The race for more powerful and efficient AI hardware gained momentum this week as Intel and Google announced latest chips designed to assist them change into more independent of NVIDIA technology.
It looks like latest AI models are released every week. Behind each release are weeks of coaching on cloud computing data centers, most of that are powered by NVIDIA GPUs.
Intel and Google have each announced latest in-house AI chips that allow large AI models to be trained and deployed faster while using less power.
Intel's Gaudi 3 AI accelerator chip
Intel might be higher known for the chips that power your PC, but on Tuesday the corporate announced its latest AI chip called Gaudi 3.
NVIDIA's H100 GPUs make up the vast majority of AI data centers, but Intel says Gaudi 3 “delivers a median of fifty% higher inference and a median of 40% higher power efficiency than Nvidia H100 – at a fraction of the price.”
An enormous think about Gaudi 3's power efficiency is that Intel used Taiwan Semiconductor Manufacturing Co.'s 5nm process to make the chips.
Intel didn't provide pricing details, but when asked the way it compares to NVIDIA's products, Das Kamhout, vice chairman of Xeon software at Intel, said: “We expect it to be very competitive.”
Dell Technologies, Hewlett Packard Enterprise, Lenovo and Supermicro will likely be the primary to deploy Gaudi 3 of their AI data centers.
Intel CEO Pat Gelsinger summed up the corporate's AI ambitions by saying, “Intel is bringing AI all over the place across the enterprise, from the PC to the info center to the sting.”
The #IntelGaudi 3 #AI Accelerator offers a highly competitive alternative to NVIDIA's H100 with higher performance, increased scalability and PyTorch integration. Discover other essential product benefits. https://t.co/sXdQKjYFw0 pic.twitter.com/iJSndBQkvT
Google's Arm and TPU upgrades
On Tuesday, Google announced its first custom Arm-based CPUs that it plans to make use of to power its data centers. The latest chip, called Axion, is a direct competitor to CPUs from Intel and AMD.
Google claims Axion offers “30% higher performance than the fastest general-purpose Arm-based instances available within the cloud today, as much as 50% higher performance, and as much as 60% higher power efficiency than comparable current-generation x86-based instances.”
Google's latest Arm-based Axion CPU. Source: Google
Google has moved several of its services reminiscent of YouTube and Google Earth to its current-generation Arm-based servers, which is able to soon be upgraded with the Axion chips.
With a strong Arm-based option, customers can more easily migrate their CPU-based AI training, inference and other applications to Google's cloud platform without having to revamp them.
For large-scale model training, Google has largely relied on its TPU chips as an alternative choice to NVIDIA hardware. These are also upgraded with a single latest TPU v5p, which now comprises greater than double the variety of chips in the present TPU v4 pod.
TPU v5p, our strongest and scalable TPU, is now generally available! #GoogleCloudNext pic.twitter.com/mmfWlzHeqs
Google doesn't need to sell its latest Arm chips or its TPUs. The company desires to advance its cloud computing services reasonably than change into a direct hardware competitor to NVIDIA.
The updated TPUs will strengthen Google's AI hypercomputer, enabling large-scale AI model training. The AI hypercomputer also uses NVIDIA H100 GPUs, which Google says will soon get replaced by NVIDIA's latest Blackwell GPUs.
Demand for AI chips isn't more likely to decelerate anytime soon, and it's looking less like a one-horse race from NVIDIA than before.
