Recently, Intel unveiled two new CPUs that will utilize artificial intelligence which will be the first time the company will use AI based chips. According to Tech Radar, the two chips are part of the Nervana Neural Network Processor (NPP) line built for large computing centers where one of them will handle interference while the other will be in-charge of training AI. The first processor is called Nervana NNP-T, also called Spring Crest which has been designed to train AI systems. It has been equipped with 24 Tensor processing clusters which will be especially used for powering neural networks. The second processor is called the Nervana NNP-I, also called Spring Hill which will be the interference SoC.
Google laid down its path forward in the machine learning and cloud computing arenas when it first unveiled plans for its tensor processing unit (TPU), an accelerator designed by the hyperscaler to speeding up machine learning workloads that are programmed using its TensorFlow framework. Almost a year ago, at its Google I/O event, the company rolled out the architectural details of its second-generation TPUs – also called the Cloud TPU – for both neural network training and inference, with the custom ASICs providing up to 180 teraflops of floating point performance and 64 GB of High Bandwidth Memory. In addition, the TPU2 boards can be linked together by a fast, dedicated network to create TPU pods, essentially multi-petaflop supercomputers designed for machine learning jobs. As we at The Next Platform have noted before, Google has positioned the TPU2 to accelerate deep learning workloads for consumer-facing workloads, ranging from search and maps to voice recognition and emerging areas like training self-driving vehicles. However, much of the talk around TPU2 has been just that: talk.
At a recent investor conference, Intel (NASDAQ:INTC) CEO Brian Krzanich offered significant insight into the company's view of -- and strategy to win in -- the market for chips that handle artificial intelligence/machine learning workloads. Krzanich started by noting that artificial intelligence and machine learning (I'll refer to these as simply "machine learning" from here on out) represent a relatively small part of the overall data center market today, but acknowledged that it's the quickest-growing data center workload. Moreover, Krzanich even said that, over the long term, Intel expects it to be the largest workload within the data center. "Where will it end [up]? Will it be 25% of the workload?
The world's largest computer chip boasts a whopping 1.2 trillion transistors and a staggering 400,000 cores, dwarfing the latest generation AMD Ryzen and Intel Core processors in both regards. Even so, it's small enough to fit inside a computer that stands just a bit over 2 feet tall. That system is the Cerebras CS-1, and you'll notice I'm not calling it a PC. It doesn't qualify as one (so don't expect to see any Black Friday deals on this rig). The CS-1 is a computer, though, and according to Cerebras there is no other system in the world that is faster at processing artificial intelligence (AI) workloads.
These days, just about everyone in the technology industry is talking Artificial Intelligence (AI) and Machine Learning (ML). There's a huge amount of excitement and a rush to be the first to get it right. What you might have noticed in this dialogue is that almost everyone is talking big, powerful, Neural Network accelerators as an essential part of bringing ML to life on your device – and whilst it's true that they have a significant role to play, they're just one part of the story.