Open Source General Computing Chip Platform（GPGPU）

"Blue-porcelain" is a GPGPU design proposed by Advanced Architecture Laboratory of Shanghai Jiao Tong University, which is an open source general computing chip platform. "Blue-porcelain" architecture is proposed to support large-scale parallel general computing, AI computing, and scientific computing. It provides flexible programmability to conform to programmers' programming habits and minimize the difficulty of software development.

Why Open Source

The fundamental problem comes from software. Traditional software architecture and hierarchy went through great changes before a new order can be established. At present, the problem of "reinventing the wheel" exists in the chip industry at the software level. Each company may re-develop a framework that can be only applied to its own products, and the general framework support cannot meet the needs of algorithm diversity. We can imagine that the cost is huge if we want to establish a great chip environment as the existing platform.

Therefore, we try to solve the problem of "capital" with free of charge, solve the problem of "talent" with openness, and solve the problem of "time" with open source.

Goals and plans

Our goal is to combine three basic targets into one project, including a professional textbook, a core course and an open source platform.

Textbooks and Courses

We have published the textbook "General Graphics Processor Design - GPGPGPU Programming Model and Architecture Principles" so far, which has been widely recognized by the industry. The core curriculum is also being actively promoted.

Open Source Platform

In the open source general computing chip platform, we define and implement a popular GPGPU design by providing a basic instruction set, a hardware architecture and a simulator. At present, there are several open source GPGPU simulators in academia, some of which are still implemented using older instruction sets or intermediate languages, and some of which are improved and extended based on RISC-V CPU instruction sets to realize the functions of GPGPU. Our platform is based on advanced GPGPU architecture of industry in recent years, and can realize complete functions of a GPGPU. It is not based on an existing open source instruction set, but an independent and complete GPGPU architecture and instruction design, so it can be better and wholly integrated into the existing GPGPU ecosystem.

Plan

We will release open source functional simulator and performance simulator in stages, and build basic test sets and advanced test sets for function verification. At present, the functional simulator of Blue-porcelain architecture has been released. In the open source project, 30 parallel computing programs are included as the basic test set to verify the basic functions. In the future, we will continue to optimize the GPGPU architecture, and update the functional simulator and the performance simulator. At the same time, there are more than 100 programs under development in the advanced test set, and we will continue to update and gradually release the tests.

In the open source platform, we focus on the design of GPGPU architecture and instructions, but do not involve specific chip definition and manufacturing process, nor restrict the implementation of chip details. Users can carry out specific hardware design and development based on the open source platform or participate in the construction of the platform in the form of software development to enrich the ecology of the open source GPGPU. We hope to build an open source general computing chip platform that can serve the industry and academia by providing architecture and instruction design, GPGPU simulator and test set software. At the same time, we want to foster talents for the GPGPU industry, build a knowledge hierarchy, coordinate the cooperation mode within the industry, and form a good open ecology, which will energize follow-up work.

The project itself is still at a very early stage, but many hands make light work. Teachers, students and engineers from all walks of life who are interested in the project are welcome to give suggestions for the general computing chip platform. Also, we welcome everyone's participation in the subsequent development to build an open source GPGPU ecosystem. Contact us