Yesterday, AMD had its first financial analyst day in several years. These events are important for several reasons, not least of which is because they give investors and enthusiasts a cohesive picture of where AMD is and where it hopes to go. Unlike other press events, which tend to be focused on a single product or product family, analyst days are broader and cover a wider range of topics.
AMD shared a lot of material at this event, and it was easily the most positive report the company has given in years. There’s good reason for that. Ryzen 5 and Ryzen 7 have been highly successful, the PS4 Pro refresh is in-market, Microsoft’s Project Scorpio is moving towards a holiday launch, and there’s a new high-end Ryzen challenger moving into play this summer. I’ve already shared thoughts on several specific AMD projects, so I’m going to round up some of the other material here, and share some thoughts overall.
Vega and Ryzen feature heavily in AMD’s plans to compete across the desktop and mobile space. To-date, we’ve only seen Polaris and the Ryzen 5 and 7 families, but AMD has bigger plans. The slideshow below steps through AMD’s overall evaluation of its Total Addressable Market (TAM) in the PC space, and how Ryzen is expected to change it.
If you’ve flipped through the slideshow, you’ll have noticed that AMD is making some potent claims about its future APU performance. Specifically, AMD is claiming it can deliver 50 percent more CPU performance, 40 percent more GPU performance, and a 50 percent reduction in power consumption, thanks to a combination of better CPU efficiency, 14nm improvements relative to its older 28nm process, and the increased efficiency and performance per watt of its new Vega GPU core.
Obviously we’ll need to see shipping hardware to evaluate these claims, but they all pass the sniff test. Just moving to 14nm would have given AMD a significant improvement over 28nm, and the Ryzen CPU cores is vastly more power efficient than the old Excavator core. The situation is a little different on the GPU side, where we expect that most of the performance improvements are likely delivered by improved memory bandwidth. On-die GPUs are almost always memory-bandwidth limited, and a shift to DDR4-3200 would give the GPU core far more bandwidth than the DDR3-2133 that Kaveri typically used. We don’t know exactly which 7th generation APU configuration AMD was comparing against, but nothing here seems unusually odd.
As for Vega, AMD did indeed demo and promise that a version of its high-end GPU would be in market by the end of June, but the GPU it showcased — the Radeon Vega Frontier Edition — is also explicitly aimed at content creators, engineers, and HPC workloads. It’s supposedly faster than Nvidia P100 in one benchmark (DeepBench, a test I’m not familiar with and haven’t evaluated).
The Radeon Vega Frontier Edition offers 1.5x more single-precision floating point performance than Fury X (13 TFLOPS vs 8.6 TFLOPS) and 3x the FP16 performance. Tempting as it might be, we do not recommend making any comparisons between AMD and NV cards based on TFLOPS performance. Not only is it a theoretical metric, it’s only useful for comparing GPU performance between cards in the same family (and only barely useful there).
Vega’s consumer iteration was conspicuous by its absence, but we’ve heard more information on that topic is coming in the not-too-distant future.
I have to qualify my subhead because much of what AMD discussed yesterday, like its upcoming server launches, deep learning initiatives, and mobile APUs haven’t launched yet. As of today, right now, there are still a number of areas where the company’s fortunes are poised to improve, even if they haven’t done so yet.
But for the first time in a decade, it feels like AMD is finally pointed in the right direction. No, Zen doesn’t completely sweep Intel, who retains a moderate advantage in single-threaded code. AMD’s overall recovery is still fragile — it needs its APU, server, and high-end GPU launches to all go well, and future iterations of Vega and Zen need to arrive on-time and without drama.
But I agree with Patrick Moorhead, of Moor Insights and Strategy, who told me, “AMD has a tremendous amount of upside. AMD doesn’t currently operate in the server market and with Epyc, they will do that. AMD also doesn’t participate in the high end graphics and machine learning market and very well could with Vega.”
“Could,” of course, does not mean “will.” There are going to be plenty of gimlet eyes cast towards AMD, which has a history of not delivering everything it promises. But there was a time when AMD was ascendant, with a clear leadership position in the multi-socket server market, a 64-bit CPU when Intel was stuck on 32-bit, and industry-leading performance in nearly every benchmark on the planet. Ryzen and Vega are the best chance AMD will ever have to reestablish itself as a serious player in every computing segment. So far, the company seems determined to achieve it.