On Aug. 5, PC processor giant Intel (NASDAQ:INTC) launched the first microprocessors based on its new Skylake micro-architecture. These first chips are "unlocked" -- which means the speeds at which the chips run can be tuned by the end user -- and aimed at PC enthusiasts as well as gamers.
These processors plug into motherboards that use Intel's LGA 1151 socket and the company's Z170 chipset, the company's latest "mainstream" platform. However, I believe that going forward Intel will stop targeting enthusiasts with this "mainstream" platform and will transition them to the company's "high-end desktop" platforms.
This transition, if my prediction is correct, should happen in the 2017 timeframe when the company launches its higher-end Skylake-E processors. Allow me to explain why I believe this to be the case.
One chipset for all?
A typical Intel desktop processor requires a second, external chip known as a "platform controller hub." With how Intel does things today, the "mainstream" desktop platforms have their own platform controller hubs (these are generally also used for Intel's high-performance laptop platforms), while the "high-end desktop" chips have a separate chip, which are essentially the same ones used with Intel's high-end Xeon processors.
However, it seems that this might be set to change.
Take a look at the following infographic from website IT Sitio:
Note that the successor to the one-socket Grantley platform is a platform known as Basin Falls. Note that the platform is expected to use the "Kaby Lake" platform controller hub rather than use the Lewisberg platform controller hub that the two-socket-and-above Purley server platform is expected to use.
Kaby Lake, as Intel officially announced on its most recent earnings call, is a third-generation 14-nanometer family of "mainstream" processors coming to PCs. The fact that the one-socket version of the Skylake-based Xeon processors (and, by extension, the Skylake-based high-end desktop processors) will use the same platform controller hub as the "mainstream" Kaby Lake processors is particularly interesting.
Is Intel laying the foundation for an enthusiast transition to high-end desktop platforms?
By making the high-end desktop platform use the same platform controller hub as the mainstream platforms, I suspect that Intel is planning to try to transition PC enthusiasts almost completely away from the mainstream platforms and toward the high-end desktop platforms.
However, Intel will have to make some broader changes to its processor offerings if it's going to actually pull off such a transition. For example, in the mainstream platform, Intel offers everything from $50 dual core Celeron chips to high-performance unlocked quad-core Core i7 processors.
In contrast, on the high-end desktop platform today, Intel generally offers three Core i7 processors, with the cheapest model starting at just under $400 and the top model going to $999. Intel would need to expand the range of its processor offerings to include chips at the approximately $250 level.
With Skylake for high-end desktops, I could see the following potential product stack:
- $249 -- Six-core/six-thread unlocked Core i5 processor.
- $349 -- Six-core/12-thread unlocked Core i7 processor.
- $599 -- Eight-core/16-thread unlocked Core i7 processor.
- $999 -- 10-core/20-thread unlocked Core i7 processor.
I believe that such a product stack would be warmly received by PC enthusiasts and would allow Intel to maintain a comfortable lead over its competition, should the Zen-based products that rival Advanced Micro Devices (NASDAQ:AMD) plans to roll out in late 2016 for this space turn out to deliver strong per-core performance.
Two major hurdles
There is one major hurdle that I see here: motherboard price. One of the attractive aspects of Intel's mainstream platform for enthusiasts is that the motherboards for the high-end desktop processors are, on average, pricier than their mainstream counterparts.
Additionally, the high-end platforms tend to require quad-channel memory for optimal performance (i.e., memory must be inserted in multiples of four), while the mainstream ones only need two channels for optimal performance, making the high-end platforms potentially more expensive to set up.
For example, a Z97 board for mainstream Haswell can be had for as low as $85 on Newegg ($75 after a $10 mail-in-rebate card), while the cheapest X99 board for high-end desktop Haswell that I can find is $220 ($199 after $20 mail-in-rebate card).
I don't think these issues will be show-stoppers, though. Although there is almost certainly an inherent cost-adder associated with supporting the high-end desktop chips over the mainstream chips, the current high-end desktop boards are relatively "premium" boards from the get-go.
If Intel widens the range of products that it offers on the platform and positions it as the enthusiast platform, then board makers could develop lower-cost boards to go along with them. Right now, the high-end desktop chips start at relatively high prices and as a result the boards are packed with features commensurate with premium processors.
One more thing
A last thing worth noting is that we've seen is that the high-end desktop chips tend to lag the mainstream chips in terms of either architecture or manufacturing technology (depending on whether it's a "tick" or a "tock" year).
To show this point, the current high-end desktop chips are currently using Intel's Haswell architecture while the latest mainstream platform chips are now beginning a transition to Skylake; a two-generation difference. The high-end desktop platform should transition to Broadwell in early 2016, narrowing the gap to one generation.
However, with the delay of Intel's 10-nanometer ramp and associated Cannonlake mainstream processors, I think there's a chance that the delta between the two platforms in terms of architecture and manufacturing will shrink substantially.
I believe that Intel's mainstream platforms will continue to use the Skylake CPU architecture until the second half of 2017 when the company transitions to 10 nanometers and the Cannonlake architecture. Cannonlake should essentially be a "refined" Skylake and probably won't bring more than 5% per-clock performance.
The Skylake high-end parts that will probably be in the market when Cannonlake arrives should be able to more than make up a slight per-clock performance deficit with higher core counts.