The last major CPU microarchitecture from Advanced Micro Devices (AMD +0.89%) was Bulldozer, which was launched in 2011. Three revisions since then -- Piledriver, Steamroller, and Excavator --have brought improvements, but this entire generation of CPUs has been a disaster for AMD. The company's server market share has essentially vanished, and its PC business has been generating massive losses. Then-AMD VP Andrew Feldman had some harsh words for AMD's efforts back in 2013: "Bulldozer was without doubt an unmitigated failure. We know it."
Bulldozer, and its subsequent revisions, simply couldn't match the performance of competing processors from Intel (INTC 1.49%). Single-threaded performance was downright awful. In Anandtech's original review of AMD's Bulldozer-based FX-8150, it pointed to a comparable Intel processor as having a 40%-50% single-threaded performance advantage.
AMD's next microarchitecture, called Zen, is set to launch in 2016, and it's clear that the company is looking to correct its previous mistakes. The focus will be on increasing single-threaded performance, and AMD is making big promises on that front. Because AMD's business has deteriorated so much over the past few years, Zen needs to succeed. If AMD can't get back in the game with Zen, the company may not survive.
A new focus
There are two ways to increase single-threaded performance. The first is to increase the clock speed, which causes the processor to use more power and generate more heat. The second is to increase the number of instructions that a core can process each clock cycle.
Bulldozer's focus on multi-core performance left it at a big disadvantage compared with Intel chips when it came to instructions per clock, or IPC, and while the three revisions brought improvements, it wasn't nearly enough. Excavator, for example, improved IPC by as much as 15% compared with Steamroller.
AMD expects the first iteration of Zen to improve IPC by a whopping 40%. The first products, set for a 2016 launch, will be high-end desktop CPUs, meant to compete with Intel's high-end offerings. Zen is expected to be built on a 14nm process, and with Intel recently announcing that its 10nm process will be delayed to 2017, AMD and Intel could both be selling 14nm CPUs at the same time next year.
If AMD can launch Zen on time, and if the chips meet expectations, AMD could have a winning product on its hands. But neither of those things is a given. Bulldozer was supposed to close the gap with Intel -- but it didn't. AMD's latest high-end graphics card, the Fury X, was supposed to have a big advantage over NVIDIA thanks to its use of high-bandwidth-memory. It didn't.
AMD has been good at making promises, but it needs to get better at delivering on them.
Last chance
AMD is saddled with quite a bit of debt, about $2.27 billion at the end of the latest quarter. The good news is that none of that debt matures until 2019. At that time, $600 million of debt will need to be paid off or refinanced, with another $450 million maturing in 2020.
AMD is on the hook for about $160 million of interest payments annually, and the company has been burning through cash in recent years.During the past twelve months, AMD's free cash flow was a loss of $187 million, but with $829 million of cash remaining, AMD should be able to fund itself through 2019, even if things don't get any better.
But things need to get better. If Zen fails to revitalize AMD's PC and server businesses, the company will be in much worse shape financially by the time 2019 rolls around. AMD would likely need to raise additional debt in order to fund its operations, something that may not even be possible if its finances have deteriorated. And refinancing maturing debt could lead to even higher interest payments, making any sort of recovery even more difficult.
It looks like AMD is making the right choices with Zen, focusing on the area where Bulldozer was weak. With Intel essentially having a monopoly in the server-chip market, along with incredible margins, Zen should be able to win some market share as long as its performance is in the ballpark of Intel's server chips, something that Bulldozer was unable to accomplish. But if Zen ends up being a flop, AMD's chances of survival will be seriously reduced.
