INTEL IS GETTING BIG INTO ARTIFICIAL INTELLIGENCE and has revealed two AI-centric CPUs to challenge the likes of Google and Amazon in the machine learning world.
The two chips will be the first to form Intel's Nervana Neural Network Processor (NNP) lineup, one of which will take care of the training of an AI system while the other handles inference - the way AI and machine learning algorithms are actually put to use in real-world situations.
The Nervana NNP-T, codenamed Spring Crest, is the training chip, hence the T in its name. The chip comes sporting 24 Tensor processing clusters designed to specifically for powering neural networks, which a high-efficiency HBM2 memory architecture, 60MB of on-hip distributed memory, and up to 119 TOPS (theoretical operations per second).
In essence, this is an SoC designed to provide all the bits to effectively train an AI system on dedicated hardware, rather than trying to do so solely with general compute processors like Intel's own Xeon CPUs.
The Nervana NNP-I, codenamed Spring Hill, is the inference SoC. It uses Intel's 10-nanometre process technology and comes with Ice Lake cores to put trained AI systems into action. There's other good stuff like support for four 64GB LPDDR4x modules for high-speed memory bandwidth and 12 "Inference Compute engines", to put AI models to work.
The idea of both SoCs is that Intel will offer chips designed to take care of a lot of the heavy lifting in AI workloads in data centre environments. They are designed to run with rather than replacing Xeon CPUs, which can do such work but aren't tailored specifically for such tasks.
So the Nervana can be seen more as rivals, or at least contemporaries, to the likes of Google's Tensor Processing Unit, Nvidia's NVDLA-based tech and Amazon's AWS Inferentia chips.
One could argue that Intel is a bit late to the game here; given it dominates the data centre world, it could have been expected to have hit the AI chip arena earlier.
But then again Intel's Xeon CPUs were ticking along fine, and there's an argument that it's only now that more AI-centric hardware is needed as smarter tech takes off that the appetite for specific AI hardware is healthy enough to see Intel go to the efforts of creating the Nervana chips.
As for the average PC enthusiast, Nervana might not seem that relevant to you, but tech from the SoCs could trickle down into future Core processors, with CPUs that could have dedicated areas for carrying out machine learning tasks locally, such as voice recognition or clever computer vision on laptops.
That's a bit of speculation, but the with Nervana and Intel finally launching its Ice Lake 10nm processors, the chipmaker has demonstrated it's still one to watch, even if AMD keeps grabbing headlines with its third-generation Ryzen processors. µ
OK Google, explain 'imminent disappointment'
We'd have called it Bridget
Investor leverages his $1.2bn stake in PC maker
Social network handed over info in 88 per cent of cases