Last March, AMD unveiled a roadmap for its GPUs that predicted a one-two punch. Polaris would arrive in mid-2016, while Vega, the company’s big GPU follow-up, would slip in at the end of the year. This arrangement meant that AMD and Nvidia would effectively launch on different cycles, with Nvidia doing a typical top-down refresh, while AMD went with an unusual midrange cycle first.
When AMD revealed these plans in March, it was actually a modified launch plan from what the company had told reviewers to expect. At its Sonoma event in December 2015, AMD implied it would launch a larger variant of Polaris first, followed by a brand-new architecture, Vega, later in the year. Those plans had changed by March, with Fat Polaris vanishing off the roadmap and Vega pulling in. Now, according to AMD’s own investor presentations, these plans have changed.
Here’s the thing about product launch timing. When a company only gives a vague date like “H1” or “H2,” it almost always means the product will launch towards the rear of the relevant window. AMD repeatedly stated it would launch Polaris in H1 2016 and it debuted the card in late June — just before H1 turns into H2. Companies typically flip to half-years as opposed to quarterly targets when they don’t want to admit a new product is farther away than investors or consumers might like. “H1” sounds like “Q1” and offers hope for an earlier debut, while “Q2” forces the company to admit that a product won’t debut until at least April of the following year.
I want to stress that these naming conventions are typical, not absolute. Maybe AMD is being conservative. Maybe they’re planning to debut the card earlier than expected and tweak Nvidia’s nose. But historically, when AMD says “H1,” they mean “May / June timeframe.” Other companies use the phrase similarly.
If Vega has really been pushed back to May or June, it’s a serious blow to AMD’s graphics strategy. A six-month delay between Nvidia and AMD’s refresh cycles was acceptable, a year-long delay is not. Vega was supposed to be AMD’s chance to catch up to Nvidia by delivering a top-end GPU refresh that could match Pascal in terms of overall performance and power efficiency. By the time Vega actually sees the light of day, Nvidia will be well at work on their own follow-up.
The only silver lining in all of this is that GPU sales are actually a comparatively small part of AMD’s profits and Polaris’ midrange position should still help them sell hardware at important mass-market price points.
The good news on Zen is that AMD is still targeting a Q1 2017 launch date. We’ve discussed performance several times in the last few weeks, and Zen is probably far more important to AMD’s bottom line than its graphics business. While AMD no longer breaks out GPUs as a separate segment, during the time it did so (approximately 2007 – 2012) its GPU business was worth between $300 – $400 million per quarter, with only a 3-4% net profit margin. Total APU and CPU sales were typically $800 – $1.2 billion per quarter over the same period. Profit margins on CPUs were far better than profits on GPUs.
It makes sense for AMD to focus on kicking the products out the door that stand to make it the most money. It’s also important, however, to manage expectations about what Zen is and isn’t likely to deliver. While I expect Zen to be much faster than Excavator and offer performance far more competitive with Intel, it probably won’t close the IPC and clock speed gaps in a single leap. The inevitable pull-down on clock speeds and AMD’s lower TDP targets (Zen supposedly tops out at 95W) will also have an impact on which frequencies AMD can hit. Intel’s eight-core chips are all 140W TDP designs and they top out at 3.7GHz. Piledriver may have hit 4.7GHz base / 5GHz Turbo with the FX-9590, but we’ve heard no rumors that AMD intends to offer a 220W TDP part this time around.
Turning to AMD’s semicustom business, the company notes that it expects to recognize $1.5 billion in new revenue over the next 3-4 years thanks to Project Scorpio and “x86 and ARM opportunities.” Oddly, the PS4 Neo isn’t listed here, even though AMD is universally believed to be driving Sony’s next-gen console.
$1.5 billion in revenue over 3-4 years works out to $500 – $375 million per year, but this is either low-balled or reflects a sharp decrease in expected revenue from semicustom designs. For the full year ended December 31 2015, AMD recorded $2.1 billion in the sales for the enterprise, embedded, and semicustom market. We know that most of this revenue is coming from PlayStation 4 and Xbox One sales, and we know both platforms are set for a major refresh cycle in 2016/2017. AMD may be playing things conservatively and predicting lower sales than actually occur. AMD’s operating income on console sales isn’t great — it earned $215 million on sales of $2.186 billion in 2015 and its profits from semicustom are expected to continue to decrease as the PlayStation 4 and Xbox One age. The company has always disclosed that its margins would be smaller over time, but has declined to give details past that.
One final note: For all the talk of the so-called “PC Master Race,” it’s amusing to look at AMD’s balance sheet and realize that consoles, not PC hardware, is keeping the company alive. In 2015, AMD’s computing and graphics business reported sales of $1.805 billion for the year and a net loss of $502 million. The console business, in contrast, reported sales of $2.186 billion and $215 million in profits. With console margins shrinking, AMD needs Zen to fire on all cylinders when it launches in six months — and turn the CPU business around.
As for Vega, none of our contacts at AMD responded to confirm or deny the delay rumors. We will update this story if we hear from the company.