Phytium's latest desktop and server processors have been adapted to DeepSeek LLM

Recently, Phytium Tengyun S5000C and Tengrui D3000 have successfully achieved end-to-end support for DeepSeek’s full range of large language models (LLMs). This achievement spans both data center and terminal environments, with local deployment inference efficiency standing toe-to-toe with solutions utilizing foreign architectures of similar products.

In recent years, Phytium has emerged as a proactive player in the evolution of AI, driving deep integration between domestic chips and AI technology. In July 2023, Phytium collaborated with its ecosystem partners to launch a fully localized AI computing center solution, delivering intelligent computing services from edge to cloud across a wide array of industries. In 2024, the DeepSeek-V2.5 236B model was already running stably on the Phytium Tengyun S5000C platform. With 8 domestic AI accelerators, this Phytium Tengyun S5000C-powered system delivered performance that reached approximately 80% of that of foreign platforms, which enables its real-world industry adoption.

During the 2024 Spring Festival, DeepSeek released its latest DeepSeek-V3 and R1 models, which quickly became a tech sensation due to their exceptional performance and cost-efficiency, offering developers and enterprises more efficient and flexible solutions for private deployment. Phytium’s technical team wasted no time in validating their local deployment on Tengyun S5000C and Tengrui D3000 in its AI lab, creating an integrated solution that empowers users to build DeepSeek LLM intelligent computing environments for domestic end-to-end platforms.

On the data center side, servers powered by Phytium’s Tengyun S5000C can collaborate with multiple AI accelerator cards to achieve deployment of both distilled and full versions of DeepSeek models. Phytium’s AI Lab has conducted tests on the distilled version DeepSeek-R1-Distill-Llama-70B using an 8-card single-server platform, achieving an actual inference speed of 22 tokens/s. This local deployment inference efficiency stands toe-to-toe with solutions utilizing foreign architectures of similar products. The full version DeepSeek R1 671B model adopts a distributed inference framework and can be successfully deployed on a four-machine, 32-card AI server cluster, providing a complete hardware and software solution for enterprise-level private deployment.
On the user terminal side, desktops powered by Phytium Tengrui D3000 can locally deploy the distilled version DeepSeek-R1 7B model when paired with one domestic inference card, enabling AI PC capabilities to help users establish secure and reliable local knowledge assistants.

The Phytium platform has achieved end-to-end support for localized deployment of full model series of DeepSeek, as well as many open-source and leading domestic LLMs. It offers a fully localized, hardware-software integrated, cost-effective, secure, and open full-stack solutions for both enterprise and personal applications. Together with industry customers and ecosystem partners, Phytium is driving the digital transformation and intelligent upgrading of industries!