» Current | 2019 | 2018 | 2017 | Subscribe

Linley Newsletter

Year in Review: Data Centers Accelerate AI Processing

December 25, 2018

Author: Bob Wheeler

In a year dominated by AI, you’d be excused for thinking traditional server CPUs were dying. Yet Intel’s data-center revenue (mainly server processors) grew 26% in the first nine months of 2018, reaching $16.9 billion. By contrast, Nvidia’s data-center revenue (mainly accelerators) rose an astounding 70% from a much smaller base, reaching $2.3 billion in that period. Although AI-accelerator growth slowed in 2018, both established vendors and startups continue to invest heavily in this technology.

Owing to AMD’s resurgence, the merchant market for server processors is refocusing on the x86 ISA, undercutting the struggling Arm camp. Although the first-generation Epyc fell short of sales expectations, AMD is already sampling its 7nm Rome processor, which should drive further share gains in 2019. Meanwhile, Intel is shipping (but hasn’t officially announced) Cascade Lake-SP, which brings new features to the existing Purley platform. Although Arm got a boost from Amazon’s launch of Arm-based EC2 instances, the cloud provider’s in-house Graviton processor doesn’t create direct demand for merchant vendors.

Although cloud operators such as Google have deployed AI-specific hardware, merchant vendors have mostly repurposed existing designs for AI. Recently, the horde of startups developing AI-specific architectures began to deliver production devices. Graphcore, Habana, and Wave are three such companies targeting data centers. These vendors are all chasing Nvidia, which continues to optimize its Tesla GPU cards for AI, where they primarily accelerate deep-neural-network (DNN) training.

Intel and Nvidia continue to benefit from the growth in cloud data centers, but in-house ASICs can displace their products. Examples include Amazon’s processors, Google’s TPUs, and Huawei’s AI chips. Merchant processors and accelerators must demonstrate superior efficiency if they hope to succeed at hyperscalers. In the longer term, a big shakeout in AI-chip vendors is coming, but next year should see more new entrants than failures.

Subscribers can view the full article in the Microprocessor Report.

Subscribe to the Microprocessor Report and always get the full story!

Purchase the full article

Events

Linley Fall Processor Conference 2019
October 23 - 24, 2019
Hyatt Regency, Santa Clara, CA
Linley Spring Processor Conference 2019
Held April 10 - 11, 2019
Proceedings available
More Events »

Newsletter

Linley Newsletter
Analysis of new developments in microprocessors and other semiconductor products
Subscribe to our Newsletter »