Examining RISC-V CPU and Imagination GPU Co-design through the K3 Multi-level Time-Space Model How to Build High-Performance SoC Capabilities

With the rapid growth in demand for edge AI and high-performance computing, the division of labour within the processor industry is undergoing a transformation. Recently, Arm has unveiled its in-house AI chip, a move that has drawn increased attention to the openness and ecosystem neutrality of the IP model.

Against this backdrop, the value of RISC-V CPU IP has become even more apparent: it offers chip manufacturers greater autonomy and flexibility, whilst helping to build a more open and stable industry ecosystem. At the same time, SoC design is shifting from a focus on enhancing single-core computing power towards system-level optimisation through the collaboration of multiple computing units. The synergy between RISC-V CPUs and GPUs is becoming a key capability underpinning intelligent end devices such as AI PCs, embodied intelligent robots, automotive systems and industrial applications.

The recent launch of the K3 processor by Jindie Timespace serves as a prime example of this trend in practice.

Breakthroughs in RISC-V CPUs: From General-Purpose Computing to High-Performance Platforms

Jindie Shikong is a RISC-V architecture AI CPU chip design company. Its K3 chip, launched this year, is based on Jindie Shikong’s proprietary RISC-V CPU IP (X100). With a maximum clock speed of 2.4GHz, its single-core performance rivals that of mainstream high-performance CPU architectures, whilst delivering approximately 130K DMIPS of general-purpose computing power. Furthermore, the chip integrates AI computing power of up to 60 TOPS INT4, supporting on-device inference for 30-billion-parameter large models.

This design exemplifies a significant direction for RISC-V in the high-performance SoC sector:

it exists not merely as a general-purpose computing core, but as a foundational platform that integrates AI capabilities with system capabilities.

From this perspective, RISC-V is evolving from an ‘optional architecture’ to a ‘platform-level architecture’.

From ‘computational capability’ to ‘practical usability’: system capabilities are key

In practical applications, CPU and AI computing power alone are insufficient to support a complete system.

Whether it be AI PCs, robots or various smart devices, they all require the ability to handle graphical interfaces, multimedia processing and the operation of an operating system. This means that SoC design must shift from ‘providing computing power’ to ‘supporting a complete system’.

Against this backdrop, the synergy between GPUs and RISC-V CPUs has gradually emerged as one of the key approaches to building high-performance SoCs.

Take the K3 as an example: by integrating the Imagination PowerVR GPU – IMG BXM 4-64 – into the system, the chip delivers not only high-performance general-purpose computing and AI capabilities but also comprehensive graphics and parallel computing capabilities.

The value delivered by this combination is evident on multiple levels:

• Provides a complete graphical environment for Linux distributions such as Ubuntu and other operating systems (Ubuntu is one of the most widely used Linux distributions in the world)
• Supports standardised graphics and computing interfaces such as Vulkan
• Compatible with open-source GPU graphics stacks, lowering the barrier to entry for developers
• Expands parallel computing capabilities, such as camera stitching or object detection.