- Openi added Google TPU to reduce dependence on NVIDIA GPUS
- Openi pressure has been highlighted to diversify the computing options by adopting a TPU
- Google Cloud Openi wins as a user despite competitive dynamics
Openi has allegedly started using Google’s tanker processing units (TPU) for Chat GPT and other products.
A report from ReutersWhich refers to the source familiar with the move, notes that this is the first major shift from Openi’s NVIDIA hardware, which has so far created the backbone of the opening stack.
Google is leasing TPU through its cloud platform, adding Open AI to the growing list of external users, including Apple, Entropic, and Safe Sprinkle.
You can like
Don’t give up nvidia
Although rented chips are not Google’s latest TPU model, this agreement reflects efforts by Openi to reduce and diversify the costs and diversify more than both NVIDIA and Microsoft Ezoor.
The decision comes when the use of chat GPT as well as workloads, which now serve more than 100 million active users daily.
This demand represents a considerable share in the annual computing budget of $ 40 billion.
Google’s V6E “Trailium” TPU is designed to diagnose a stable state and offer high -end the higher therpes with lower -end GPUs than GPU.
Although Google refused to comment and Openi did not respond immediately ReutersArrangements suggest deepening infrastructure options.
Openi continues to rely on Microsoft -backed Azor for most of its deployment (Microsoft is somehow the company’s largest investor), but the GPU’s surrounding supply issues and pricing pressures have exposed the dangers of depending on a vendor.
Mixing Google not only improves the ability to measure the opener’s computer, but also align with the wider industry trend towards combining hardware sources to take advantage of prices.
There is no suggestion that Openi is considering fully abandoning NVIDIA, but adding Google’s TPU to further control over costs and availability.
The extent to which the Openi can connect this hardware to its stack remains to be seen, especially the software, looking at the long -standing dependence on the CUDA and NVIDIA tooling of the environmental system.
