Intel Optics Closes All Openings Into The Server Market

| About: Intel Corporation (INTC)

Summary

Bigger, more powerful data centers and supercomputers require higher speeds and more processors.

Can't be done with existing switching fabrics and speeds.

Intel has been co-developing the solution with key customers for more than 10 years.

Deliveries from production start this year.

Every so often someone starts up a company or a new product line in an existing company to address the server market with a non-Intel-based product. AMD (NASDAQ:AMD), ARM (NASDAQ:ARMH), Qualcomm (NASDAQ:QCOM) and others are now looking to expand into the profitable and growing server market. But Intel (NASDAQ:INTC) just closed and locked the front door.

Data centers have been growing, both in numbers and in size to the point where there are significant problems that needed solution:

1. The amount of power being used in data centers is on the order of 2.5% of the available power in the US. Looking at it in another way, data centers use approximately all of the power generated by wind in the US.

2. The size of some of these data centers is so large that they can't be expanded further without degrading latency and bandwidth. And the cumulative throughput is so high that the bit error rate further degrades the performance because of the use of forward error detection and correction.

Intel and others have been aware of these problems for some time and concluded more than 10 years ago that something had to be done to address the problems.

And they decided to disaggregate the highly compacted sub-modules into compute nodes, memory and switches to improve predictability, serviceability and software portability.

Intel decided that the best solution would be to reduce the power dissipated in the network fabric by going optical. This would essentially eliminate power dissipation in all of the electrical wiring, carry the signals between the various server modules, memories, and system I/O. But this required development of key technologies and components.

Intel was already working on fiber networks but now they needed a higher speed network, and with the disaggregated architecture they needed optical switches. They contracted with key researchers in universities in the US and they published some of this work about five years ago. They took out a raft full of patents and then went into the development of all the required components in this new architecture. It's called Omni-Path. And it's described in detail here.

And it's been demonstrated by 20-plus companies - members of the Intel Fabric Builders Companies (ISC November 17, 2015). And it's being delivered to customers within the next month or so.

It's software compatible at the applications level with the Intel True Scale architecture.

It allows continuous improvement through higher and higher levels of integration and through replacement of electrical interconnect with highly integrated optical interconnect and switches, including on-chip lasers and fiber interconnect inside of server chips, including MIC class chips. The process for doing this has already been developed and it involves composite wafers that include Silicon (for the microprocessors and light guides and the laser resonator cavity) and Indium Aluminum Phosphide (for the lasers).

Integrating the lasers removes power lost in the (formerly copper) interconnect and in the electronic circuits that were driving the copper interconnect. The power lost in the laser drivers is much smaller and the power lost in the fiber interconnect is negligible.

The latency of the architecture is greatly improved because the multi-port switches allow simpler switching networks. The error correction delays are almost eliminated because they are done within the fabric of the system rather than by forward error correction.

No, this is not April 1. This is the product of a 10-year long (at least) managed program for the development of a new fabric for data centers and supercomputers that allows continuous scaling to very large fabrics with more than 17,000 nodes and very small chips containing optical transmitters and receivers.

And, ARM, AMD and Qualcomm are going to have a hell of a time breaking into this and avoiding patent infringement. Besides, they have 10 years of catch-up work to do.

Disclosure: I am/we are long INTC.

I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.