“SANTA CLARA, Calif., and TOKYO – Feb. 7, 2022 – NVIDIA and SoftBank Group Corp. (“SBG” or “SoftBank”) today announced the termination of the previously announced transaction whereby NVIDIA would acquire Arm Limited (“Arm”) from SBG. The parties agreed to terminate the Agreement because of significant regulatory challenges preventing the consummation of the transaction, despite good faith efforts by the parties. Arm will now start preparations for a public offering."
To understand why this is so big, let’s do a thought experiment. Think about the process training process of an AI. The reason AI has had such a boom since 2012, is due to the introduction of training deep neural networks on Graphical Processing Units (GPUs). This drastically reduced the time it took to train an AI model. To understand why GPUs accelerated AI, it is best explained by a CPU vs GPU mythbusters video referenced below. When training, currently the only commercially viable option is using Nvidia’s compute stack: CUDA and cuDNN. There is an open alternative being developed called SYCL (referenced below), but unsure of it’s use at this time.
In 2014, Nvidia brought out the Jetson TK1 DevKit. We will come to understand why this is important later, but for now the only thing to understand is it is a small/powerful ARM-based Linux computer.
Google then introduced a Tensor Processing Unit (TPU) in 2016, which is hardware designed specifically for machine learning (and tensorflow their AI framework). In 2020 Nvidia then announced it’s own version of a TPU, called a Tensor Core.
The AI industry is shifting to something called edge-computing, which is outside the scope of this zet, but have referenced below. Essentially it is just shifting the AI from running in a cloud provider, to running it on a small device out in world. The Nvidia Jetson is one such device.
Not only is ARM used for Nvidia’s Jetson range, but it is also what most (if not all) mobile phones and smaller devices use (will leave that as a challenge for you to go spelunking on your own).
#arm #nvidia #jetson #edge-computing #ai #cuda #cudnn #gpu #tpu