» Current | 2020 | 2019 | 2018 | Subscribe

Linley Newsletter

Analysts’ Choice Winners for 2019

January 21, 2020

Author: The Linley Group

Habana’s Goya inference chip is our Best AI Accelerator. On ResNet-50, it outperforms Nvidia’s T4 by 4x for offline applications and a huge 7x for real-time applications with a batch size of one. Unlike most of Nvidia’s competitors, Habana has demonstrated strong performance on multiple neural networks. Our Best PC or Server Processor is AMD’s Rome, officially called the second-generation Epyc processor. Sporting up to 64 cores, Rome easily grabbed the performance lead in mainstream two-socket servers. Rome tops Intel’s latest Xeons in cache size and memory bandwidth and is also the first x86 processor with PCI Express Gen4.

Huawei’s Kirin 810 stands out as the Best Mobile Processor. Compared with the mid-premium Qualcomm Snapdragon 730 and MediaTek Helio P90, which debuted in phones at about the same time, the Kirin 810 delivers better CPU performance, better GPU performance, and better AI-Benchmark scores. For Best Processor IP, we selected Arm’s Mali-G77, the first GPU to implement that vendor’s new Valhall architecture. Arm expects the new GPU to increase area efficiency and power efficiency by 30% compared with last year’s Mali-G76. These gains place the G77 well ahead of other licensable GPUs.

Our Best Embedded Processor award goes to Syntiant’s NDP101, a low-power SoC for smart speakers. Instead of accessing the cloud, it employs a tiny AI engine to immediately recognize up to 64 verbal commands, all while consuming 0.2mW, far less than any competing device. We selected Broadcom’s Monterey switch as Best Networking Chip because it’s unique in addressing 5G wireless infrastructure. It’s the only switch chip to integrate CPRI ports and perform radio-over-Ethernet (RoE) encapsulation, handling fronthaul for 4G remote radio heads. It also provides 25Gbps Ethernet ports for 5G remote units.

Intel wins our Best Technology Award for its Foveros 3D-stacking technology. By stacking multiple die vertically using proprietary vias, it’s much denser than attaching chiplets using an Embedded Multi-Die Interconnect Bridge (EMIB). Foveros allows Intel to manufacture each die in a different IC technology to reduce cost.

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

Free Newsletter

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

Events

Linley Fall Processor Conference 2020
October 20-22 and 27-29, 2020 (All Times Pacific)
Register Now!
More Events »