What are the best models to run on AMD RX 7800 XT (16GB)?

Based on model rankings and 16 GB usable VRAM: Top Overall: Qwen 3.8 27B (INT4) ~28 tok/s. Top Coding: Qwen3.5-27B (INT4) ~28 tok/s. Optimal Daily Driver: Phi-4 Reasoning Plus (INT4) ~41 tok/s. Maximum Model Size: ERNIE-4.5-VL-28B-A3B (INT4) ~126 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 AMD RX 7800 XT (16GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4	28 tok/s	~4.7s
#83	Qwen3.5-27B	27B	INT4	28 tok/s	~4.7s
#100	Phi-4 Reasoning Plus	14B	INT4	41 tok/s	~2.1s
#101	MiMo V2 Flash	15B (309B active)	INT4	3 tok/s	~47.1s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4	121 tok/s	~776 ms
#109	Gemma 4 12B	11.95B	INT4 Q8	47 tok/s 29 tok/s	~1.8s ~2.1s
#113	Qwen3.5-9B	9B	INT4 Q8	80 tok/s 49 tok/s	~1.4s ~1.4s
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	96 tok/s 61 tok/s 30 tok/s	~1.1s ~1.1s ~1.3s
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	27 tok/s 16 tok/s 8 tok/s	~4.3s ~4.3s ~4.3s
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	139 tok/s 94 tok/s 55 tok/s	~625 ms ~628 ms ~636 ms
#136	GPT-OSS 20B	21B (3.6B active)	INT4	58 tok/s	~794 ms
#146	Granite 4.2 8B	8B	INT4 Q8	87 tok/s 55 tok/s	~1.2s ~1.2s
#150	Gemma 4 E4B	8B	INT4 Q8	87 tok/s 55 tok/s	~1.2s ~1.2s
#158	GLM-4V	9B	INT4 Q8	58 tok/s 39 tok/s	~1.4s ~1.5s
#159	Gemma 3 27B	27B	INT4	28 tok/s	~4.7s
#160	Phi-4	14B	INT4 Q8	56 tok/s 30 tok/s	~2.1s ~2.5s
#169	OLMo 3 7B Think	7B	INT4 Q8	70 tok/s 50 tok/s	~1.1s ~1.1s
#172	DeepSeek-R1 14B	14B	INT4 Q8	57 tok/s 31 tok/s	~2.1s ~2.5s
#173	Ministral 3 14B	14B	INT4	41 tok/s	~2.1s
#177	Gemma 3 12B	12B	INT4 Q8	63 tok/s 36 tok/s	~1.8s ~2.0s
#181	Mistral-Small-2501	24B	INT4	33 tok/s	~3.8s
#182	Ministral 3 8B	8B	INT4 Q8	64 tok/s 45 tok/s	~1.2s ~1.2s
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	196 tok/s 147 tok/s 95 tok/s	~334 ms ~336 ms ~340 ms
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	196 tok/s 147 tok/s 95 tok/s	~334 ms ~336 ms ~340 ms
#185	Qwen2.5-14B	14B	INT4 Q8	56 tok/s 30 tok/s	~2.1s ~2.5s
#186	Gemma 2 27B	27B	INT4	28 tok/s	~4.7s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	119 tok/s 78 tok/s 45 tok/s	~785 ms ~789 ms ~799 ms
#189	Qwen3-4B	4B	INT4 Q8 FP16	139 tok/s 94 tok/s 55 tok/s	~625 ms ~628 ms ~636 ms
#190	DeepSeek-R1 8B	8B	INT4 Q8	89 tok/s 55 tok/s	~1.2s ~1.2s
#193	Ministral 3 3B	3B	INT4 Q8 FP16	114 tok/s 90 tok/s 60 tok/s	~482 ms ~484 ms ~490 ms
#195	Qwen3-14B	14B	INT4 Q8	56 tok/s 30 tok/s	~2.1s ~2.5s
#197	Gemma 2 9B	9B	INT4 Q8	80 tok/s 49 tok/s	~1.4s ~1.4s
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	96 tok/s 61 tok/s 30 tok/s	~1.1s ~1.1s ~1.3s
#199	Qwen3-8B	8B	INT4 Q8	87 tok/s 55 tok/s	~1.2s ~1.2s
#200	Gemma 3 4B	4B	INT4 Q8 FP16	139 tok/s 94 tok/s 55 tok/s	~625 ms ~628 ms ~636 ms
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	98 tok/s 62 tok/s 32 tok/s	~1.1s ~1.1s ~1.2s
#202	Llama 3.1 8B	8B	INT4 Q8	87 tok/s 55 tok/s	~1.2s ~1.2s
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	211 tok/s 162 tok/s 107 tok/s	~287 ms ~288 ms ~291 ms
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	307 tok/s 288 tok/s 252 tok/s	~51 ms ~51 ms ~51 ms
