Xu famous: “Computing energy is – and can proceed to be – key to AI. That is very true in China.” He reiterated that Huawei’s purpose is to sustainably meet long-term computing demand by constructing superpods and superclusters with the semiconductor manufacturing course of nodes which might be virtually out there to the Chinese language mainland.
A superpod is a single logical machine, made up of a number of bodily machines that may study, suppose, and motive as one. Xu unveiled the corporate’s latest superpod merchandise: the Atlas 950 SuperPoD (with 8,192 Ascend NPUs) and the Atlas 960 SuperPoD (with 15,488 Ascend NPUs). These two superpods will ship an industry-leading efficiency throughout a number of key metrics, together with the variety of NPUs, whole computing energy, reminiscence capability, and interconnect bandwidth.
Primarily based on publicly introduced product roadmaps from friends within the {industry}, these superpods are at the moment probably the most highly effective superpods on the planet, and can stay so for years to return.
Xu additionally introduced the Atlas 950 SuperCluster (with greater than 500,000 Ascend NPUs) and Atlas 960 SuperCluster (with a couple of million Ascend NPUs), that are large-scale computing clusters comprised of a number of Huawei SuperPoDs. These too are poised to outperform all different computing clusters available on the market.
With the world’s strongest superpods and superclusters, Xu mentioned Huawei has what it takes to supply considerable computing energy for ongoing, fast developments in AI, each now and sooner or later.
Xu went on to introduce the TaiShan 950 SuperPoD, the world’s first general-purpose computing superpod. The TaiShan 950 SuperPoD, mixed with Huawei’s distributed GaussDB, can function a viable various to mainframes and mid-range computer systems, and even Exadata database servers.
To this point, a significant bottleneck for large-scale AI computing infrastructure is interconnect expertise, specifically the bodily limitations of present cable expertise (each optical and copper) to hyperlink up huge numbers of chips and superpods over lengthy distances whereas sustaining a dependable, high-speed, and low-latency connection. Huawei has honed its connectivity experience over the previous three many years. And by combining this experience with various methods improvements, the corporate has overcome these challenges.
The result’s UnifiedBus, a groundbreaking interconnect protocol for SuperPoDs. Xu additionally launched the technical specs for UnifiedBus 2.0, within the hopes that {industry} companions will undertake this protocol to develop extra UnifiedBus-based merchandise and elements and collectively create an open UnifiedBus ecosystem.
“Superpods and superclusters powered by UnifiedBus are our reply to surging demand for computing, each right now and tomorrow,” concluded Xu. “Our purpose is to maintain pushing developments in AI to create better worth.”
Themed All Intelligence, HUAWEI CONNECT 2025 will delve into AI throughout three dimensions: technique, expertise, and ecosystems. On the occasion, Huawei will deep dive into its newest strategic initiatives and unveil its all-new digital and clever infrastructure merchandise, scenario-specific options for industries, and growth instruments.
The occasion will run from September 18 to twenty on the Shanghai World Expo Exhibition & Conference Middle and Shanghai Expo Middle.
For extra data, please go to HUAWEI CONNECT 2025 on-line at www.huawei.com/en/occasions/huaweiconnect
Join our each day information round-up!
Give your online business an edge with our main {industry} insights.
