Intel made announcements about upcoming AI solutions at 2019 Intel AI Summit. Those include Intel Nervana Neural Network Processors (NNP) for training (NNP-T1000) and inference (NNP-I1000) for cloud and data center customers, as well as Gen 3 Intel Movidis “Keem Bay” VPU.
We’ll focus on the latter in this post with Intel claiming similar performance as NVIDIA Jetson AGX Xavier at much lower power consumption with claims of up to 4.7 times more power efficiency when using ResNet-50 benchmark inference measurement using INT8 with a batch size of 1.
Considering Jetson Xavier AGX has a ~30W power budget, that would mean Movidius “Keem Bay” consumes around 6 Watts. Compared to Myriad X MA2085, the new Gen3 VPU is said to have more than 10 times the inference performance. Intel did not provide any TOPS figure, but considering the company announced 1 TOPS of neural compute performance for Myriad X, one may expect Keem Bay to deliver 10 TOPS. The higher performance & efficiency are partially made possible thanks to a new memory architecture with a 64-bit memory bus. The chip will be supported by OpenVINO Toolkit just like its predecessor.
Intel went on to claim the upcoming Movidius Keem Bay is about 4 times faster than NVIDIA TX2 and 1.25 times faster than Huawei Ascent 310, but with much higher power efficiency and considerably smaller footprint when expressed in inferences per mm2.
Movidius Keem Bay VPU is scheduled to launch in H1 2020 in a variety of form factors including M.2 and PCIe cards with one or more VPU.
You can watch the one-hour presentation at 2019 Intel AI Summit in the video embedded below. Keem Bay section starts at the 24:50 mark.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress
