IT Home, October 29th on-site news: XHEART XinXin Aviation today officially released its large model chip X7 Ultra. With the core concept of "large models defining chips," the chip architecture is designed based on the actual needs of mass-produced models, and the computing power of this large model reaches 700 TOPS.
XHEART Xin Xin Hang Tu CEO Li Zonglin believes that as intelligent driving evolves from multiple small models to end-to-end large models, the way chips are designed also needs to change accordingly. Therefore, XHEART decided at the beginning of its entrepreneurship in 2023 to start from the model requirements in real applications, reverse-engineer the chip architecture, reorganize computing, storage, and data flow, and reserve space for future model evolution.
According to the introduction, the new XinXin HangTu X7 Ultra large model chip boasts 700 TOPS of large model computing power, and the mass-produced model has an actual measurement alignment of 99.99%.
This chip is specifically designed for large models and features a super-large core NPU architecture. The data interaction within a core is compared to the data interaction between cores, resulting in an interconnection speed increase of TB orders of magnitude. In addition, the chip's multi-level cache architecture supports large capacity and high concurrency, which can reduce inference latency and power consumption.
New Core Journey X7 Ultra is equipped with a dedicated high-precision large model operator acceleration engine. It undergoes extreme hardware optimization for a series of key operators of large models such as Layernorm, Softmax, Transpose, etc. It natively supports FP16 / FP32 high-precision vector calculations to ensure model accuracy.
The hardware parallelism of this chip, NPU, when unlocked, can fully unleash the potential of large model inference, with a computational utilization rate of 98.7%. Compared to industry-leading products, its memory bandwidth consumption is reduced by 80%, and its energy efficiency is increased by 100%.
New Core Journey X7 Ultra makes its debut in China with mass production Armv9. It meets automotive-grade standards CPU and features a 20% improvement in performance and a 10% increase in energy efficiency. Additionally, the large-cluster architecture enhances multi-core collaboration and L3 memory access performance. It integrates a brand-new SVE2 vector processing engine, offering stronger DSP performance compared to the previous generation NEON.
IT Home, equipped with the new Xinhangtu intelligent driving large model chip. X7 Ultra Core highlights at a glance:
