Microprocessor giant Intel (INTC 5.41%) recently announced that it had begun shipping a "preliminary production" version of its next-generation Xeon server processor family. The new chips are based on the company's Skylake processor architecture and are manufactured in the company's 14-nanometer+ technology.
Intel's IvyTown server processor. Image source: Intel.
It was previously believed, thanks to a slide deck that leaked onto the Web last year, that those next-generation processors would pack up to 28 processing cores -- a respectable step up from the 24 found in the current-generation Broadwell-EP-based processor family.
However, a new leak -- this time, a purported listing of an engineering sample of the upcoming Skylake Xeon chips on Chinese website Taobao -- suggests that the top Skylake Server chips will have 32 cores. (Credit goes to AnandTech user Sweepr for first spotting this.)
Let's take a closer look at what this would mean for Intel's data-center business.
Some business background
Intel has stated in the past that it aims to deliver at least a 20% performance enhancement, SKU-to-SKU, when it introduces a next-generation server processor family.
This means, for example, that when this year's Xeon E5 2698 v4 is replaced by a Xeon E5 2698 v5, the latter will be at least 20% faster than the former. Intel has also in the past indicated that it aims to keep pricing roughly flat between generations for a given processor model -- something that's corroborated by the public pricing that Intel publishes for its chips.
So, in effect, performance per dollar for Intel's data-center chips is expected to go up at a steady, robust pace with each product cycle.
However, investors might know that one of the key growth drivers for Intel's data-center business has been an increase in average selling prices. That increase comes when data-center processor buyers opt to buy processors higher up in Intel's product stack.
So, for example, let's suppose that the last time a company bought Intel-powered servers, it opted for machines powered by Xeon E5-2660 chips, with a list price of $1,329. Then, when the server buyer comes back for new systems, it opts for Xeon E5-2680 v4 chips, with a list price of $1,745.
In that case, the server buyer isn't buying any more chips than it did last time, but it is buying more expensive processors to better serve its needs. From a sales perspective, Intel isn't selling more units to this customer than it did before, but it's selling more computing power than it did before -- and it's getting paid for it.
It's in Intel's best interests to keep adding chips to the top of its Xeon processor stack to continue to encourage upsell to increasingly powerful and expensive processor models.
About that 32-core Xeon...
With all that background in mind, we can now understand why Intel's reported 32-core Skylake Xeon processor is so interesting.
The highest-end Xeon E7 v4 processor was the Xeon E7 8890 v4. This part included 24 cores, which ran at a base clock of 2.2GHz and a maximum turbo clock of 3.4GHz. The prior-generation part, the Xeon E7 8890 v3, had 18 cores, which ran at a base clock of 2.5GHz and a maximum turbo speed of 3.3GHz. That's 33% more cores at roughly the same speeds, translating into about 33% more performance.
Had Intel's top Skylake Xeon processor included "just" 28 cores, then the generation-over-generation processor core count from the top Xeon E7 v4 to the Xeon E7 v5 would have gone up just 16.7%.
There's additional performance to be had from the new Skylake architecture, and Intel could have adjusted frequencies to make up for the smaller core count increase, but the latter may have had a negative impact on power efficiency.
By bringing out a 32-core model at the very top of its next-generation Xeon processor stack, Intel ensures that it can deliver at least the same kind of performance boost at the very top of its Xeon E7 v5 stack as it did with the Xeon E7 v4 chips.
Intel's data-center business is expected to grow by less than 10% this year. Given the recent slowdown in its growth rate, Intel can't afford to let up on the performance improvements that it brings to the table.