#208	Llama 3 8B	8B	INT4 Q8	87 tok/s 55 tok/s	~1.2s ~1.2s
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	293 tok/s 255 tok/s 198 tok/s	~105 ms ~105 ms ~106 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	143 tok/s 98 tok/s 57 tok/s	~595 ms ~599 ms ~606 ms
#212	Ministral-8B-2410	8B	INT4 Q8	87 tok/s 55 tok/s	~1.2s ~1.2s
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	222 tok/s 173 tok/s 116 tok/s	~254 ms ~256 ms ~258 ms
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	163 tok/s 115 tok/s 70 tok/s	~479 ms ~481 ms ~487 ms
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	273 tok/s 230 tok/s 171 tok/s	~140 ms ~140 ms ~142 ms
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	143 tok/s 98 tok/s 57 tok/s	~595 ms ~599 ms ~606 ms
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	114 tok/s 98 tok/s 68 tok/s	~360 ms ~361 ms ~396 ms
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	93 tok/s 59 tok/s 29 tok/s	~1.1s ~1.1s ~1.3s
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	254 tok/s 209 tok/s 150 tok/s	~176 ms ~177 ms ~178 ms
#220	Gemma 2 2B	2B	INT4 Q8 FP16	196 tok/s 147 tok/s 95 tok/s	~334 ms ~336 ms ~340 ms
#221	Phi-3-medium	14B	INT4 Q8	56 tok/s 30 tok/s	~2.1s ~2.5s
#222	OLMo 3 7B Instruct	7B	INT4 Q8	70 tok/s 50 tok/s	~1.1s ~1.1s
#224	Qwen2-7B	7B	INT4 Q8 FP16	96 tok/s 61 tok/s 30 tok/s	~1.1s ~1.1s ~1.3s
#225	Phi-3-small	7B	INT4 Q8 FP16	96 tok/s 61 tok/s 30 tok/s	~1.1s ~1.1s ~1.3s
#226	Gemma 3 1B	1B	INT4 Q8 FP16	254 tok/s 209 tok/s 150 tok/s	~176 ms ~177 ms ~178 ms
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	93 tok/s 59 tok/s 29 tok/s	~1.1s ~1.1s ~1.3s
#229	Gemma 1 7B	7B	INT4 Q8	70 tok/s 50 tok/s	~1.1s ~1.1s
#230	Gemma 1 2B	2B	INT4 Q8 FP16	207 tok/s 153 tok/s 97 tok/s	~334 ms ~336 ms ~339 ms
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	77 tok/s 56 tok/s 33 tok/s	~920 ms ~924 ms ~1.0s
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	77 tok/s 56 tok/s 33 tok/s	~920 ms ~924 ms ~1.0s
#233	ChatGLM-6B	6B	INT4 Q8 FP16	77 tok/s 56 tok/s 33 tok/s	~920 ms ~924 ms ~1.0s
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	304 tok/s 269 tok/s 215 tok/s	~88 ms ~88 ms ~89 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	304 tok/s 269 tok/s 215 tok/s	~88 ms ~88 ms ~89 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	300 tok/s 280 tok/s 242 tok/s	~55 ms ~56 ms ~56 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	300 tok/s 280 tok/s 242 tok/s	~55 ms ~56 ms ~56 ms
-	Falcon-1B	1B	INT4 Q8 FP16	265 tok/s 216 tok/s 153 tok/s	~176 ms ~176 ms ~178 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	254 tok/s 209 tok/s 150 tok/s	~176 ms ~177 ms ~178 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	222 tok/s 173 tok/s 116 tok/s	~254 ms ~256 ms ~258 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	222 tok/s 173 tok/s 116 tok/s	~254 ms ~256 ms ~258 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	211 tok/s 162 tok/s 107 tok/s	~287 ms ~288 ms ~291 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	206 tok/s 157 tok/s 102 tok/s	~302 ms ~303 ms ~307 ms
