AMD GPU roadmap reveals HBM2-powered Vega and Navi

For the first time in a long time, AMD has an honest-to-goodness GPU product roadmap, and it names the company's next two GPU architectures: Vega and Navi.

Revealed at this year's Game Developers Conference during the AMD "Capsaicin" event, the roadmap is light on details, but it does hint at what the future might hold for AMD's GPUs. For starters, Vega is due to arrive early in 2017, not long after the release of Polaris. Interestingly, the roadmap explicity calls out Vega's use of HBM2 memory, a feature that was originally supposed to be a part of the upcoming Polaris architecture and used alongside traditional GDDR5 memory.

In an interview with PC Perspective, AMD's Raja Koduri hinted that Polaris would instead use HBM1, rather than its more accomplished counterpart. Of particular concern is that current HBM1 implementations are limited to 4GB stacks—HBM2 can go as high as 16GB—which could carry over to Polaris. Given that GPU memory requirements have skyrocketed over the past year, that could put Polaris at a significant disadvantage, particularly when Nvidia has confirmed that at least some Pascal parts will use HBM2.

AMD also confirmed that there will be at least two GPUs released under Vega: Vega 10 and Vega 11. This is similar to Polaris, which consists of the Polaris 10 and Polaris 11 GPUs, the latter of which is presumed to be a smaller GPU intended for laptops and mid-range graphics cards.

The final GPU on the roadmap, Navi, is set to arrive in 2018 following the release of Vega. Its listed features—"scalability" and "nextgen memory"—don't give much away, but that it arrives so soon after Vega hints that it may not be a direct replacement. Instead, it might be a mobile part—or, perish the thought, even an APU part.

Regardless, that AMD has a confident product roadmap marks a dramatic change for a company that has had a rather rocky few years. Sticking Raja Koduri at the top to lead the recently formed Radeon Technologies Group seems to be paying off. With Nvidia's GPU Technology Conference just around the corner, though, AMD shouldn't get too comfortable just yet.