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 AMD Ryzen AI 9 HX 370 (64GB system, 48GB GPU-accessible) (October 2026)

Q: What are the best models to run on AMD Ryzen AI 9 HX 370 (64GB system, 48GB GPU-accessible)?

Based on model rankings and 48 GB usable VRAM: Top Overall: Qwen 3.8 27B (INT4) ~6 tok/s. Top Coding: Qwen3.5-27B (INT4) ~6 tok/s. Optimal Daily Driver: Phi-4 Reasoning Plus (INT4) ~7 tok/s. Maximum Model Size: AliceAI Foundation 80B A3B Base (INT4) ~23 tok/s.

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4 Q8	6 tok/s 3 tok/s	~24.6s ~24.8s
#48	Agnes 3.0 Flash	33B	INT4 Q8	5 tok/s 3 tok/s	~29.9s ~30.1s
#72	K2 Horizon MoVA 36B A4B	36B (4B active)	INT4 Q8	27 tok/s 17 tok/s	~4.8s ~4.8s
#83	Qwen3.5-27B	27B	INT4 Q8	6 tok/s 3 tok/s	~24.6s ~24.8s
#92	Sarvam-30B	32B (2.4B active)	INT4 Q8	37 tok/s 26 tok/s	~3.4s ~3.4s
#94	Qwen3.6 35B A3B	35B (3B active)	INT4 Q8	32 tok/s 22 tok/s	~3.9s ~3.9s
#96	Muse Glimmer 30B	30B	INT4 Q8	5 tok/s 3 tok/s	~27.3s ~27.4s
#100	Phi-4 Reasoning Plus	14B	INT4 Q8 FP16	7 tok/s 5 tok/s 3 tok/s	~13.0s ~13.0s ~13.2s
#101	MiMo V2 Flash	15B (309B active)	INT4 Q8 FP16	1 tok/s 1 tok/s 1 tok/s	~269.3s ~270.9s ~274.4s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4 Q8	29 tok/s 19 tok/s	~4.4s ~4.4s
#109	Gemma 4 12B	11.95B	INT4 Q8 FP16	9 tok/s 6 tok/s 3 tok/s	~11.1s ~11.2s ~11.3s
#113	Qwen3.5-9B	9B	INT4 Q8 FP16	16 tok/s 9 tok/s 5 tok/s	~8.4s ~8.4s ~8.5s
#114	Qwen3.5-35B-A3B	35B (3B active)	INT4 Q8	32 tok/s 22 tok/s	~3.9s ~3.9s
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	20 tok/s 12 tok/s 6 tok/s	~6.6s ~6.6s ~6.7s
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	5 tok/s 3 tok/s 1 tok/s	~26.4s ~26.6s ~26.9s
#129	Gemma 4 31B	30.7B	INT4 Q8	5 tok/s 3 tok/s	~27.9s ~28.1s
#131	Nemotron 3.5 Lightning	30B (3B active)	INT4 Q8	11 tok/s 10 tok/s	~3.9s ~3.9s
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	33 tok/s 20 tok/s 10 tok/s	~3.8s ~3.9s ~3.9s
#133	Qwen3-30B-A3B	30B (3B active)	INT4 Q8	33 tok/s 22 tok/s	~3.8s ~3.9s
#136	GPT-OSS 20B	21B (3.6B active)	INT4 Q8 FP16	12 tok/s 10 tok/s 7 tok/s	~4.2s ~4.2s ~5.0s
#146	Granite 4.2 8B	8B	INT4 Q8 FP16	18 tok/s 10 tok/s 5 tok/s	~7.5s ~7.5s ~7.6s
#150	Gemma 4 E4B	8B	INT4 Q8 FP16	18 tok/s 10 tok/s 5 tok/s	~7.5s ~7.5s ~7.6s
#151	Nemotron 3.5 Lightning 30B A3B	30B (3B active)	INT4 Q8	33 tok/s 22 tok/s	~3.8s ~3.9s
#153	Qwen3-32B	32B	INT4 Q8	5 tok/s 3 tok/s	~29.1s ~29.2s
#154	Qwen2.5-72B	72B	INT4	2 tok/s	~70.6s
#155	Mistral Large 3	41B (675B active)	INT4	1 tok/s	~587.9s
