Hardware · Rockchip

What AI models can the Rockchip RK3588 / RK3588S (16 GB) run?

RAM: 16.0 GB
Usable for models: 13.6 GB
Memory bandwidth: 34 GB/s
Runs models on: CPU cores
Released: 2022

397 / 622 LLMs run Best: K2 Horizon 7B 54 / 61 Image models run Best: FLUX 3 Image Klein 15 / 43 Video models run Best: LTX-Video 2B 65 / 70 Decision models run Best: OpenAI Privacy Filter

Models run on the CPU cores here, so the memory’s speed and how fast those cores can read it set the pace: small models (about 1-4B) are usable, bigger ones are slow. We plan around 13.6 GB of the 16.0 GB (what Linux leaves free for a model). 6 TOPS NPU (three cores). llama.cpp and Ollama use the CPU. Rockchip’s RKLLM toolkit (and front ends such as RKLLama) runs converted 8-bit models on the NPU; generation is still limited by memory speed, so it isn’t much faster than the CPU, but it leaves the CPU free.

More about this hardware

Maker	Rockchip
Type	Single-board computer or edge AI board
Share of memory for models	85% (what Linux typically leaves free on a board this size)
Chip	Rockchip RK3588
AI accelerator	6 TOPS NPU (three cores). llama.cpp and Ollama use the CPU. Rockchip’s RKLLM toolkit (and front ends such as RKLLama) runs converted 8-bit models on the NPU; generation is still limited by memory speed, so it isn’t much faster than the CPU, but it leaves the CPU free.
Boards with this chip	Orange Pi 5 / 5 Plus / 5 Max / 5 Ultra · Radxa ROCK 5B / 5B+ / 5 ITX · FriendlyElec NanoPC-T6 · Khadas Edge2 · Turing RK1 · Vicharak Axon · KICKPI K8C · Seeed reComputer RK3588
Memory type	LPDDR4X-4224, 64-bit (boards with LPDDR5 reach about 44 GB/s)
Bandwidth figure	33.8 GB/s, the chip’s rated memory speed
Note	4× Cortex-A76 at 2.4 GHz plus 4× Cortex-A55, Mali-G610 GPU. Speeds assume llama.cpp on the CPU: Turing Pi measured 11.25 tokens/s for Llama 3.2 3B at Q4_K_M on an RK3588, about 60% of the memory’s rated speed. The RK3588S is the same chip with fewer I/O lanes. Use a fan or a large heatsink.
Specifications	www.orangepi.org

Apps that run models well on the Rockchip RK3588 / RK3588S (16 GB)

Linux on ARM boards

For a typical model that fits in memory. Each model page picks the best app for that model on this hardware. All apps

3 products with the Rockchip RK3588 / RK3588S (16 GB)

Orange Pi 5 PlusOrange Pi · Board · 4, 8, 16 or 32 GBProduct page (opens in a new tab)
ROCK 5B+Radxa · Board · 4, 8, 16, 24 or 32 GBProduct page (opens in a new tab)
reComputer RK3588-30Seeed Studio · Board · 8 or 16 GB · $279–$349Product page (opens in a new tab)

Best LLMs for the Rockchip RK3588 / RK3588S (16 GB)

Quality	Model	Quant	Memory	Speed	Runs
77Great quality (A) · 77 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking.	K2 Horizon 7BInstitute of Foundation Models	`Q5_K_M`	9.3 GB	3.2 tok/s	Slow
66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking.	Mellum2.1 Thinking (12B-A2.5B)JetBrains	`Q4_K_M`	9.2 GB	13.2 tok/s	Runs well
58Good quality (B) · 58 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking.	Maple Preview (20B-A1B)DeepGrove	`Q3_K_M`	11.7 GB	27.3 tok/s	Runs welltight fit
75Great quality (A) · 75 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking.	MiMo V2.6 Distill Qwen 9BXiaomi	`Q5_K_M`	8.6 GB	3.0 tok/s	Slow
64Good quality (B) · 64 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking.	Gemma 4 26B-A4BGoogle	`Q2_K`	12.4 GB	14.3 tok/s	Runs welltight fit
74Great quality (A) · 74 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking.	SenseNova Loop 9B PreviewSenseTime	`Q4_K_M`	8.1 GB	3.6 tok/s	Slow
73Great quality (A) · 73 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking.	Ornith 1.5 9BDeepReinforce	`Q4_K_M`	7.8 GB	3.6 tok/s	Slow
71Great quality (A) · 71 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking.	TMax 9BAi2	`Q5_K_M`	8.3 GB	3.2 tok/s	Slow
72Great quality (A) · 72 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking.	Qwen3.5 9BAlibaba	`Q4_K_M`	7.8 GB	3.6 tok/s	Slow
61Good quality (B) · 61 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking.	Raon-1 mini Preview (30B-A4B)KRAFTON	`Q2_K`	13.3 GB	12.7 tok/s	Runs welltight fit
69Good quality (B) · 69 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking.	Gemma 4 12BGoogle	`Q3_K_M`	8.2 GB	3.7 tok/s	Slow
71Great quality (A) · 71 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking.	Ornith 1.0 9BDeepReinforce	`Q4_K_M`	7.8 GB	3.6 tok/s	Slow

