Intel has unveiled new details of its product roadmaps for 2015 and beyond, and it’s now clear that the company is pushing Skylake full steam ahead, despite Broadwell’s exceptionally late arrival. Intel now expects to introduce a bevy of 14nm products based on both architectures in 2015, with additional 14nm Atom hardware replacing Bay Trail as well.
First, a bit of a mea culpa is in order — I initially predicted that Skylake could slip into 2016 based on the Broadwell delay and the unlikely prospect that Intel would launch multiple architectures within the same year. It’s now clear that the company does intend to go this road, though it’s still possible that it will arrange its introductions in a manner that doesn’t leave Skylake overwriting just-launched Broadwell hardware.
Desktop and mobile
Here’s how the introduction is going to happen, courtesy of ZDNet:
Core M (Broadwell’s ultra-mobile flavor) is already shipping in a few early devices and will ramp up through the end of the year. By this coming spring, it’ll have branched out and taken over the product stack Haswell currently occupies, with fifth-generation refreshes for the entire laptop/convertible market. In the back half of 2015 we have new budget product launches, including Braswell (more data on that in a moment), and finally Skylake with its new architecture on 14nm.
That’s it for Intel’s official statements. According to WCCFTech, Intel will also launch new desktop parts next year, with a Core i7 5000 unlocked CPU (Broadwell-K) and a second set of desktop SKUs dubbed the Core i7-6000 family, or Skylake-S. Broadwell-K is reportedly compatible with the Z97 family of chipsets that are already shipping, while Skylake-S will require a new motherboard.
Broadwell is the 14nm refresh of Haswell, with a die shrink and a handful of minor improvements to the CPU, but not much more. Skylake, in contrast, is the full architecture refresh — so what are its (rumored) features?
No More FIVR: Intel’s Fully Integrated Voltage Regulator has always been rumored to cause the company headaches. Whether or not that’s true, we do know that putting the voltage regulator next to the CPU can cause additional heat to build up on-chip, and that heat only increases as the CPU clock and voltage increase. Skylake is supposed to dump the FIVR design, though it’s possible the ultra-low power CPUs will retain it to maximize power savings.
New instruction sets: Skylake is expected to include the updated form of AVX with support for 512-bit registers (AVX 3.2/512F), along with specialized instructions for executing SHA-1 and SHA-2 securely. Intel MPX (Memory Protection Extensions) and Intel ADX (Multi-Precision Add-Carry Instruction Extensions) are also expected with Skylake.
Core M (top) vs. Haswell. Broadwell-Y is a much, much smaller package — ideal for tablets.
New GPU enhancements: While integrated graphics remains unacceptable for a majority of gamers, steady improvement in this area has become the norm, even from Intel. Skylake will retain the maximum 128MB EDRAM L4 cache, but should incorporate additional GPU-side performance improvements. DirectX 12 should be supported on this GPU family, and we’ll see more desktop chips debuting with some version of the L4 cache.
Skylake will also support DDR4, though WCCFTech thinks this will be SKU-specific, with some chips staying on DDR3 through the end of 2015. There’s also rumor of a new “UniDIMM” form factor, that could bring interoperability with DDR3 or DDR4 in the same board. Desktop Skylake chips will still top out at four cores + HyperThreading; if you want more than eight threads you’ll need to move to the E-variant of the platform. Intel may also push its “no cables” idea for the platform (we were dubious on that score but are willing to be convinced).
Exactly how much raw performance Skylake will offer remains to be seen. Haswell looked fabulous on paper, but in non-AVX optimized code the actual real-world performance gain was around 7%, clock-for-clock, over Ivy Bridge. Devil’s Canyon pulled away from the older Ivy Bridge parts, but it did that the old fashioned way — with clock speed.
Whether Intel will attempt to clock Skylake more aggressively is an open question at this point. The first SKUs won’t be unlocked, so desktop enthusiasts may choose to wait for the Skylake-K variant, coming in 2016.
Intel’s tablet future…
Tablet and mobile
The other major component of Intel’s roadmap is its tablet and cellular business. In 2014, the company set a target of shipping 40 million total Atom SoCs, and was willing to take a bath on cost to get there. By the end of the year, Intel will have racked up $3 billion or more in total losses — but it’ll likely make its 40-million-SoCs goal.
In 2015, Intel wants to move towards a cost structure that allows it to make money. The launch of SoFIA, the TSMC-built, 3G-capable, budget cellular platform is seen as critical to that goal. Intel’s Rockchip collaboration will ship in 2015 as well, and the company hopes to extend that partnership over the long term.
Intel wants to ramp its baseband products extremely quickly, shifting from SoFIA 3G with dual-core Atom processors at the end of this year to quad-core Atom and 3G by the middle of 2015, and quad-core Atom+LTE sampling by the end of 2015. Presumably these dates refer to product samples rather than vendor introductions; it’s not uncommon for vendor launches to lag chipset production by 6-9 months. That suggests we could see LTE-based Atom products in market by the early part of 2016. Intel will shift SoFIA production back to its own fabs and a 14nm process node at some point in 2016, which implies a late 2016/early 2017 introduction for those products.
In tablets, Intel will continue to lean on the combination of its separate XMM 7260 modem and the Bay Trail/Moorefield combination until the back half of the year, when its 14nm die-shrink of Atom (codenamed Airmont) will be paired with a refreshed, purportedly Broadwell-based GPU.
Then Goldmont (refreshed Atom architecture) debuts in 2016, followed by a unified SoFIA device that will combine the tablet and smartphone markets with a single, addressable product that can match both.
Putting it all together
Intel’s roadmap continues to reflect a company in transition. While it may be on much firmer footing than its erstwhile competitor, AMD, it has to make fundamental changes to its cost structure if it wants to compete with ARM products and still make money. The SoFIA project is a short-term plan to change that. Meanwhile, making Broadwell potentially compatible with already existing motherboards could win Intel some small kudos from the enthusiast community, while it looks to Skylake to deliver the performance gains the HPC and workstation markets still require.
Most of Intel’s key challenges will be in the mobile and tablet markets throughout the next two years. The desktop refresh cycles remain interesting, but the lack of serious competition through at least 2016 means that this space should be fairly quiet. Intel needs to continue to deliver on its promises of decreased mobile power consumption — and if early reviews are anything to go by, it may need to have some conversations with OEMs about proper positioning for Core M products.
If you’re on the fence about upgrading to Devil’s Canyon now or waiting for Skylake in 2015, I don’t necessarily expect it to make much difference. Waiting may buy you a chip that’s 5-10% faster in specific metrics, but the days of meteoric gains from shipping code in conventional software are long gone. There’s no 14nm magic on the horizon that’s expected to deliver the kind of clock speed gains Intel would need to pull off a 15-25% year-on-year scaling, and absent stronger competition, the company is under relatively little pressure to push the performance envelope in its core markets.