#156	GLM-4	32B	INT4 Q8	4 tok/s 2 tok/s	~29.1s ~31.8s
#157	DeepSeek-R1 70B	70B	INT4	2 tok/s	~63.1s
#158	GLM-4V	9B	INT4 Q8 FP16	11 tok/s 7 tok/s 4 tok/s	~8.4s ~8.5s ~8.6s
#159	Gemma 3 27B	27B	INT4 Q8	6 tok/s 3 tok/s	~24.6s ~24.8s
#160	Phi-4	14B	INT4 Q8 FP16	10 tok/s 6 tok/s 3 tok/s	~12.9s ~13.0s ~13.2s
#161	Llama 3.3 70B	70B	INT4	2 tok/s	~63.1s
#162	DeepSeek-R1 32B	32B	INT4 Q8	5 tok/s 3 tok/s	~29.1s ~29.2s
#163	Hunyuan Standard	52B (389B active)	INT4	1 tok/s	~339.6s
#164	Qwen2.5-32B	32B	INT4 Q8	5 tok/s 3 tok/s	~29.1s ~29.2s
#168	Magistral Small	24B	INT4 Q8	5 tok/s 3 tok/s	~22.0s ~22.1s
#169	OLMo 3 7B Think	7B	INT4 Q8 FP16	14 tok/s 9 tok/s 5 tok/s	~6.6s ~6.6s ~6.7s
#170	OLMo 3.1 32B Instruct	32B	INT4 Q8	4 tok/s 2 tok/s	~29.1s ~31.8s
#171	Llama 3.1 70B	70B	INT4	2 tok/s	~63.1s
#172	DeepSeek-R1 14B	14B	INT4 Q8 FP16	11 tok/s 6 tok/s 3 tok/s	~12.9s ~13.0s ~13.2s
#173	Ministral 3 14B	14B	INT4 Q8 FP16	7 tok/s 5 tok/s 3 tok/s	~13.0s ~13.0s ~13.2s
#177	Gemma 3 12B	12B	INT4 Q8 FP16	12 tok/s 7 tok/s 4 tok/s	~11.1s ~11.2s ~11.3s
#179	Qwen2-72B	72B	INT4	2 tok/s	~70.6s
#180	OLMo 3 32B Base	32B	INT4 Q8	4 tok/s 2 tok/s	~29.1s ~31.8s
#181	Mistral-Small-2501	24B	INT4 Q8 FP16	6 tok/s 4 tok/s 2 tok/s	~21.9s ~22.1s ~26.2s
#182	Ministral 3 8B	8B	INT4 Q8 FP16	12 tok/s 8 tok/s 5 tok/s	~7.5s ~7.5s ~7.6s
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	58 tok/s 36 tok/s 20 tok/s	~2.0s ~2.0s ~2.1s
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	58 tok/s 36 tok/s 20 tok/s	~2.0s ~2.0s ~2.1s
#185	Qwen2.5-14B	14B	INT4 Q8 FP16	10 tok/s 6 tok/s 3 tok/s	~12.9s ~13.0s ~13.2s
#186	Gemma 2 27B	27B	INT4 Q8	6 tok/s 3 tok/s	~24.6s ~24.8s
#187	OLMo 3.1 32B Think	32B	INT4 Q8	4 tok/s 2 tok/s	~29.1s ~31.8s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	27 tok/s 16 tok/s 8 tok/s	~4.8s ~4.9s ~4.9s
#189	Qwen3-4B	4B	INT4 Q8 FP16	33 tok/s 20 tok/s 10 tok/s	~3.8s ~3.9s ~3.9s
#190	DeepSeek-R1 8B	8B	INT4 Q8 FP16	18 tok/s 10 tok/s 5 tok/s	~7.5s ~7.5s ~7.6s
#191	Llama 3 70B	70B	INT4	2 tok/s	~63.1s
#193	Ministral 3 3B	3B	INT4 Q8 FP16	25 tok/s 18 tok/s 11 tok/s	~3.0s ~3.0s ~3.0s
#195	Qwen3-14B	14B	INT4 Q8 FP16	10 tok/s 6 tok/s 3 tok/s	~12.9s ~13.0s ~13.2s
#196	Kimi Linear 48B A3B Instruct	48B (3B active)	INT4	9 tok/s	~4.2s
#197	Gemma 2 9B	9B	INT4 Q8 FP16	16 tok/s 9 tok/s 5 tok/s	~8.4s ~8.4s ~8.5s
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	20 tok/s 12 tok/s 6 tok/s	~6.6s ~6.6s ~6.7s
#199	Qwen3-8B	8B	INT4 Q8 FP16	18 tok/s 10 tok/s 5 tok/s	~7.5s ~7.5s ~7.6s
