The big news of the day, based on a leaked Intel roadmap, is that Broadwell
won't launch in 2014. Instead, we'll see a Haswell
refresh with unknown performance characteristics (slightly higher clocks are the best bet) and, late in the year, a "Haswell-E" server part on the X99 chipset. Ivy Bridge-E, the six-core LGA2011 CPU, is still set for later this year, on the X79 chipset. The big assumption coming out of this is that Intel has delayed 14nm production and that Broadwell wont' ship at all in 2014. Based on sources we've spoken to, this isn't accurate.
The flaw in the current story is that it implicitly assumes that desktop is the focus of Intel's business and that the company would naturally deploy new process nodes and technologies there before bringing them out anywhere else. For decades, that's been true -- but from Intel's perspective, investing money in desktop first and foremost is yielding smaller and smaller returns. When Intel did the major Haswell briefings in early May, the company talked candidly about a 6-8 year replacement cycle for desktops.
Last year, Intel discussed moving to a desktop cycle that was essentially focused on every other year, with BGA parts on new ticks and tocks available on the standard cadence, but socketed chips launching at longer intervals. There were also rumors that slumping demand in the market for new parts had led Intel to delay some of its 14nm transition plans, and the difficulty of moving to new process nodes is a factor that's impacted everyone across the industry.
Slow Refreshes Change Everything:
Haswell, Intel's latest and best microarchitecture
to date, is focused on reducing mobile power consumption
, improving IPC (instructions per clockcycle), and driving into smaller form factors. The 8-10% IPC improvement it offers over Ivy Bridge is actually quite significant given the current state of CMOS manufacturing, but the response from readers to the new desktop chips has been pretty lukewarm. In mobile, Haswell is more exciting. This fact underlines a hard truth of the modern era -- desktop performance alone isn't exciting enough to drive purchases.
When Intel chose to integrate a voltage regulator into Haswell, it made a decision that would improve the chip's low power capability, but increase thermal density. That was always going to hurt high-end scaling. In other words, Intel made a decision that prioritized mobile and low power consumption over high desktop performance. As the benefits of scaling to new process nodes fall, Intel is repeatedly choosing to focus what benefits it can grasp on the low power consumption side of the equation.
What's the point of moving to Broadwell? Lower power. Do desktop users care about lower power? Not very much.
For Intel, that means reevaluating where it puts its efforts. If pushing to new process nodes means building lower TDP hardware, Santa Clara is going to do so -- it badly wants to be moving into smartphones and tablets, and it needs to maintain a manufacturing advantage to make that happen. The notebook refresh cycle remains substantially faster than desktop, for multiple reasons: Laptops are more difficult to upgrade, repair, and are more susceptible to damage due to their intrinsic portability. Given the relatively high probability of hardware failure within 3-4 years of owning a laptop, Intel can count on a replacement cycle that's fairly consistent. Tablets and smartphones are likely to follow similar trajectories, provided it can win more SKUs in these markets. Again, that's a reason to push forward on Broadwell, not pull back.
I suggest that this desktop roadmap, even if accurate, isn't actually the whole picture. Desktop just isn't that important to Intel, anymore -- and judging by sales numbers, it's that important to anyone else, either.