Google launches the challenge to NVIDIA in the AI supercomputer sector. The company, in a document that will be officially presented at Isca 2023, highlights how its artificial intelligence models are faster and present a lower environmental impact compared to solutions that take advantage of NVIDIA hardware. At the basis of the goal achieved by Google there is TPU V4 (Tensor Processing Unit), can guarantee up to 1.7x faster performance with up to 1.9x greater efficiency.
Google challenges NVIDIA: the goal is the leadership of the AI supercomputer sector
Prova Bitdefender. Leader in Cybersecurity
TPU V4 is nothing new: the system was launched during 2020. Google today unveils the results after 3 years of operation in a corporate datacenter located in Oklahoma. The chip, as clarified by Google, was also used for train AI image generator, Midijourney.
It should be noted, in any case, that Google and its TPU V4 will have to deal with the new NVIDIA H100 chip. The new solution from NVIDIA, in fact, should guaranteeand four times the performance of the previous generation (A100) which Google refers to for its comparison with NVIDIA.
In any case, Google has a new answer ready. Over the next few months, in fact, the company should present a new version of the TPU.