Facebook is reportedly planning to build its own custom chips to deploy in its data centers in a push towards AI and machine learning.
Wall Street Journal reported that Facebook (FB, Financial) is hiring engineers with experience in building custom processors from scratch. Google (GOOG, Financial), Amazon (AMZN, Financial) and Microsoft (MSFT, Financial) are already known for building in-house custom chips in order to reduce reliance on external suppliers. Google AI engine tensor flow utilizes the capabilities of a custom home grown chip. Amazon is also building a chip to improve AI capabilities of Alexa. Facebook seems to be the last to jump on the AI and machine learning hardware bandwagon.
Points of interest
The development of custom chips, if successful, will reduce Facebook’s reliance on third party silicon suppliers for its data centers.
We already know Facebook’s white box strategy of using home grown equipment like Wedge (networking switch) and Voyager (transponder) in the networking arena. Now, with the chip development, the company will effectively remove some of the component providers for network devices from its data centers. This could very much lower Facebook’s operational costs in the long-run.
This development isn’t a threat for Intel and Advanced Micro Devices
The company is hiring engineers for application specific processors (ASICs) and field programmable arrays (FGPAs). Facebook can potentially use its application-specific processors in its upcoming hardware devices, including video chat devices and virtual reality devices. Moreover, the company is interested in creating a Google-like voice assistant.
Regarding FGPAs, machine learning inference is a great example, along with network protocol acceleration/inspection, image processing as shown, and others that can benefit from the reprogrammable nature of FPGAs, noted Robert Hormuth, vice president and fellow, server CTO, Dell EMC.
This indicates that Facebook has no plans to compete with the likes of Intel (INTC, Financial) and Advanced Micro Devices (AMD, Financial) in the central processing market. Facebook might be trying to create semiconductors for its networking equipment, data centers and artificial intelligence engines.
In an astonishing revelation, Jeff Bezos claimed that the number of prime subscribers has crossed more than a 100 million.
"In 2017, Amazon shipped more than 5 billion items with Prime worldwide, and more new members joined Prime than in any previous year – both worldwide and in the U.S.," noted CEO Jeff Bezos in the latest shareholder letter.
Point of interest
Paid prime membership includes prime video.
Netflix (NFLX, Financial) has 125 million members across the globe. It seems like Amazon is catching up to Netflix in terms of the reach of prime video. However, time spent on the platform is higher for Netflix as compared to Amazon Prime. Nonetheless, 100 million prime subscribers provide Amazon with a foundation for competition in the streaming arena.
EPYC continues to garner attention; Cray, a manufacturer of supercomputers, has added Advanced Micro Device’s EPYC to its CS500 portfolio of super computers.
Cray Inc. (CRAY, Financial) added EPYC 7000 processors with a support for up to 32 cores and eight DDR4 memory channels per socket. Cray noted that EPYC's programming libraries not only support but enhance the performance of EPYC server processors. According to Advanced Micro Device’s vice president of data centers, Cray is the first vendor to offer the optimized programming environment for EPYC.
Points of interest
Intel has been the lone supplier of CPUs to Crays in the recent past.
It’s been five years since the last time Crays offered a design powered by AMD silicon. This indicates that Advanced Micro Devices has come a long way, as EPYC has gained traction for the past year or so.
High performance computing is EPYC’s core strength.
This development also consolidates the view that EPYC is quite competitive to Intel’s offerings when it comes to high performance computing. According to AnandTech, EPYC beats Intel in most of the memory intensive workloads. In short, Advanced Micro Devices will give Intel a run for its money in the servers space.
Software optimization support: not very healthy for EPYC?
Because Advanced Micro Device’s vice president of data center noted that Crays is the first vendor to offer optimized programming environment, it raises the question of whether optimized solutions are available from server providers likeDell (DVMT, Financial) EMC. If not, Advanced Micro Devices will have to face an uphill battle against the established server base of Intel.
Disclosure: I have no positions in any stocks mentioned and no plans to initiate any positions within the next 72 hours.
