Chip giant Intel's (NASDAQ:INTC) second-largest business by both revenue and operating profit, and arguably its most important business from a growth perspective, is its data center group, or DCG for short. Last quarter, DCG saw revenue and operating profit grow 27% and 64.8%, respectively.
Although the company spent a lot of time at its recent Data-Centric Innovation Summit event talking about the increasing role that sales of non-processor products -- like Ethernet adapters, Optane DC Persistent Memory, and silicon photonics connectivity products -- will play for DCG, the reality is that last quarter, nearly 92% of DCG's revenue came from so-called platform sales -- or, in other words, processors and their accompanying chipsets.
With that said, Intel did shine a light on its data center processor release plans at the aforementioned event. Let's take a closer look at what the company had to say.
Cascade Lake in 2018
Intel DCG chief Navin Shenoy said that the company's next data center processor, which goes by the code name Cascade Lake, is "on track to ship at the end of 2018."
This new processor family, the company claims, will run at higher frequencies than the previous-generation Xeon Scalable products; offer support for the company's Optane DC Persistent Memory (products that allow Intel's DCG to expand the company's served addressable market); incorporate so-called security mitigations to guard against the Spectre and Meltdown security exploits; and even incorporate the company's "Intel Deep Learning Boost" instructions to accelerate certain artificial intelligence workloads.
"This is going to be a very attractive product and I expect this to ramp very fast in 2019," Shenoy said.
Cascade Lake will be built using a derivative of the company's 14nm technology. Intel's 2016 server processors, which were based on the company's Broadwell architecture, were manufactured on the company's first-generation 14nm technology. The Xeon Scalable processors that launched last year, which are based on the company's Skylake architecture, are built using an enhanced version of that technology, marketed as 14nm+.
Back in 2017, Intel said that it had created a third-generation version of its 14nm technology, known as 14nm++, that offered a performance improvement over 14nm+. At the time, Intel said that 14nm++ offered a 26% performance improvement at the same power or a 52% power reduction at the same performance as the original 14nm.
Cascade Lake, Intel tells me, incorporates "[manufacturing] process enhancements beyond 14nm++."
The 2019 and 2020 offerings
Shenoy then said that "toward the end of 2019" it would launch a new processor known as Cooper Lake, which will be manufactured using a variant of the company's 14nm technology that Intel tells me will incorporate "further process enhancements beyond what [Intel] deliver[s] in Cascade Lake." The processor, Shenoy asserted, is "going to generate and deliver a significantly better generation-on-generation performance improvement."
It'll also support the bfloat16 numeric format, which the executive says "is principally used for [machine learning] training kinds of workloads."
Earlier in Shenoy's presentation, the executive disclosed that the company sold over $1 billion worth of its Xeon processors to customers looking to run artificial intelligence workloads. In a separate presentation, Intel executive Naveen Rao said that the market opportunity for data center processors sold to run artificial intelligence workloads would grow from $2.5 billion in 2017 to between $8 billion and $10 billion by 2022.
Although Intel has made it clear that it's designing chips specifically to run artificial intelligence workloads (the revenue I mentioned before came from sales of the company's general purpose Xeon data center processors), the company has said that it's "reinventing Xeon for [artificial intelligence]" via a combination of hardware and software advancements.
Shortly after the launch of Cooper Lake, Shenoy says that the company will introduce its first data center processor that's manufactured using its 10nm technology called Ice Lake. Cooper Lake and Ice Lake are set to share the same platform and will, in fact, coexist in the marketplace.
In an interview with AnandTech, Intel executive Lisa Spelman indicated that the Ice Lake data center processor is set to arrive in "the middle of 2020."