People have always been fighting over silicon. Ever since IBM launched the first Intel-powered PC/AT or 80286 in 1984, it’s been a battle between the likes of HP, Compaq, Dell, and IBM to see who gets the latest chips to market first. Fast-forward to present times, and although AMD and NVidia have made an impression on the market, the majority of cloud real estate is still being built on Intel. Though home PC sales may have dropped dramatically compared to what they used to be, Intel chips are still a hot property among the public and private cloud vendors.
It’s funny that the latest Intel chips are now appearing directly in the cloud rather than through traditional server vendors like HP or Dell. What’s funnier is that if it weren’t bad enough that the cloud was eating away the physical server market, they now seem to have first pick with regards to top-of-the-line computer chips as well. This is an indication of the rapid growth of cloud computing and the steady rate at which enterprises are moving their datacenters to the cloud.
Intel’s declining PC market
Intel has been at the forefront of microprocessor technology since they invented the world’s first microprocessor in 1971. In fact, they have been so dominant that they have often been compared to Microsoft with regards to monopolization, and, like Microsoft, have also seen their share of anti-competitive lawsuits. The alignment with Google that was announced in November 2016 seems to have paid off since Google is the only one offering the new Xeon chips, codenamed Skylake. Though a small move in the high-level chess match that’s being played out for cloud dominance, it’s a significant move nonetheless.
As organizations shift slowly away from the declining PC market, datacenter business revenue rose by about 9 percent to $4 billion, and this probably explains Intel’s disposition to favor cloud vendors over physical server vendors like HP and Dell. The illusion that the new Skylake announcement creates that Intel favors Google over other cloud vendors might look ridiculous at first, especially since Intel and Microsoft are old friends and AWS and Intel have a deep working relationship.
“Wintel” Azure
Windows computers working on Intel chips were the Android phones of the PC era, and the dynamic duo aren’t done yet. Azure is the first public cloud to include an Intel-initiated and container-oriented Clear Linux OS. Clear Linux is only for x86 chips and aims to let customers use the isolation of VM technology along with the deployment benefits of containers. Again, this doesn’t mean the “Wintel” relationship is exclusive. Rather, this looks to be more on the lines of Microsoft repairing their old anti-Linux image and supporting anything and everything open source and Linux related.
In fact, Microsoft being as ambitious as ever, is already moving past Intel with regards to its Azure cloud and has begun testing Windows Server OS on ARM-based servers made by Qualcomm and Calvium. This could also be the reason Microsoft is on the naughty list and didn’t get first pick of the Skylake goodies.
ARMs and ammunition
Another company interested in ARM architecture is none other than public cloud leader AWS. Their subsidy “Annapurna Labs” announced it would start selling ARM-based chips for storage, IOT, cloud and networking. ARM is probably to Intel what Linux was to Windows. Some call it a “thorn in the flesh,” and it could be a real problem for Intel in this sector. Similar to the free software movement, an ARM-compatible chip called “Amber” is available through the open source hardware community called OpenCores under the condition that all updates or modifications are shared with the community. Their customers include NVidia, Qualcomm, Broadcom, Apple, Samsung, and Applied Micro, among others. The 32-bit ARM architecture is the most widely used with reference to mobile and handheld devices already, and with AWS and Microsoft creeping in on Intel’s last safehouse in the cloud, maybe it’s no surprise they’ve chosen Google for Skylake.
Custom silicon
This choice could also be part of a larger bid from Intel to build customized silicon for the giant cloud providers based on their specifications. Around two years ago, AWS announced the E5-2666 V3 (Haswell) processors that were custom-made by Intel for C4 instances. Around the same time, another announcement was made that “customizable cloud chips” were being manufactured. Diane Bryant of Intel was quoted as saying “it’s not clear how many customers will need this customization but the ones that do will be big,” and she was right. The new announcement states, “We optimized Skylake for Google Compute Engine’s complete family of VMs.” Brian Krzanich, Intel’s chief executive, was apparently trying to be economical while finding new business in the cloud. The new features in the Xeon chip are made possible by fusing the chip with a programmable integrated circuit. Not only does this allow Intel to customize chips en masse, but it also allows them to be economically viable.
To be fair, Intel has never played favorites, and it can’t really afford to. AWS has promised its customers it will have the Skylake processors early in 2017 and they would be “customized” for AWS. This customization of silicon seems to be the trend here, and different public cloud vendors probably have different specifications that they would like met based on their customer requirements. The fact that Intel is willing to bend over backwards to accommodate these giant cloud vendors, who are known to be ruthless negotiators, just goes to show how important the cloud sector is for Intel now.
Chasing a mirage
Though Google is making much about the new processor capabilities and is even “screening” potential clients to see if their needs match the new offerings, it won’t be long before AWS and Azure have their own customized Skylake chips. The new chips are said to double the floating-point performance for the heaviest calculations when compared to previous versions of the chip and improve application performance by up to 30 percent. The fields most noted to feel the improved performance are in graphic-intensive operations like scientific modeling, 3-D rendering, analytics, genome research, and engineering simulations. What’s interesting is that’s almost exactly what AWS says the older Xeon Haswell processors are best suited for as well.
If anyone is actually planning to shift to Google Cloud just for the new Skylake chips, they better be prepared to keep shifting, as the cloud giants play cat and mouse with the latest silicon. What we do know is that though Moore’s law states that the number of transistors in an integrated circuit will double every year, that law could soon meet its demise. If the latest Skylake chips are being designed for Google, the next lot could be designed for someone else. However, adding an instance that runs on the Skylake chips to your hybrid cloud sounds like a great idea. Apart from basic bragging rights that comes with using the latest tech, it will probably also be a lot of fun to test your applications in terms of how much better the new chips actually are.
