The most popular and comprehensive Open Source ECM platform
Let us know more
AI Inference Processors: Dedicated Chips Handling Separate AI Workflows is not the Future
Every tech company nowadays of any weight has taken it on their own to design an AI chip specific to their business.
Amazon’s chip is called Inferentia, designed for ‘power workloads’ of thousands of teraflops of processing. Intel has a chip called Myriad 2 AI which is designed for complex viewing and imaging applications. Huawei had the Ascend 910 for data centers and the Ascend 310 targeting smartwatches and smartphones. Google has had their general-purpose TPU for some time. Imagination Power VR has GPUs and chips that target smartphones, smart cars, Iot and cameras. Nvidia has been the lead of the AI chip market. There’s Alibaba, Microsoft, and Facebook too.
But is it all just hype? Or at least severely overhyped? Some analysts think so. Richard Kingston, Vice President of Market Intelligence, said that “now we are in 2019, the reality of what AI can do and where it will feature is much more realistic.”
Kingston also argues that standalone AI chips are not the future. Adding AI as a separate chip with its own workflow while needing to coordinate with standard processors is complex and slows things down.