-	Falcon-3B	3B	INT4 Q8 FP16	171 tok/s 118 tok/s 71 tok/s	~479 ms ~481 ms ~487 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	168 tok/s 145 tok/s 107 tok/s	~224 ms ~225 ms ~227 ms
-	Phi-1	1.3B	INT4 Q8 FP16	168 tok/s 145 tok/s 107 tok/s	~224 ms ~225 ms ~227 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	168 tok/s 145 tok/s 107 tok/s	~224 ms ~225 ms ~227 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	164 tok/s 116 tok/s 70 tok/s	~479 ms ~481 ms ~487 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	163 tok/s 115 tok/s 70 tok/s	~479 ms ~481 ms ~487 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	163 tok/s 115 tok/s 70 tok/s	~479 ms ~481 ms ~487 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	139 tok/s 94 tok/s 55 tok/s	~625 ms ~628 ms ~636 ms
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4	137 tok/s	~634 ms
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4	137 tok/s	~634 ms
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4	126 tok/s	~721 ms
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4	126 tok/s	~721 ms
-	Phi-2	2.7B	INT4 Q8 FP16	119 tok/s 95 tok/s 64 tok/s	~438 ms ~440 ms ~445 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	114 tok/s 90 tok/s 60 tok/s	~482 ms ~484 ms ~490 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	114 tok/s 90 tok/s 60 tok/s	~482 ms ~484 ms ~490 ms
-	Falcon-7B	7B	INT4 Q8 FP16	100 tok/s 62 tok/s 33 tok/s	~1.1s ~1.1s ~1.2s
-	Falcon3-7B	7B	INT4 Q8 FP16	96 tok/s 61 tok/s 30 tok/s	~1.1s ~1.1s ~1.3s
-	Hy-MT2 7B	7B	INT4 Q8 FP16	96 tok/s 61 tok/s 30 tok/s	~1.1s ~1.1s ~1.3s
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	93 tok/s 59 tok/s 29 tok/s	~1.1s ~1.1s ~1.3s
-	LensVLM 9B	9B	INT4 Q8	80 tok/s 49 tok/s	~1.4s ~1.4s
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8	80 tok/s 49 tok/s	~1.4s ~1.4s
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	77 tok/s 56 tok/s 33 tok/s	~920 ms ~924 ms ~1.0s
-	Yi-6B	6B	INT4 Q8 FP16	77 tok/s 56 tok/s 33 tok/s	~920 ms ~924 ms ~1.0s
-	Falcon3-10B	10B	INT4 Q8	73 tok/s 45 tok/s	~1.5s ~1.5s
-	Falcon2-11B	11B	INT4 Q8	71 tok/s 42 tok/s	~1.6s ~1.7s
-	Hunyuan Lite	7B	INT4 Q8	70 tok/s 50 tok/s	~1.1s ~1.1s
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4	70 tok/s	~616 ms
-	OLMo 3 7B Base	7B	INT4 Q8	70 tok/s 50 tok/s	~1.1s ~1.1s
-	SEA-LION-7B	7.1B	INT4 Q8	69 tok/s 50 tok/s	~1.1s ~1.1s
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8	69 tok/s 50 tok/s	~1.1s ~1.1s
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8	64 tok/s 45 tok/s	~1.2s ~1.2s
-	Typhoon-2-8B	8B	INT4 Q8	64 tok/s 45 tok/s	~1.2s ~1.2s
-	GLM-4-9B	9B	INT4 Q8	58 tok/s 39 tok/s	~1.4s ~1.5s
-	GLM-4-9B-Chat	9B	INT4 Q8	58 tok/s 39 tok/s	~1.4s ~1.5s
-	GLM-4-9B-Chat-1M	9B	INT4 Q8	58 tok/s 39 tok/s	~1.4s ~1.5s
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8	58 tok/s 38 tok/s	~1.4s ~1.5s
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8	58 tok/s 38 tok/s	~1.4s ~1.5s
-	Yi-9B	9B	INT4 Q8	58 tok/s 39 tok/s	~1.4s ~1.5s
-	GreenMind-14B-R1	14B	INT4	41 tok/s	~2.1s

AMD RX 7800 XT (16GB)

VRAM Scaling by Context Length

Best Models to Run on AMD RX 7800 XT (16GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions