Zhaoxin Launches DeepSeek-R1 Models for All CPU Types

Key Takeaways

1. Zhaoxin has launched DeepSeek-R1 Distill models across all its products, including desktops and servers, showcasing adaptability with multiple model versions up to 671B.
2. Their CPUs are compatible with Linux, Windows, and local Chinese operating systems, and support local GPU AI acceleration cards for diverse AI applications.
3. The KaiXian KX-7000 desktop processor runs the DeepSeek-R1-7B model, offering smart features like document management and coding support, integrated with popular applications.
4. The Lianhe Donghai XRS302 workstation, equipped with domestic GPU cards, effectively runs multiple DeepSeek-R1 models, optimized for low-latency and quiet operation.
5. The Kaisheng KH-40000/32 processor can independently run the DeepSeek-R1-671B model and supports up to 2TB of memory, showcasing powerful server-grade AI processing capabilities.


Zhaoxin has recently announced the successful launch of the DeepSeek-R1 Distill models across all its products, ranging from standard desktop computers to powerful servers.

The Chinese CPU manufacturer revealed that their processors can now handle multiple versions of the DeepSeek-R1 Distill model. This includes the 1.5B, 7B, 14B, 32B, 70B, and the massive 671B variant. This deployment covers all of their x86-compatible devices, showcasing the adaptability of their technology.

Broad Compatibility

In addition, Zhaoxin’s CPUs are designed to work seamlessly with Linux, Windows, and various indigenous Chinese operating systems. They are also fully compatible with local GPU AI acceleration cards. This combination creates a strong computing platform suitable for a wide range of AI applications.

Desktop AI Deployment

The KaiXian KX-7000 desktop processor has shown it can successfully run the DeepSeek-R1-7B model when used with domestically produced graphics and acceleration cards. By utilizing DeepSeek’s advanced language understanding, this system offers features such as smart document management, data analysis, and coding support, all integrated smoothly with popular applications like WPS Office and VSCode.

The KX-7000 belongs to Zhaoxin’s latest series of desktop processors. It uses the “Century Avenue” microarchitecture, which features innovative chiplet interconnections, includes 8 CPU cores with 32MB of shared L3 cache, and can achieve speeds of up to 3.7GHz.

Workstation Performance

The Lianhe Donghai XRS302 workstation, powered by the KH-40000/16 server processor, has successfully implemented the DeepSeek-R1-7B, 14B, and 32B models. Equipped with four high-performance domestic GPU inference acceleration cards, this workstation offers a budget-friendly yet powerful solution for AI inference tasks.

Zhaoxin has optimized its CPU-GPU interconnect system to enhance data flow while reducing latency. Additionally, the workstation features a low-power, quiet design, making it ideal for office environments that require AI inference without excessive noise and energy consumption.

Server-Grade AI Processing

On a more advanced level, the Kaisheng KH-40000/32 processor features a dual-socket configuration with a total of 64 cores and can support up to 2TB of memory. When paired with local accelerator cards, this CPU effectively deployed the DeepSeek-R1-70B model.

Moreover, the KH-40000/32 was able to run the entire DeepSeek-R1-671B model independently, without the need for external acceleration cards. This achievement highlights the impressive capabilities of Zhaoxin’s server CPUs to handle models with hundreds of billions of parameters with ease.

Although Zhaoxin’s claims are noteworthy, it is important to approach these performance outcomes skeptically. Like any benchmarks provided by vendors, actual performance can vary from controlled testing situations, and the company might showcase optimized scenarios that portray their products in the most favorable manner.

Source:
Link

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *