Chipmaker Intel (NASDAQ:INTC) just announced that the company, in partnership with supercomputer vendor Cray (NASDAQ:CRAY), has won a contract to build two supercomputers for the U.S. Department of Energy's Argonne National Laboratory. The more powerful of the two computers is called Aurora, while its lesser sibling is known as Theta.
Theta comes in 2016, Aurora in 2018
According to Intel, the first system slated to come online is Theta. Expected in 2016, Theta will deliver 8.5 petaflops of compute performance and consume 1.7 megawatts of power. Intel says that it will be powered by both Intel Xeon processors as well as high-performance, compute-oriented Xeon Phi processors.
The second system to come online is the more powerful Aurora. Intel says that this system is expected to deliver 180 petaflops of compute power. This is said to deliver 18 times more computing power than the IBM Blue Gene/Q-based Mira supercomputer that is currently in use at the Argonne National Laboratory. Intel says that this increase in computing performance comes at only 2.7 times more power consumption.
Aurora is expected to come online in 2018, will have over 50,000 compute nodes, and should be based on Intel 10-nanometer Xeon Phi processors, codenamed Knights Hill.
Intel's HPC strategy is looking really good
Intel has been bullish on its opportunity in the high-performance computing market for quite some time, and as an Intel investor, I am really liking the company's strategy.
Not only does Intel offer standalone Xeon processors in this market, but it offers the high-performance computing tuned Xeon Phi chips in both co-processor form as well as in standalone form. Additionally, Intel is able to offer much more than processors into such systems.
To illustrate, note that Intel says that the Aurora computer will feature Xeon Phi processors, Intel's Omni-Path interconnect, a "new non-volatile memory architecture," and "advanced file system storage using Intel Lustre software."
For this deal, then, Intel is bringing processors, flash memory, interconnects, silicon photonics, and even software. It is practically a one-stop shop for high-performance computing, which means more dollar content and ultimately, more revenue per supercomputer for Intel.
What does this mean for the Intel data center group?
Intel says that the deal is reportedly valued at "more than $200 million." While impressive, a deal of this size is not going to move the needle for Intel, which generated just under $55.9 billion in revenue last year.
A bunch of such wins in aggregate is, of course, enough to have a financial impact. Seeing Intel win high-profile designs with its suite of technologies is a pretty good sign that its technology offerings in this space are competitive.
Putting this into context, Intel executives say that the "technical computing" (which includes high-performance computing) market segment is expected to grow significantly between 2014 and 2018, as shown below:
As long as Intel continues to deliver competitive Xeon and Xeon Phi processors and related platform components, then this forecast seems pretty reasonable.
Speaking of "related platform components," Intel did say that in 2014, its data center revenue was roughly 88% CPU-related revenue and 12% "non-CPU" revenue. By 2018, Intel expects that to shift to about 78% CPU-related revenue and 22% "non-CPU." Given what we are seeing with the Aurora supercomputer, where Intel is providing many non-CPU components, this seems like a perfectly plausible prediction.