181	Mistral-Small-2501	24B	INT4	25 tok/s	~2.7s
#182	Ministral 3 8B	8B	INT4 Q8 FP16	47 tok/s 32 tok/s 18 tok/s	~943 ms ~953 ms ~1.1s
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	162 tok/s 116 tok/s 71 tok/s	~261 ms ~263 ms ~269 ms
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	162 tok/s 116 tok/s 71 tok/s	~261 ms ~263 ms ~269 ms
#185	Qwen2.5-14B	14B	INT4 Q8	40 tok/s 24 tok/s	~1.6s ~1.6s
#186	Gemma 2 27B	27B	INT4	22 tok/s	~3.1s
#187	OLMo 3.1 32B Think	32B	INT4	13 tok/s	~4.3s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	92 tok/s 58 tok/s 32 tok/s	~610 ms ~616 ms ~630 ms
#189	Qwen3-4B	4B	INT4 Q8 FP16	108 tok/s 71 tok/s 40 tok/s	~486 ms ~491 ms ~502 ms
#190	DeepSeek-R1 8B	8B	INT4 Q8 FP16	67 tok/s 40 tok/s 21 tok/s	~937 ms ~947 ms ~969 ms
#193	Ministral 3 3B	3B	INT4 Q8 FP16	87 tok/s 67 tok/s 43 tok/s	~377 ms ~380 ms ~388 ms
#195	Qwen3-14B	14B	INT4 Q8	40 tok/s 24 tok/s	~1.6s ~1.6s
#197	Gemma 2 9B	9B	INT4 Q8 FP16	59 tok/s 36 tok/s 18 tok/s	~1.1s ~1.1s ~1.2s
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	72 tok/s 44 tok/s 24 tok/s	~825 ms ~833 ms ~852 ms
#199	Qwen3-8B	8B	INT4 Q8 FP16	65 tok/s 40 tok/s 21 tok/s	~937 ms ~947 ms ~969 ms
#200	Gemma 3 4B	4B	INT4 Q8 FP16	108 tok/s 71 tok/s 40 tok/s	~486 ms ~491 ms ~502 ms
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	74 tok/s 45 tok/s 24 tok/s	~824 ms ~833 ms ~852 ms
#202	Llama 3.1 8B	8B	INT4 Q8 FP16	65 tok/s 40 tok/s 21 tok/s	~937 ms ~947 ms ~969 ms
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	177 tok/s 129 tok/s 81 tok/s	~224 ms ~226 ms ~231 ms
#205	Command R	35B	INT4	12 tok/s	~4.7s
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	285 tok/s 263 tok/s 220 tok/s	~41 ms ~41 ms ~42 ms
#208	Llama 3 8B	8B	INT4 Q8 FP16	65 tok/s 40 tok/s 21 tok/s	~937 ms ~947 ms ~969 ms
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	268 tok/s 224 tok/s 164 tok/s	~83 ms ~84 ms ~85 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	112 tok/s 74 tok/s 42 tok/s	~463 ms ~468 ms ~478 ms
#212	Ministral-8B-2410	8B	INT4 Q8 FP16	65 tok/s 40 tok/s 21 tok/s	~937 ms ~947 ms ~969 ms
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	188 tok/s 140 tok/s 89 tok/s	~199 ms ~201 ms ~205 ms
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	130 tok/s 88 tok/s 51 tok/s	~373 ms ~377 ms ~385 ms
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	244 tok/s 197 tok/s 138 tok/s	~110 ms ~111 ms ~113 ms
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	112 tok/s 74 tok/s 42 tok/s	~463 ms ~468 ms ~478 ms
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	87 tok/s 74 tok/s 53 tok/s	~283 ms ~285 ms ~291 ms
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	70 tok/s 43 tok/s 23 tok/s	~858 ms ~867 ms ~887 ms
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	223 tok/s 176 tok/s 119 tok/s	~138 ms ~139 ms ~142 ms
#220	Gemma 2 2B	2B	INT4 Q8 FP16	162 tok/s 116 tok/s 71 tok/s	~261 ms ~263 ms ~269 ms
