Final Thoughts on AMD's Core Technology Update Presentation
Moving Along to 2015 and 2016
This is the year that we will start to see some serious shifts from AMD when it comes to supporting ARM. They announced their Project Skybridge initiative that will enable the sharing of the same socket with an x86 processor as well as an ARM based processor. This looks like it will be an extension of the current AM1 socket that currently houses the low power Kabini products.
On the x86 side we will see the introduction of a 20 nm based product based on the Puma+ architecture. Puma is the basis of the current Beema/Mullins parts. These will be low power products that will be true SOCs, just as the current AM1 Kabinis are. These apparently will encompass full HSA functionality with the latest GCN based graphics cores. These will not be high performance cores, but rather will be aimed at the low power/energy efficient market.
The ARM side gets a bit more interesting. AMD will go over the Cortex A57 cores with a fine toothed comb and implement as many power saving tricks to make this a much more efficient core than what ARM is licensing to its other partners. This seems to be AMD’s way of dipping their toes into designing their own ARM cores. This could actually be the first ARM core that will fully implement HSA functionality. AMD will be combining their A57 design with the latest generation GCN architecture to enable HSA.
AMD will have two separate 64 bit architectures that will fit into the same socket. This will allow quite a bit of flexibility for the company when selling these products to end users. AMD will have a unified socket architecture that will support these very different parts. This is a net positive for the company as they will have a solid foundation for these products since they share the same infrastructure. For consumers this means they have some decent options for how they want to implement their end solution. I am not talking buyers off of Newegg, but rather groups higher up the food chain that will provide their end users with customized, low power solutions that can address the Windows/Linux/Android environments.
AMD’s ARM offering is a taste of things to come. AMD will work hard to fully integrate their semi-custom A57 product into the HSA fold. It is a good starting point for the company and developing a workflow and gather traces of what ARM is good (and not so good) at.
This looks to be a year the rubber really meets the road. Just as we were promised “Fusion” from when AMD bought ATI in 2006, AMD is promising two entirely new architectures to address multiple markets, all featuring HSA functionality.
AMD has made the jump from being a licensee of specific cores to being a licensee of the entire ISA. AMD is now standing on the same ground as Qualcomm, NVIDIA, Samsung, and a select few others. They are able to custom design their own ARMv8 based cores that will differ significantly from what ARM and its other partners offer. This allows AMD a lot of leeway to implement new features and structures that could give it an edge.
We do not know much about AMD’s ARM design other than it is codenamed K12. We have no idea what kind of power range this product will span, but my guess is that it will be relatively low powered. Think 65 watts and below. AMD does not list “High Performance Server” or “High Performance Desktop” as targets for this chip. Instead, they are looking at dense server, embedded, semi-custom, and ultra-low power client.
x86 is also getting attention here with a brand new core that will also be introduced in 2016. There is no code name attributed to this product, but it appears to be stepping away from the Bulldozer architecture that didn’t turn out to be as shiny for AMD as they had hoped. Jim Keller is heading the teams that are designing both the new ARM and x86 cores. He was one of the brains behind the original Athlon and Athlon 64 and left the company right before the A64 was released. AMD has a lot of design experience with solid IPC/low power designs (Jaguar, Puma+) as well as being able to clock products up fairly high (current Vishera, Steamroller cores). This new core will also have to stretch from top to bottom in terms of power consumption. I highly doubt we will ever see a 125 watt TDP desktop processor from AMD again, but we might see one in the 65 to 95 watt range.
AMD has proven to be a resilient company in the face of some serious financial hardships. Rory and his band of merry executives have really turned the company around. This turnaround has a price though. AMD is no longer competing with Intel on the high end/high performance x86 markets. I doubt AMD will ever produce another FX-8000 or FX-9000 type of part ever again. From here on out it looks like much more power efficient APUs will be the desktop offerings. We can only hope that AMD reaches the same level of IPC and power efficiency that Intel has with Haswell. AM3+ is the last socket of its kind, and from here on out we will have products that will have to fit in FM2+’s 100 watt TDP envelope.
Now, we have to wonder if the Excavator core will ever see the light of day? We will see Excavator based APUs in 2015, but how far will AMD go with this family of products? Roadmaps that we have previously seen may in fact be obsolete with where the company is moving. AMD has wisely decided to stop competing with Intel exclusively, and instead is competing against all of the ARM partners as well. This strategy does have its risks, but AMD is in a better position to compete against the likes of Qualcomm, Samsung, and others. They have certainly trained well for this new market, considering that they have survived (and at times thrived) some forty years while going head-to-head against the 800 pound gorilla of the semiconductor world.