For as long as I can remember, I have been following roadmap leaks in the PC space from Intel (INTC) and Advanced Micro Devices (AMD), and generally speaking, these have been pretty straightforward to interpret. However, in this new world of hyper-segmentation and intense focus on mobile, things have gotten a little tricky to understand, particularly on the Intel side of things. In this article, I'd like to "set the record straight" so that investors do not get the wrong idea from the numerous sites on the web spreading what appears to be faulty analysis of good leaks.
14nm Is Not Delayed To 2015
Intel bears have pointed to the following leaked roadmap slide in an attempt to prove that Intel is running into problems with its 14nm process and that processors based on that process will not be available until 2015:
However, this is actually not the case. See, following the 22nm "tock" known as Haswell, Intel indeed has a processor core known as Broadwell. The rub, unfortunately, is that this processor will apparently not be coming to the socketed desktop format, but will instead only be coming in "BGA" (that means the chip is soldered to the motherboard) format, implying that it is a part aimed exclusively at mobility and integrated form factors such as high end tablets, notebooks, and Ultrabooks.
Now, from a business perspective, this makes complete sense, so let me explain.
Why Broadwell Won't Be Coming To Socketed Desktops
When Intel released a socketed "Ivy Bridge" for the desktop, it seemed pointless from the point of view of the folks who buy these things. Sure the die size was smaller, and the integrated graphics were better, but these are the sorts of things that mean absolutely nothing to the socketed desktop crowd. The majority of folks buying these chips will either find the integrated graphics already "good enough", or if they need more 3D gaming performance, will be buying separate cards from either Nvidia (NVDA) or Advanced Micro Devices. Finally, while "ticks" do wonders to reduce power consumption, this is not something that people buying "towers" (where socketed chips/boards go) care too much about. As long as the electric bill isn't enormous and the heat dissipated isn't terrible, then the customer base here cares much more about performance than anything else. And, did I mention, that this customer base is a rather niche market?
See, Broadwell will be the same sort of thing; a minimal improvement on the CPU side of things from Haswell (think 3-5% performance per clock improvement), but a solid improvement in power consumption, a big jump in graphics (Broadwell gets the new "Gen 8" graphics micro-architecture), and overall an excellent improvement for mobile users who care about performance within a power constrained, battery-life driven environment.
So, from Intel's perspective, why should it release a new chip for a user base when that new chip will not improve meaningfully on any vectors that this user base cares about? Broadwell is for all-in-one desktops, notebooks, Ultrabooks, and tablets, and it makes sense for Intel to not waste any of its 14nm capacity in this space when its 22nm (which should be very mature and cheap by then) parts will do the trick.
In addition, it seems that Intel has something even better than "Broadwell" planned for the socketed desktop: "Haswell-E".
Haswell-E: First Details Revealed
The enthusiast crowd ("-E" stands for "enthusiast") doesn't want integrated graphics. This crowd wants gobs of performance, lots of cores, insane memory bandwidth, and doesn't care so much about power because they are very likely running systems with 1200W power supplies, 2-3 200W+ graphics cards, and more memory than an iPad has storage space. Intel, instead of bringing a socketed Broadwell which will be met by enthusiasts with a yawn, is apparently going to be stepping up its game on the "-E" parts in the Broadwell time frame:
This is what desktop enthusiasts want. More cores, lots of memory bandwidth, big caches, lots of PCIe Gen 3 lanes, and all of this with only a moderate regard for power consumption. So, while Broadwell and the next gen 14nm Atom known as Airmont soak up the 14nm capacity for the markets that really need the latest and greatest in power saving and graphics (all at a competitive price, so smaller dies help), Haswell and Haswell-E will service the customers who would rather build towers and workstations. Intel wins because it doesn't have to worry about validating additional Broadwell SKUs for the desktop socketed platform (where it would receive a lukewarm reception at best) and instead can focus on where the money is: mobile. At the same time, Intel can use cut down server parts as "high end desktop" parts aimed at the hardest-core enthusiasts.
For investors, this is a very clear win. The best part of all of this is that the only "competitor" in this space is AMD, and that company will not be transitioning to its 28nm architectures until the end of 2013/early 2014, so it is very unlikely that AMD will be able to (profitably) compete on performance or cost against Intel's 22nm high end desktop parts.
So, the takeaway is clear: Intel's 14nm process will debut 1H 2014 with Broadwell aimed at all-in-ones, notebooks, Ultrabooks, and high end tablets where the real volume will be. The high end desktop will stay longer on 22nm Haswell until the 14nm "tock" known as Skylake, where Intel will be able to offer a meaningful CPU performance increase and platform update, which will be a compelling offering to the socketed desktop crowd (and it will be even more compelling as the users will not be upgrading from a "tick" but from the last "tock"). This all makes perfect business sense, and I believe that investors should take heart in knowing that Intel is doing the right things in each of its markets from a competitive and cost structure standpoint.