#221	Phi-3-medium	14B	INT4 Q8	40 tok/s 24 tok/s	~1.6s ~1.6s
#222	OLMo 3 7B Instruct	7B	INT4 Q8 FP16	52 tok/s 36 tok/s 21 tok/s	~830 ms ~838 ms ~857 ms
#223	Yi-34B	34B	INT4	13 tok/s	~4.5s
#224	Qwen2-7B	7B	INT4 Q8 FP16	72 tok/s 44 tok/s 24 tok/s	~825 ms ~833 ms ~852 ms
#225	Phi-3-small	7B	INT4 Q8 FP16	72 tok/s 44 tok/s 24 tok/s	~825 ms ~833 ms ~852 ms
#226	Gemma 3 1B	1B	INT4 Q8 FP16	223 tok/s 176 tok/s 119 tok/s	~138 ms ~139 ms ~142 ms
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	70 tok/s 43 tok/s 23 tok/s	~858 ms ~867 ms ~887 ms
#229	Gemma 1 7B	7B	INT4 Q8 FP16	52 tok/s 36 tok/s 21 tok/s	~830 ms ~838 ms ~857 ms
#230	Gemma 1 2B	2B	INT4 Q8 FP16	173 tok/s 121 tok/s 73 tok/s	~260 ms ~263 ms ~268 ms
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	57 tok/s 41 tok/s 25 tok/s	~717 ms ~724 ms ~740 ms
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	57 tok/s 41 tok/s 25 tok/s	~717 ms ~724 ms ~740 ms
#233	ChatGLM-6B	6B	INT4 Q8 FP16	57 tok/s 41 tok/s 25 tok/s	~717 ms ~724 ms ~740 ms
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	282 tok/s 240 tok/s 181 tok/s	~70 ms ~70 ms ~72 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	282 tok/s 240 tok/s 181 tok/s	~70 ms ~70 ms ~72 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	277 tok/s 253 tok/s 210 tok/s	~45 ms ~45 ms ~46 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	277 tok/s 253 tok/s 210 tok/s	~45 ms ~45 ms ~46 ms
-	Falcon-1B	1B	INT4 Q8 FP16	235 tok/s 182 tok/s 122 tok/s	~138 ms ~139 ms ~142 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	223 tok/s 176 tok/s 119 tok/s	~138 ms ~139 ms ~142 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	188 tok/s 140 tok/s 89 tok/s	~199 ms ~201 ms ~205 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	188 tok/s 140 tok/s 89 tok/s	~199 ms ~201 ms ~205 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	177 tok/s 129 tok/s 81 tok/s	~224 ms ~226 ms ~231 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	172 tok/s 125 tok/s 77 tok/s	~235 ms ~238 ms ~243 ms
-	Falcon-3B	3B	INT4 Q8 FP16	138 tok/s 91 tok/s 52 tok/s	~372 ms ~376 ms ~384 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	135 tok/s 114 tok/s 81 tok/s	~176 ms ~178 ms ~181 ms
-	Phi-1	1.3B	INT4 Q8 FP16	135 tok/s 114 tok/s 81 tok/s	~176 ms ~178 ms ~181 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	135 tok/s 114 tok/s 81 tok/s	~176 ms ~178 ms ~181 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	132 tok/s 89 tok/s 51 tok/s	~373 ms ~376 ms ~385 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	130 tok/s 88 tok/s 51 tok/s	~373 ms ~377 ms ~385 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	130 tok/s 88 tok/s 51 tok/s	~373 ms ~377 ms ~385 ms
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4 Q8	114 tok/s 73 tok/s	~454 ms ~537 ms
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4 Q8	114 tok/s 73 tok/s	~454 ms ~537 ms
