In an interview with VentureBeat, Raja Koduri — the head of Radeon Technologies Group — shed more light on the Polaris architecture.
AMD Polaris to get two GPUs this year
Raja Koduri confirmed that AMD/RTG has two versions of FinFET GPUs in development: Polaris 11 and Polaris 10. The first processor is allegedly planned as mid-range solution for desktops and something more of a high-end solution for notebooks. The latter should definitely make an appearance in enthusiast portfolio by replacing Fury X. Of course those codenames should end with something fancier and easier to remember as we get closer to the launch, as explained in this part of the interview:
We have some exciting hardware announcements as well. This is designed for FinFET. Our guiding principle for the Polaris architecture was power efficiency. We have the new naming scheme for our architectures. It’ll be based on galaxies, star systems, and stars. You’ll see more of this coming in the future. Polaris is the beginning of our journey through space. — Raja Koduri
What’s interesting Koduri already said that next architecture coming after Polaris (in 2017) should be 20% faster and more power efficient than Polaris this year.
When we set to design this GPU, we set a completely different goal than for the usual way the PC road maps go. Those are driven by, the benchmark score this year is X. Next year we need to target 20 percent better at this cost and this power. We decided to do something exciting with this GPU. Let’s spike it so we can accomplish something we hadn’t accomplished before.
Raja Koduri explaining Polaris 1X architecture:
Venture Beat: Is that with a generation coming in 2016?
Koduri: Yes. We have two versions of these FinFET GPUs. Both are extremely power efficient. This is Polaris 10 and that’s Polaris 11. In terms of what we’ve done at the high level, it’s our most revolutionary jump in performance so far. We’ve redesigned many blocks in our cores. We’ve redesigned the main processor, a new geometry processor, a completely new fourth-generation Graphics Core Next with a very high increase in performance. We have new multimedia cores, a new display engine.
This is very early silicon, by the way. We have much more performance optimization to do in the coming months. But even in this early silicon, we’re seeing numbers versus the best class on the competition running at a heavy workload, like Star Wars—The competing system consumes 140 watts. This is 86 watts. We believe we’re several months ahead of this transition, especially for the notebook and the mainstream market. The competition is talking about chips for cars and stuff, but not the mainstream market.
In summary, it’s fourth generation Graphics Core Next. HDMI 2.0. It supports all the new 4K displays and TVs coming out with just plug and play. It supports display core 4.3, the latest specification. It’s very exciting 4K support. We can do HEVC encode and decode at 4K on this chip. It’ll be great for game streaming at high resolution, which gamers absolutely love. It takes no cycles away from games. You can record gameplay and still have an awesome frame rate. It’ll be available in mid-2016.
The whole interview focuses on Virtual Reality, where Koduri explains that GPUs have to be capable of displaying content at 16K resolution and 240 frame rate, in order to offer mirror-like quality. This can take years to achieve, but at least we know where GPUs are heading.
Radeon 400 series speculation
After hearing the news from Koduri we finally have more information about the GPUs in Radeon 400 series. Because only two new GPUs are planned, we may still see few rebrands in Radeon 400 series (two graphics processor will not fill whole portfolio).
Polaris 10 and 11 are clearly replacing GPUs in mid-range and high-end segment. It is very unlikely Fiji would not return in Radeon 400 series. It is also very unlikely Polaris 10 and 11 would only be offered in one memory configuration. The mid-range Polaris should end up with various GDDR5 solutions, while high-end Polaris could benefit from adjustable number of stacks in High-Bandwidth-Memory (HBM). Pitcairn, a 4 years old design, will definitely be sent to retirement. We don’t know what will happen to Hawaii and Tonga though, since both are relatively new designs.
High-end Polaris, Fiji, Hawaii, Tonga and mid-range Polaris sound good to me. But let’s be honest, what we all to hear is whether high-end Polaris is actually faster than NVIDIA’s Pascal GP104, but the answer is still few months ahead.