Rockchip SoC Product Line Evolution: A Technical Roadmap from PMP Controllers to AIoT Platforms

Rockchip SoC Product Line Evolution: A Technical Roadmap from PMP Controllers to AIoT Platforms

Disclaimer: All data in this article are sourced from public materials (Wikipedia, official Rockchip releases, and industry media reports). Some information on new products (the RK36xx series) is based on public disclosures from 2025-2026 and represents preliminary specifications; the final mass-produced products may differ.

1. Introduction

Rockchip Electronics was founded in 2001. Looking back over its more than two decades of product-line evolution, a clear transition path emerges-from "multimedia decoder chips" to "general-purpose application processors" and then to "AIoT platforms." This evolution not only reflects Rockchip's own accumulation of technical capability but also mirrors the structural shifts across the broader consumer-electronics and embedded-chip markets.

This article follows the generational progression of Rockchip's SoC products and examines the evolution of its technical roadmap and market positioning across four dimensions: CPU architecture selection, process-node advancement, multimedia capability evolution, and NPU adoption.

2. Product Generation Overview

Generation / Series	Release Year	Process Node	CPU Architecture	Target Market	Flagship Product
RK26xx	2006	180nm	ZSP DSP core	PMP/MP4	RK2606
RK27xx	~2007	Not disclosed	ARM7EJ-S + ZSP500 DSP	PMP/MP4	RK2706
RK28xx	~2009	65nm	ARM926EJ-S	MID / Tablet	RK2808A
RK29xx	2011	55nm	Cortex-A8	Tablet / TV Stick	RK2918
RK30xx / RK31xx	2012–2013	40nm / 28nm	Cortex-A9	Tablet / TV Stick	RK3066, RK3188
RK32xx	2014	28nm	Cortex-A17	Tablet / Dev Board	RK3288
RK33xx	2016	28nm	Cortex-A72 + A53	Tablet / Chromebook / SBC	RK3399
RK35xx	2020–2024	22nm / 8nm	Cortex-A55 / A76 / A72	AIoT / Edge Computing / IPC	RK3566, RK3588
RK36xx	2026 (est.)	4–5nm	Cortex-A730 + A530	High-end AIoT / Robotics	RK3668, RK3688

3. CPU Architecture Selection: Keeping Pace with ARM's Core Roadmap

Throughout its history, Rockchip has been a faithful adopter of the ARM architecture, and its CPU core selections have nearly completely tracked the evolution of ARM's application processors.

3.1 The Classic ARM Era (2006-2010)

Early Rockchip products centered on multimedia decoding, and their architecture evolved from DSP to a dual-core ARM+DSP configuration:

• RK26xx series (2006): Built on a 0.18 µm (180 nm) process and integrating a single DSP core. The DSP employs a RISC-based superscalar 5-stage pipeline with two dual arithmetic logic units (ALUs) and two multiply-accumulate units (MACs), capable of executing up to 4 instructions per clock cycle. The RK2606A was the flagship product of this period, supporting MPEG-4 video decoding.

• RK27xx series (~2007): Adopted a dual-core ARM + DSP architecture. The primary core was an ARM7EJ-S (ARMv5TE architecture) running up to 200 MHz, paired with a ZSP500 DSP coprocessor at up to 160 MHz. With a chip clock of 400 MHz, processing performance rose from 100 MIPS in the RK26 era to 800 MIPS. This generation was the first to support RM/RMVB video playback.

• RK28xx series (~2009): Built on a 65 nm process, this was Rockchip's first 65 nm chip to enter high-volume mass production. The core was an ARM926EJ-S running at around 600 MHz and supporting 720p HD video decoding. The RK28 series also integrated a DSP coprocessor (ZSP800 family).

Chips of this period primarily targeted PMPs (portable media players) and early MIDs (mobile internet devices)-a highly competitive segment with limited profit margins.

3.2 The Cortex-A Era (2011-2015)

The RK2918 (2011) was Rockchip's first Cortex-A8 product, built on a 55 nm process and among the industry's first chips with hardware support for Google WebM VP8 decoding. Rockchip then quickly followed ARM's Cortex-A9 roadmap:

• RK3066 (2012): Dual-core Cortex-A9, 40 nm process, widely used in Android tablets

• RK3188 (2013): Quad-core Cortex-A9 @ 1.6 GHz, 28 nm HKMG process-China's first 28 nm quad-core chip

• RK3288 (2014): Quad-core Cortex-A17, 28 nm process, equipped with a Mali-T760 MP4 GPU and supporting 4K H.265 decoding

The Cortex-A12/A17 naming controversy surrounding the RK3288 is an industry anecdote-in October 2014 ARM formally folded Cortex-A12 into the Cortex-A17 brand, effectively "auto-upgrading" Rockchip's spec sheet.

3.3 The big.LITTLE Era (2016-2020)

The RK3399 (2016) was Rockchip's first big.LITTLE product: dual-core Cortex-A72 + quad-core Cortex-A53 on a 28 nm process with a Mali-T860 GPU. It became a popular choice in the open-source single-board computer (SBC) market and was adopted by products such as the Asus Chromebook Flip, Pine64 Pinebook Pro, and Rock Pi 4. Google also selected its high-yield variant as the OP1 platform reference design for Chromebooks.

3.4 The Diversified-Core Era (2020-2025)

Starting with the RK3566, Rockchip's product line began to exhibit a diversified core-selection strategy:

• RK3566/RK3568: Quad-core Cortex-A55, targeting the entry-level AIoT market, 22 nm process

• RK3576: Quad-core Cortex-A72 + quad-core Cortex-A53, 8 nm process, 6 TOPS NPU

• RK3588: Quad-core Cortex-A76 + quad-core Cortex-A55 (DynamIQ architecture), 6 TOPS NPU, 8 nm process

3.5 Next Generation: The RK36xx Series (Announced 2025, Mass Production Expected 2026)

Based on publicly disclosed preliminary specifications:

• RK3668:4×Cortex-A730 + 4×Cortex-A530,~200K DMIPS,5–6nm

• RK3688: 8× Cortex-A730 + 4× Cortex-A530, 250-300K DMIPS, 4-5 nm (some sources also mention an 8 nm process, which may indicate a different variant or a discrepancy between information sources)

A noteworthy change: the RK36xx series shifts entirely to the Cortex-A730/A530 core combination, a new core architecture ARM introduced in 2024-2025. The A730 is positioned as a performance core and the A530 as an efficiency core, replacing the classic A76/A55 pairing.

4. NPU Adoption and Evolution

Rockchip's NPU strategy has progressed along a path from "discrete DSP" to "integrated NPU" and then to "NPU generational upgrades":

4.1 The CEVA DSP Phase

The RV1108 (2017) integrated a CEVA XM4 DSP as a vision-processing accelerator-Rockchip's first attempt to enter the AI vision domain. The DSP handled 1440p30 H.264 codec operations alongside a Cortex-A7 host controller, forming an early "general-purpose core + dedicated accelerator" architecture.

4.2 The First-Generation NPU

The RK3399Pro (2019) integrated an NPU for the first time, delivering 3.0 TOPS (INT8) of compute. A derivative of the RK3399, it added a neural-network accelerator on top of the existing CPU+GPU, initially targeting the AI development-board market.

4.3 Mainstreaming of Mid-Range NPUs

The RK3566 (2020) integrates a 1 TOPS NPU supporting INT8/INT16

• Both the RK3576 and RK3588 (2022-2023) provide 6 TOPS of NPU compute and support multiple precisions including INT4/INT8/INT16/FP16/BF16/TF32

4.4 Next Generation: RKNN-P3

Rockchip unveiled the RKNN-P3 NPU architecture in 2025, implemented in the RK36xx series:

• RK3668:16 TOPS

• RK3688:32 TOPS

This level of compute already approaches that of some edge GPU accelerators, giving Rockchip's high-end SoCs the potential to support more complex vision models (such as lightweight Transformer architectures) and on-device deployment of large models with up to 20B parameters.

5. Generational Evolution of Multimedia Capabilities

Multimedia processing is one of Rockchip's core competencies; the table below summarizes the evolution of codec capabilities across generations:

Product	Video Decode	Video Encode	ISP
RK2808A	720p H.264	None	None
RK2918	1080p H.264 + VP8	Encode supported	None
RK3066	1080p multi-format	1080p H.264	None
RK3188	1080p multi-format	1080p H.264	None
RK3288	4K H.264/H.265/VP9 @60fps	1080p H.264	13M
RK3399	4K H.264/H.265/VP9 @60fps	1080p H.264	13M
RK3566	4K H.264/H.265/VP9 @60fps	1080p H.265/H.264	8M ISP 2.0 3F HDR
RK3576	8K@30fps + 4K@120fps(H.265/HEVC、VP9、AVS2、AV1)	4K@60fps(H.265/H.264)	16M ISP, multi-channel MIPI CSI-2
RK3588	8K decode	4K encode	Multi-channel
RK3668	8K H.264/H.265/AV1 @60fps	8K @30fps	AI-enhanced ISP @ 8K30fps
RK3688	16K @30fps	8K @60fps	AI-enhanced ISP @ 8K60fps

Key trends:

1. Lagging catch-up in encoding: In Rockchip's early products, encoding capability long trailed decoding. Starting with the RK3576, HEVC (H.265) encoding first reached 4K resolution, and the RK36xx series achieves 8K encoding.

2. AV1 support: AV1 decoding support was introduced starting with the RK3576, reflecting the industry's shift toward royalty-free codecs.

3. AI-enabled ISP: The ISP in the RK36xx series incorporates AI-enhanced processing, shifting from traditional ISP algorithms to a hybrid architecture of "traditional pipeline + AI post-processing."

6. Process-Node Evolution and Process Selection

Generation	Process Node	Notes
RK26xx	180nm	DSP core
RK28xx	65nm	Rockchip's first high-volume 65 nm mass-produced chip
RK2918	55nm
RK3066	40nm
RK3188	28nm HKMG	China's first 28 nm quad-core chip
RK3288	28nm HKMG
RK3399	28nm HKMG
RK3566	22nm
RK3576	8nm
RK3588	8nm LP
RK3668	5–6nm
RK3688	4–5nm	Some sources mention 8 nm; pending official confirmation

It is worth noting the "fluctuation" in process-node selection: the RK3576 adopts an advanced 8 nm process, while some contemporaneous products still use mature nodes-indicating that Rockchip pursues a differentiated strategy across its product line, with high-end products chasing advanced nodes and cost-sensitive products flexibly choosing mature nodes. The RK3668/RK3688 skip 7 nm and go directly to the 5-6 nm / 4-5 nm range, showing that the high-end product line now demands higher performance and lower power.

7. Evolution of Market Positioning

7.1 The Low-Cost Tablet Era (2010-2015)

During this phase, Rockchip's main battleground was the entry-level (sub-RMB 1,000) Android tablet market, where it competed directly with MediaTek (MTK) and Allwinner. The RK3188 and RK3288 were its most successful products of this period, capturing a large share of the white-label tablet market through their excellent price-performance ratio.

7.2 Open-Source Hardware and Single-Board Computers (2016-2020)

The RK3399 unexpectedly became a favorite of the open-source hardware community. Its open documentation strategy and relatively complete Linux support led many SBC products to adopt Rockchip solutions. During this period Rockchip shifted from a purely consumer-electronics focus to a dual-track strategy of "consumer + developer."

7.3 AIoT and Edge Computing (2020-Present)

Starting with the RK3566, Rockchip clarified its AIoT strategic direction. Its product line is organized into three tiers:

• Entry level: RK3308 (audio only), RK3326/PX30 (entry-level vision)

• Mid-range: RK3566/RK3568 (mainstream AIoT)

• High-end: RK3576/RK3588 (edge AI computing)

The RK36xx series, announced in 2025, further pushes the product line toward a flagship AIoT positioning, with target applications including robotics and high-end vision devices-competing differentially with NVIDIA's Jetson series (not via a GPU path, but through an NPU path).

8. Summary and Outlook

Rockchip's product-line evolution reveals the following patterns:

1. CPU selection tracks ARM closely: Rockchip launches corresponding products no later than 1-2 years after ARM releases a new core, reflecting its close partnership with ARM.

2. The NPU has gradually become a differentiator: from discrete DSP to iterative in-house NPU architectures, the NPU has evolved from an "add-on feature" into a "core selling point," and the RK3688's 32 TOPS of compute can already support on-device large-model deployment.

3. Multimedia capabilities keep scaling upward: a decoding path of 4K -> 8K -> 16K, and codec support progressing from H.264 -> H.265 -> AV1.

4. Market positioning is migrating from low cost to high value: tablet SoC -> SBC chip -> AIoT platform, with the ASP (average selling price) trending upward.

5. Pragmatic process selection: rather than chasing the newest node, Rockchip flexibly chooses based on product positioning, reflecting the cost-control mindset typical of Chinese Fabless companies.

As planned, the RK3688 is slated for release in 2026. This will be an important window for observing Rockchip's competitiveness in the high-end AIoT market.

Appendix: Rockchip SoC Architecture Evolution Roadmap

This article is compiled and analyzed based on public information. Product specifications referenced herein are subject to Rockchip's official final releases.

ONE MORE THING

Thinkcore Team is an embedded board solution provider, developer, and motherboard supplier, specializing in the design, manufacturing, sales, and technical sharing of embedded motherboards. Our chipset portfolio includes Rockchip, Allwinner, and others. We welcome inquiries and collaboration!

E-mail:  developer@thinkcore.cn

This article is reprinted from here

https://mp.weixin.qq.com/s/fZsbn_lk83FVSp8vpwM1kw