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4	110 tok/s	~478 ms
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4	110 tok/s	~478 ms
-	Hy-MT2 30B A3B	30B (3B active)	INT4	108 tok/s	~485 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	108 tok/s 71 tok/s 40 tok/s	~486 ms ~491 ms ~502 ms
-	Phi-2	2.7B	INT4 Q8 FP16	91 tok/s 72 tok/s 47 tok/s	~343 ms ~346 ms ~353 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	87 tok/s 67 tok/s 43 tok/s	~377 ms ~380 ms ~388 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	87 tok/s 67 tok/s 43 tok/s	~377 ms ~380 ms ~388 ms
-	Falcon-7B	7B	INT4 Q8 FP16	75 tok/s 46 tok/s 24 tok/s	~824 ms ~833 ms ~852 ms
-	Falcon3-7B	7B	INT4 Q8 FP16	72 tok/s 44 tok/s 24 tok/s	~825 ms ~833 ms ~852 ms
-	Hy-MT2 7B	7B	INT4 Q8 FP16	72 tok/s 44 tok/s 24 tok/s	~825 ms ~833 ms ~852 ms
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	70 tok/s 43 tok/s 23 tok/s	~858 ms ~867 ms ~887 ms
-	LensVLM 9B	9B	INT4 Q8 FP16	59 tok/s 36 tok/s 18 tok/s	~1.1s ~1.1s ~1.2s
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8 FP16	59 tok/s 36 tok/s 18 tok/s	~1.1s ~1.1s ~1.2s
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	57 tok/s 41 tok/s 25 tok/s	~717 ms ~724 ms ~740 ms
-	Yi-6B	6B	INT4 Q8 FP16	57 tok/s 41 tok/s 25 tok/s	~717 ms ~724 ms ~740 ms
-	Falcon3-10B	10B	INT4 Q8 FP16	54 tok/s 32 tok/s 16 tok/s	~1.2s ~1.2s ~1.3s
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4 Q8	54 tok/s 45 tok/s	~446 ms ~487 ms
-	Falcon2-11B	11B	INT4 Q8 FP16	52 tok/s 30 tok/s 14 tok/s	~1.3s ~1.3s ~1.5s
-	Hunyuan Lite	7B	INT4 Q8 FP16	52 tok/s 36 tok/s 21 tok/s	~830 ms ~838 ms ~857 ms
-	OLMo 3 7B Base	7B	INT4 Q8 FP16	52 tok/s 36 tok/s 21 tok/s	~830 ms ~838 ms ~857 ms
-	SEA-LION-7B	7.1B	INT4 Q8 FP16	51 tok/s 36 tok/s 21 tok/s	~841 ms ~850 ms ~869 ms
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8 FP16	51 tok/s 36 tok/s 21 tok/s	~841 ms ~850 ms ~869 ms
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8 FP16	47 tok/s 32 tok/s 18 tok/s	~943 ms ~953 ms ~1.1s
-	Typhoon-2-8B	8B	INT4 Q8 FP16	47 tok/s 32 tok/s 18 tok/s	~943 ms ~953 ms ~1.1s
-	GLM-4-9B	9B	INT4 Q8 FP16	42 tok/s 29 tok/s 16 tok/s	~1.1s ~1.1s ~1.2s
-	GLM-4-9B-Chat	9B	INT4 Q8 FP16	42 tok/s 29 tok/s 16 tok/s	~1.1s ~1.1s ~1.2s
-	GLM-4-9B-Chat-1M	9B	INT4 Q8 FP16	42 tok/s 29 tok/s 16 tok/s	~1.1s ~1.1s ~1.2s
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8 FP16	42 tok/s 29 tok/s 16 tok/s	~1.1s ~1.1s ~1.2s
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8 FP16	42 tok/s 29 tok/s 16 tok/s	~1.1s ~1.1s ~1.2s
-	Yi-9B	9B	INT4 Q8 FP16	42 tok/s 29 tok/s 16 tok/s	~1.1s ~1.1s ~1.2s
-	GreenMind-14B-R1	14B	INT4 Q8	30 tok/s 20 tok/s	~1.6s ~1.6s
-	Falcon-40B	40B	INT4	15 tok/s	~4.9s

NVIDIA RTX PRO 4000 Blackwell SFF (24GB)

VRAM Scaling by Context Length

Best Models to Run on NVIDIA RTX PRO 4000 Blackwell SFF (24GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions