Hygon Unveils Innovative Plans for China’s Domestic Tech Market: Next-Gen C86 Chips Featuring +15% IPC, SMT4, and Competing with Intel Xeon 6

China’s leading semiconductor manufacturer, Hygon, has unveiled a plan to develop six new chips targeted at the domestic technology market, featuring the advanced C86 CPU as the flagship product.

Introducing Hygon’s Cutting-Edge C86 CPUs: Competing with Intel’s Xeon Lineup

Based in Beijing, Hygon has established itself as a prominent player in the chipmaking landscape, committed to meeting the demands of the Chinese market. Their earlier collaboration with AMD to produce C86 chips garnered significant interest, paving the way for the forthcoming next-generation C86 processors.

The recently announced lineup will include six new chips aimed at various domestic applications. The standout offering is the next-gen C86 CPU, designed for versatile computing needs. Currently, Hygon’s C86-4G series caters to both mainstream consumers and enterprise clients, while the anticipated C86-5G series promises enhanced performance and capabilities.

The upcoming Hygon C86 series boasts a newly designed architecture, featuring an impressive IPC (Instructions Per Cycle) increase of over 15%.Earlier indications suggested a potential 17% IPC boost for this series. Furthermore, this new lineup will support multi-threading with SMT4, allowing each core to manage four simultaneous threads, enhancing its capabilities for enterprise and server applications. The addition of AVX512 instructions will elevate vector computing power, while integrated AI acceleration support for INT8 and BF16 instructions stands out as a critical feature.

Enhanced microarchitecture, promising over 15% IPC improvement.
Supports SMT4 multi-threading to efficiently handle high-demand scenarios.
AVX512 instruction set compatibility for advanced vector processing.
New AI acceleration features with INT8 and BF16 support.

Performance claims indicate that Hygon’s next-generation C86 chips will directly compete with Intel’s Xeon 6 series, a significant step toward aligning China’s domestic chip production capabilities with major international players.

In addition to CPU development, Hygon is also broadening its portfolio with a new GPU accelerator named “DCU, ”engineered for AI training and computational tasks. This accelerator will showcase a full-precision GPGPU architecture along with ultra-high-speed interconnect capabilities. It is anticipated that this chip will rival NVIDIA’s A100 based on the Ampere architecture.

Full-precision GPGPU architecture designed for complex calculations.
Comprehensive support for FP64, PF16, and BF16 precision formats.
Fast inter-chip interconnect for enhanced data transfer rates.
Compatible with advanced HBM memory solutions.

The upcoming product range will also include a PCIe 5.0 Switch, a Scale-Up Interconnect Switch, a 400G Network Interface chip, and a ScaleFabric 400/800G switch. These innovations are critical for establishing Hygon’s proprietary AI and high-performance Enterprise ecosystem. Detailed specifications of the chips are as follows:

Hygon PCIe 5.0 Switch:

Designed for high-speed I/O operations.
Positioned to compete with Broadcom’s premium PCIe switching products.
Offers 104 lanes for expanded connectivity.

Scale-Up Interconnect Switch:

Performance levels matching NVLINK + NVSwitch.
Facilitates ultra-high-speed interconnects for multiple GPU/CPU configurations.

ScaleFabric 400G Network Interface Chip:

Delivers 400Gb/s port speed with native RDMA protocol.
Features credit-based flow control and lossless characteristics.
Communication latency close to 0.93µs and supports up to 256K QP (Queue Pairs).
Plug-and-play capability, ensuring a seamless transition to 800G.
Comparable performance to NVIDIA’s InfiniBand NDR Network AICs.

ScaleFabric 400/800G Switch:

Offers port speed options of 400/800Gb/s with native RDMA support.
Provides a dense switching capacity with 80 ports at 400Gb/s and a total of 64 Tb.
Ensures a switching latency of just 260ns and rapid recovery from link failures.
Utilizes self-developed 112G high-speed SerDes IP, excelling in signal transmission capabilities.
Performance metrics on par with NVIDIA’s InfiniBand NDR Switches.

Production of these innovative chips is anticipated to commence between 2026 and 2027, with their introduction set to reshape the technological landscape in China.

Source & Images