惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

罗磊的独立博客
L
LangChain Blog
aimingoo的专栏
aimingoo的专栏
IT之家
IT之家
B
Blog
博客园_首页
博客园 - 司徒正美
有赞技术团队
有赞技术团队
博客园 - 聂微东
I
InfoQ
美团技术团队
GbyAI
GbyAI
阮一峰的网络日志
阮一峰的网络日志
H
Help Net Security
大猫的无限游戏
大猫的无限游戏
MyScale Blog
MyScale Blog
WordPress大学
WordPress大学
The GitHub Blog
The GitHub Blog
A
About on SuperTechFans
人人都是产品经理
人人都是产品经理
Microsoft Azure Blog
Microsoft Azure Blog
Engineering at Meta
Engineering at Meta
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
The Cloudflare Blog

The Register - On-Prem: Systems

Qualcomm teases agentic CPUs and smartphones Fujitsu says quantum and AI will replace mainframes in 2035 ZTE & China's NCRC partner for smart interventional medicine Core Scientific accelerates crypto-to-AI pivot Meta to use millions of AWS Graviton cores AI now gobbling up power and management chips for servers Tesla stakes AI dreams on Intel's unfinished AI chip SK Hynix breaks ground on Indiana advanced packaging plant Datacenter boom keeps dirty coal plants alive in the US AMD's Ryzen 9 9950X3D2 Dual Edition tested World's blandest man steps down from CEO job to spend more time in tastefully appointed home World's blandest man steps down from CEO job Intel eases reliance on TSMC with 'Merica-made Core Series 3 processors Intel eases reliance on TSMC with Core Series 3 CPUs Guide to GPU virtualization: passthrough, vGPU, and MIG Guide to GPU virtualization: passthrough, vGPU, and MIG Orbital datacenter startup admits launch economics don't fly AI-powered mainframe exits are a bubble set to pop Cloud-smart strategy helps Interactive meet GenAI demands Oracle taps Bloom for fuel cells to support datacenter binge Oracle taps Bloom for fuel cells to support datacenter binge UK signs Rolls-Royce SMR design deal Japan going back to the future by reviving its chip industry Japan going back to the future by reviving its chip industry AWS ponders selling its home-grown chips by the rack-load Supply chain challenges risk delaying Nvidia's Rubin GPUs Supply chain challenges risk delaying Nvidia's Rubin GPUs Supermicro investigating alleged China chip smuggling Intel trapped in Elon's reality distortion field UALink delivers 2.0 spec before v. 1.0 silicon ships
Alibaba delivers RISC-V server chip optimized for Chinese AI
Simon Sharwood Simon Sharwood · 2026-03-25 · via The Register - On-Prem: Systems

Systems

Alibaba delivers RISC-V server chip optimized to run China’s top AI models

Claims its set performance records but looks to be years behind western fare

Alibaba has revealed a new server chip that it says is the most powerful processor ever to use the RISC-V instruction set.

According to a social media post by Alibaba’s DAMO Academy, which develops some of its chips, the new XuanTie C950 is ready to power cloudy servers, generative AI workloads, high-end robotics, and edge computing devices.

“The XuanTie C950 is equipped with a self-developed AI acceleration engine, and for the first time natively supports large models with hundreds of billions of parameters, such as Qwen3 and DeepSeek V3, potentially becoming a new type of high-end CPU for the AI Agent era,” the post enthuses.

Alibaba claims the machine’s single-core general-purpose performance “exceeded 70 points in the SPECint 2006 benchmark test.” Photos from Tuesday event at which Alibaba announced the chip suggest its SPECInt 2017 benchmark result is 2.6GHz. Per analysis by Google researcher Laurie Kirk puts it nearly on par with Apple’s M1 chip – which the iGiant launched in the year 2020.

Kirk also noted that Alibaba claimed it’s implemented version 23.1 of the RISC-V RVA, a minor update proposed in August 2025. She expressed surprise Alibaba was able to use it so quickly.

A product page and spec sheet for the chip add more details – but also omit details like core count and instead offer the non-specific description of the chip as a “64-bit multi-core CPU IP.”

Based on those documents the “acceleration engine” mentioned above is probably the XuanTie Tensor Processing Engine (TPE) and supports data types from FP16 down to INT4/FP8, plus the micro-scaling formats MXFP8, MXFP4, and RVFP4. The chip can achieve 8 TOPS per TPE.

The memory subsystem apparently “comprises a high-performance multi-level cache hierarchy with an ultralow 4-cycle load-to-use L1 data cache latency, a private per-core L2 cache supporting large capacity configurations, and an MMU with multiple RISC-V virtual memory modes and two-stage address translation.”

In a section of the spec sheet discussing buses, there’s a mention of a multi-processor mode that “leverages the XL-300 interconnect to form clusters of up to 8 cores.”

Posts The Register has seen suggest Alibaba had the chip made using a 5nm process, a feat that’s not beyond some Chinese chipmakers. Whether they can build chips of that caliber in large quantities is less certain.

Alibaba CEO Yongming Wu last week acknowledged that Chinese chips lag behind those from western chipmakers. He said the company’s response is “to engage in more profound co-design with Alibaba's cloud infrastructure and the Qwen model to provide improved cost effectiveness. This is one key differentiator … that sets us apart from other chip companies.”

So while the XuanTie C950 may not be a particularly powerful chip, and Alibaba may struggle to get millions of them into production, native support for the company’s own Qwen models suggests Wu’s vision for a nicely harmonized AI stack is becoming a reality. ®