In addition to Intel and AMD, VIA in Taiwan will also build x86 processors. I wonder how many people know about it? Recently, CenTaur, a 24-year-old processor research and development department of VIA, developed the world's first x86 processor with integrated AI coprocessors, an

In addition to Intel and AMD, VIA in Taiwan will also build x86 processors. I wonder how many people know about it? Recently, CenTaur, a 24-year-old processor research and development department of VIA, developed the world's first x86 processor with integrated AI coprocessors, and has a working prototype. Chip testing began in September this year.

new processor is manufactured using TSMC's 16nm process, with a core area of ​​no more than 195 square millimeters. It uses a ring bus design inside, integrating eight x86 CPU cores, 16MB shared level 3 cache, four-channel DDR4-3200 memory controller , PCIe 3.0 controller (44 lines), south bridge and IO functions, and is a complete SoC. The biggest highlight of

is the AI ​​coprocessor "NCORE", which occupies an area of ​​about 34.4 square millimeters (17.6%). The software is mapped to PCI devices. supports the acceleration of DNN deep neural network creation and training, and claims to provide up to 20TB/s of memory bandwidth and 20 trillion AI operations per second.

main frequency can work at 2.5GHz, and actually supports AVX-512 instruction set , which is not available in AMD Zen 2 architecture.

CHA processor core diagram

CHA processor module diagram

Recently, Centaur released many architectural details of this processor, but what is interesting is that it was not released by itself, but from Linley Group, an authoritative organization in California, the processor technology of the United States and a publisher of the famous chip magazine "Microprocessor Report". The latter carefully studied Centaur's processor architecture design document, interviewed relevant designers, and gave this report.

Linley Gwennap, editor-in-chief of Microprocessor Report magazine, praised this newly designed x86 processor: "Centaur returns to the x86 market with a high-profile return, bringing innovative processor design, the entire eight high-performance CPU cores , a custom deep learning accelerator (DLA). This is the industry's first server processor design to integrate DLA. The new accelerator NCore's neural network performance is even better than the most powerful Xeon, and does not require expensive external GPU computing card assistance."

Linley Group revealed that Centaur's new x86 microarchitecture is called "CNS" . The design goal is that IPC should be higher than traditional PC processors. 4 x86 instructions can be decoded per clock cycle and 10 micro-operations are executed in parallel. The first processor is temporarily named "CHA", among which the peak performance of the AI ​​coprocessor INT8 integer is as high as 20 TOPS (20 trillion operations per second).

CNS microarchitecture diagram

NCore AI coprocessor architecture diagram

Linley Group measures the AI ​​performance of x86 processor based on authoritative MLPerf performance tests. It was found that the AI ​​inference performance of Centaur CHA processor is equivalent to 23 world-class Intel x86 cores, and the latter must support 512-bit VNNI vector neural network instructions. - In fact, Intel does not have a real 32 core product yet. The architecture of

Centaur AI coprocessor is similar to the SIMD (single instruction multi-data) concept of VNNI instructions. However, with the support of 16MB of dedicated memory and 20TB/s bandwidth, can process 32,768 data bits per clock cycle. After handing the inference processing to a special AI coprocessor, the x86 core can perform other general tasks with confidence.

Centaur also provides developers with new algorithms that take advantage of the unrivalled ultra-low inference latency of the Centaru AI coprocessor and work closely with the x86 CPU core.

At the ISC East conference held in New York State, Centaur also publicly displayed the CHA processor for the first time. In addition to traditional AI applications such as video analysis, real-time object detection and classification, it also showed off cutting-edge applications such as semantic segmentation (pixel-level image classification), human body posture estimation (simple strokes), which is eye-opening.

At present, Centaur is improving and optimizing the hardware performance and software efficiency of the new platform, and the new processor is expected to be officially put into production in the second half of next year.

Linley Group's detailed report can be downloaded

Comparison of different CPU architectures

Centaur processor test