Huawei is working on a dynamic tech strategy to elevate the performance of its AI chip in the coming years. Well, this tech plan is 4 ZettaFLOPS FP4 that will align the internal components of the Ascend NPU processor to power up its efficiency.
In September this year, Huawei launched its new-gen data center-scale AI solution. This tech can offer 1 FP4 ZettaFLOPS performance for the AI inference aspect.
FP4 ZettaFLOPS describes an enormous amount of computational power for AI tasks, particularly with a 4-bit floating-point data format. A FLOP(s) refers to Floating point Operations Per Second. It is a standard measure of computer performance.
While the Atlas 950 delivers only 1 FP4 ZettaFLOPS, Huawei is planning 4 ZettaFLOPS FP4 performance for its AI chip tech by 2028. The company revealed a roadmap of Ascend processors, including the new-gen NPUs, to accelerate AI workloads.
Since Huawei can’t access TSMC’s leading-edge process technologies or high-end HBM4 and GDDR7 memory from global companies, it is working on a new architecture and new types of memory for the upcoming Ascend 950 series.
As a result, Huawei Ascend AI chips will be able to offer multi-ZettaFLOPS efficiency. The details of this new technology are currently unclear. But as per previous reports, the company will soon launch an ultimate chipmaking infrastructure.
Meanwhile, the company is edging towards its new dream – 2nm chips for smartphones. One of the Huawei patents shows that the company is trying to make amendments in terms of chip design to make its Kirin processors more useful.
If true, Kirin will soon become equivalent to foreign smartphone chipsets, offering the best flagship-level performance to its devices. Though it would be worth seeing how much time it will take to do so!
(Image Credits: Huawei)
|| via ||
The post Huawei plans this tech strategy to boost AI chip performance by 2028 appeared first on Huawei Central.
