On the evening of September 16, CCTV’s “News Broadcast” opened like a “mysterious treasure box,” bringing us a super surprise—the long-rumored Alibaba self-developed AI chip has made its appearance!
It comes from Alibaba’s T-head PPU chip, showcasing a “kingly demeanor” right from the start.
In terms of parameters, it directly competes with NVIDIA’s H20 chip on several important specifications, even surpassing NVIDIA’s A800 chip!
It’s like a rookie challenging seasoned experts in the field, yet not falling behind.
In terms of memory, the T-head PPU integrates HBM2e, which is on par with the A800, but slightly inferior to H20’s HBM3, akin to a running race where there’s just a small distance separating them.
However, the memory capacity of the PPU is the same as that of the H20, both at 96G, making them equal in this aspect.
For inter-card communication bandwidth, the PPU offers 700GB/s, placing it between the A800 and H20, acting as a “balancer” between the two.
In terms of PCIe interface, the PPU is superior to the A800 and equally excellent as the H20.
Regarding power consumption, both the PPU and A800 are at 400W, which is lower than H20’s 550W, making it more “energy-efficient and environmentally friendly,” like a “little expert” at managing resources.
When compared to the domestic Huawei Ascend 910B chip, CCTV’s visuals show that the T-head PPU leads in most performance parameter indicators.
However, the latest model in the Ascend series is the 910C chip; if they were to compete, what sparks would fly?
It truly piques curiosity!
In fact, as early as late August, media reports indicated that Alibaba had developed a new chip for AI inference tasks to fill the market gap left by NVIDIA.
On September 11, Silicon Valley tech media also revealed that both Alibaba and Baidu are using their self-developed AI chips to train models.
Alibaba is using its self-developed chip to develop smaller AI models, while Baidu is attempting to train a new version of its Wenxin large model using the Kunlun chip P800, but both companies have not completely abandoned NVIDIA’s chips.
It’s like a complex “love triangle,” with everyone searching for the most suitable “partner” for themselves.
“AI + Cloud” has now become one of Alibaba’s two major growth engines alongside e-commerce.
In February of this year, Alibaba announced plans to invest at least $53 billion over the next three years to build cloud and AI hardware infrastructure, a staggering investment akin to “pouring heavy money” into the future of AI.
In June, Alibaba Group’s leaders also stated that AI will be the biggest driving force for change in the next decade, and Alibaba will increase investment in three areas: AI and cloud computing infrastructure, AI foundational models, and AI-native applications, as well as the AI transformation and upgrading of existing businesses.
The competition in the AI chip field has always been fierce.
NVIDIA, as the “big brother” of the industry, has maintained a significant position.
However, more and more companies are beginning to develop their own AI chips, such as Huawei, Alibaba, and Baidu, all striving to break NVIDIA’s “monopoly.”
It’s like an intense marathon race, with everyone desperately running forward, not wanting to fall behind.