Best Image models for the Rockchip RK3588 / RK3588S (16 GB)

Quality	Model	Quant	Memory	Speed	Runs
90Excellent quality (A+) · 90 of 100: how capable the model is at Q8 GGUF, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking.	FLUX 3 Image KleinBlack Forest Labs	`Q8 GGUF`	12.8 GB	16.0 min	Very slowtight fit
76Great quality (A) · 76 of 100: how capable the model is at FP8, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking.	FLUX.2 [klein] 4BBlack Forest Labs	`FP8`	10.2 GB	2.2 min	Slow
83Great quality (A) · 83 of 100: how capable the model is at Q8 GGUF, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking.	Z-Image TurboAlibaba (Tongyi)	`Q8 GGUF`	13.0 GB	7.0 min	Very slowtight fit
79Great quality (A) · 79 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking.	ERNIE-Image TurboBaidu	`Q4 GGUF`	11.3 GB	9.4 min	Very slow
81Great quality (A) · 81 of 100: how capable the model is at Q8 GGUF, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking.	Z-ImageAlibaba (Tongyi)	`Q8 GGUF`	13.0 GB	47.8 min	Very slowtight fit
87Excellent quality (A+) · 87 of 100: how capable the model is at Q8 GGUF, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking.	ERNIE-ImageBaidu	`Q8 GGUF`	11.2 GB	64.9 min	Very slowencoder unloaded after use

Best Video models for the Rockchip RK3588 / RK3588S (16 GB)

Quality	Model	Quant	Memory	Speed	Runs
49Fair quality (C) · 49 of 100: how capable the model is at FP8, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking.	LTX-Video 2BLightricks	`FP8`	10.8 GB	55.0 min	Very slow
83Great quality (A) · 83 of 100: how capable the model is at FP8, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking.	Bernini-R A14B (video generation + editing)ByteDance	`FP8`	10.7 GB	34.3 h	Very slowCPU offload
53Fair quality (C) · 53 of 100: how capable the model is at FP8, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking.	Wan 2.1 T2V 1.3BAlibaba	`FP8`	12.8 GB	8.8 h	Very slowtight fit
44Fair quality (C) · 44 of 100: how capable the model is at FP8, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking.	CogVideoX 2BZhipu AI	`FP8`	11.0 GB	2.2 h	Very slow
76Great quality (A) · 76 of 100: how capable the model is at FP16, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking.	HY-OmniWeaving (8.3B, unified video)Tencent	`FP16`	10.7 GB	28.1 h	Very slowCPU offload

Best Decision models for the Rockchip RK3588 / RK3588S (16 GB)

Quality	Model	Quant	Memory	Speed	Runs
78Great quality (A) · 78 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking.	OpenAI Privacy FilterOpenAI	`Q8_0`	2.4 GB	120 ms	Runs great
86Excellent quality (A+) · 86 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking.	Jina Reranker v3.5Jina AI	`Q8_0`	2.6 GB	2.6 s	Very slow
90Excellent quality (A+) · 90 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking.	Qwen3 Reranker 4BAlibaba	`Q8_0`	6.3 GB	17.4 s	Very slow
90Excellent quality (A+) · 90 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking.	Qwen3 Reranker 8BAlibaba	`Q8_0`	10.7 GB	35.4 s	Very slow
90Excellent quality (A+) · 90 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking.	Skywork Reward V2 Llama 3.1 8BSkywork	`Q8_0`	10.4 GB	69.6 s	Very slow
84Great quality (A) · 84 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking.	Jina Reranker v3Jina AI	`Q8_0`	2.6 GB	2.6 s	Very slow
88Excellent quality (A+) · 88 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking.	zerank-2ZeroEntropy	`Q8_0`	6.3 GB	17.4 s	Very slow
80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking.	GLiNER2.5 DecideFastino	`Q8_0`	1.2 GB	1.1 s	Slow

Best LLMs by task on this hardware Open in Model Finder

Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.