#200	Gemma 3 4B	4B	INT4 Q8 FP16	33 tok/s 20 tok/s 10 tok/s	~3.8s ~3.9s ~3.9s
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	21 tok/s 12 tok/s 6 tok/s	~6.6s ~6.6s ~6.7s
#202	Llama 3.1 8B	8B	INT4 Q8 FP16	18 tok/s 10 tok/s 5 tok/s	~7.5s ~7.5s ~7.6s
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	66 tok/s 42 tok/s 23 tok/s	~1.7s ~1.8s ~1.8s
#205	Command R	35B	INT4 Q8	3 tok/s 2 tok/s	~31.8s ~34.7s
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	164 tok/s 136 tok/s 95 tok/s	~273 ms ~275 ms ~278 ms
#208	Llama 3 8B	8B	INT4 Q8 FP16	18 tok/s 10 tok/s 5 tok/s	~7.5s ~7.5s ~7.6s
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	142 tok/s 99 tok/s 59 tok/s	~610 ms ~614 ms ~620 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	35 tok/s 21 tok/s 11 tok/s	~3.7s ~3.7s ~3.7s
#211	Mixtral-8x7B-v0.1	46.7B (7B active)	INT4 Q8	17 tok/s 10 tok/s	~7.6s ~9.0s
#212	Ministral-8B-2410	8B	INT4 Q8 FP16	18 tok/s 10 tok/s 5 tok/s	~7.5s ~7.5s ~7.6s
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	73 tok/s 47 tok/s 26 tok/s	~1.5s ~1.5s ~1.6s
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	42 tok/s 25 tok/s 14 tok/s	~2.9s ~3.0s ~3.0s
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	116 tok/s 79 tok/s 46 tok/s	~826 ms ~831 ms ~840 ms
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	35 tok/s 21 tok/s 11 tok/s	~3.7s ~3.7s ~3.7s
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	25 tok/s 21 tok/s 14 tok/s	~2.2s ~2.2s ~2.2s
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	19 tok/s 11 tok/s 6 tok/s	~6.8s ~6.9s ~6.9s
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	98 tok/s 65 tok/s 37 tok/s	~1.1s ~1.1s ~1.1s
#220	Gemma 2 2B	2B	INT4 Q8 FP16	58 tok/s 36 tok/s 20 tok/s	~2.0s ~2.0s ~2.1s
#221	Phi-3-medium	14B	INT4 Q8 FP16	10 tok/s 6 tok/s 3 tok/s	~12.9s ~13.0s ~13.2s
#222	OLMo 3 7B Instruct	7B	INT4 Q8 FP16	14 tok/s 9 tok/s 5 tok/s	~6.6s ~6.6s ~6.7s
#223	Yi-34B	34B	INT4 Q8	3 tok/s 2 tok/s	~30.9s ~33.8s
#224	Qwen2-7B	7B	INT4 Q8 FP16	20 tok/s 12 tok/s 6 tok/s	~6.6s ~6.6s ~6.7s
#225	Phi-3-small	7B	INT4 Q8 FP16	20 tok/s 12 tok/s 6 tok/s	~6.6s ~6.6s ~6.7s
#226	Gemma 3 1B	1B	INT4 Q8 FP16	98 tok/s 65 tok/s 37 tok/s	~1.1s ~1.1s ~1.1s
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	19 tok/s 11 tok/s 6 tok/s	~6.8s ~6.9s ~6.9s
#229	Gemma 1 7B	7B	INT4 Q8 FP16	14 tok/s 9 tok/s 5 tok/s	~6.6s ~6.6s ~6.7s
#230	Gemma 1 2B	2B	INT4 Q8 FP16	64 tok/s 38 tok/s 20 tok/s	~2.0s ~2.0s ~2.1s
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	15 tok/s 11 tok/s 6 tok/s	~5.7s ~5.7s ~5.8s
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	15 tok/s 11 tok/s 6 tok/s	~5.7s ~5.7s ~5.8s
#233	ChatGLM-6B	6B	INT4 Q8 FP16	15 tok/s 11 tok/s 6 tok/s	~5.7s ~5.7s ~5.8s
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	159 tok/s 113 tok/s 68 tok/s	~506 ms ~508 ms ~514 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	159 tok/s 113 tok/s 68 tok/s	~506 ms ~508 ms ~514 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	152 tok/s 125 tok/s 88 tok/s	~305 ms ~306 ms ~309 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	152 tok/s 125 tok/s 88 tok/s	~305 ms ~306 ms ~309 ms
-	Falcon-1B	1B	INT4 Q8 FP16	108 tok/s 69 tok/s 39 tok/s	~1.1s ~1.1s ~1.1s
-	Falcon3-1B	1B	INT4 Q8 FP16	98 tok/s 65 tok/s 37 tok/s	~1.1s ~1.1s ~1.1s
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	73 tok/s 47 tok/s 26 tok/s	~1.5s ~1.5s ~1.6s
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	73 tok/s 47 tok/s 26 tok/s	~1.5s ~1.5s ~1.6s
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	66 tok/s 42 tok/s 23 tok/s	~1.7s ~1.8s ~1.8s
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	63 tok/s 40 tok/s 22 tok/s	~1.8s ~1.8s ~1.9s
-	Falcon-3B	3B	INT4 Q8 FP16	45 tok/s 27 tok/s 14 tok/s	~2.9s ~2.9s ~3.0s
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	44 tok/s 35 tok/s 23 tok/s	~1.4s ~1.4s ~1.4s
-	Phi-1	1.3B	INT4 Q8 FP16	44 tok/s 35 tok/s 23 tok/s	~1.4s ~1.4s ~1.4s
-	Phi-1.5	1.3B	INT4 Q8 FP16	44 tok/s 35 tok/s 23 tok/s	~1.4s ~1.4s ~1.4s
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	43 tok/s 26 tok/s 14 tok/s	~2.9s ~2.9s ~3.0s
-	Falcon3-3B	3B	INT4 Q8 FP16	42 tok/s 25 tok/s 14 tok/s	~2.9s ~3.0s ~3.0s
-	Qwen2.5-3B	3B	INT4 Q8 FP16	42 tok/s 25 tok/s 14 tok/s	~2.9s ~3.0s ~3.0s
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4 Q8 FP16	35 tok/s 23 tok/s 12 tok/s	~3.6s ~3.6s ~4.0s
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4 Q8 FP16	35 tok/s 23 tok/s 12 tok/s	~3.6s ~3.6s ~4.0s
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4 Q8	34 tok/s 22 tok/s	~3.8s ~3.8s
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4 Q8	34 tok/s 22 tok/s	~3.8s ~3.8s
-	Hy-MT2 30B A3B	30B (3B active)	INT4 Q8	33 tok/s 22 tok/s	~3.8s ~3.9s
-	Spark X2.5 4B	4B	INT4 Q8 FP16	33 tok/s 20 tok/s 10 tok/s	~3.8s ~3.9s ~3.9s
-	Phi-2	2.7B	INT4 Q8 FP16	27 tok/s 20 tok/s 12 tok/s	~2.7s ~2.7s ~2.7s
-	MaLLaM-3B	3B	INT4 Q8 FP16	25 tok/s 18 tok/s 11 tok/s	~3.0s ~3.0s ~3.0s
-	SmolLM3 3B	3B	INT4 Q8 FP16	25 tok/s 18 tok/s 11 tok/s	~3.0s ~3.0s ~3.0s
-	AliceAI Foundation 80B A3B Base	80B (3B active)	INT4	23 tok/s	~5.7s
-	Falcon-7B	7B	INT4 Q8 FP16	21 tok/s 12 tok/s 6 tok/s	~6.6s ~6.6s ~6.7s
-	Falcon3-7B	7B	INT4 Q8 FP16	20 tok/s 12 tok/s 6 tok/s	~6.6s ~6.6s ~6.7s
-	Hy-MT2 7B	7B	INT4 Q8 FP16	20 tok/s 12 tok/s 6 tok/s	~6.6s ~6.6s ~6.7s
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	19 tok/s 11 tok/s 6 tok/s	~6.8s ~6.9s ~6.9s
-	LensVLM 9B	9B	INT4 Q8 FP16	16 tok/s 9 tok/s 5 tok/s	~8.4s ~8.4s ~8.5s
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8 FP16	16 tok/s 9 tok/s 5 tok/s	~8.4s ~8.4s ~8.5s
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	15 tok/s 11 tok/s 6 tok/s	~5.7s ~5.7s ~5.8s
-	Yi-6B	6B	INT4 Q8 FP16	15 tok/s 11 tok/s 6 tok/s	~5.7s ~5.7s ~5.8s
-	Falcon2-11B	11B	INT4 Q8 FP16	14 tok/s 8 tok/s 4 tok/s	~10.2s ~10.2s ~10.4s
-	Falcon3-10B	10B	INT4 Q8 FP16	14 tok/s 8 tok/s 4 tok/s	~9.3s ~9.3s ~9.5s
-	Hunyuan Lite	7B	INT4 Q8 FP16	14 tok/s 9 tok/s 5 tok/s	~6.6s ~6.6s ~6.7s
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4 Q8 FP16	14 tok/s 12 tok/s 9 tok/s	~3.5s ~3.5s ~3.5s
-	OLMo 3 7B Base	7B	INT4 Q8 FP16	14 tok/s 9 tok/s 5 tok/s	~6.6s ~6.6s ~6.7s
-	SEA-LION-7B	7.1B	INT4 Q8 FP16	14 tok/s 9 tok/s 5 tok/s	~6.7s ~6.7s ~6.8s
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8 FP16	14 tok/s 9 tok/s 5 tok/s	~6.7s ~6.7s ~6.8s
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8 FP16	12 tok/s 8 tok/s 5 tok/s	~7.5s ~7.5s ~7.6s
-	Typhoon-2-8B	8B	INT4 Q8 FP16	12 tok/s 8 tok/s 5 tok/s	~7.5s ~7.5s ~7.6s
-	GLM-4-9B	9B	INT4 Q8 FP16	11 tok/s 7 tok/s 4 tok/s	~8.4s ~8.5s ~8.6s
-	GLM-4-9B-Chat	9B	INT4 Q8 FP16	11 tok/s 7 tok/s 4 tok/s	~8.4s ~8.5s ~8.6s
-	GLM-4-9B-Chat-1M	9B	INT4 Q8 FP16	11 tok/s 7 tok/s 4 tok/s	~8.4s ~8.5s ~8.6s
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8 FP16	11 tok/s 7 tok/s 4 tok/s	~8.6s ~8.6s ~8.7s
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8 FP16	11 tok/s 7 tok/s 4 tok/s	~8.6s ~8.6s ~8.7s
-	Yi-9B	9B	INT4 Q8 FP16	11 tok/s 7 tok/s 4 tok/s	~8.4s ~8.5s ~8.6s
-	GreenMind-14B-R1	14B	INT4 Q8 FP16	7 tok/s 5 tok/s 3 tok/s	~13.0s ~13.0s ~13.2s
-	Falcon-40B	40B	INT4 Q8	4 tok/s 2 tok/s	~36.2s ~39.5s
-	Kimi-Dev-72B	72B	INT4	2 tok/s	~76.5s
-	Typhoon-2-70B	70B	INT4	2 tok/s	~74.4s

AMD Ryzen AI 9 HX 370 (64GB system, 48GB GPU-accessible)

VRAM Scaling by Context Length

Best Models to Run on AMD Ryzen AI 9 HX 370 (64GB system, 48GB GPU-accessible)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions