Hot on heels of inspired speech by Rory P. Read, new CEO of AMD who probably has more emotion and energy than all of former two managements (Hector’s and Dirk’s) put together, the stage was occupied by a new member of the senior AMD team, Miss Lisa Su.

Lisa is newly appointed Senior VP and General Manager of Global Business Units. She gave an overview of roadmap of upcoming products, mysteriously not disclosing any desktop parts sans APUs.

AMD Strategy for 2012-2013: Execution

As you can see in the image above, AMD roadmap for 2012-2013 is focused on Execution, and that word was said several dozen times over the course of the Financial Analyst Day 2012. The company is focusing on executing in three major areas, which is graphics – we’re going to see three GPU architectures in as many years – GCN i.e. Southern Islands in 2012, Sea Islands in 2013 and Canary Islands in 2014. As you can see, there will be quite a lot of islands in AMD’s GPU world.

AMD Client and Graphics Roadmap for 2012-2013: 40nm, 32nm and 28nm product mix

Sea Islands is perhaps the key GPU architecture for AMD in the next couple of years, since it brings the support for virtual memory, meaning GPU and CPU will finally be able to address the same memory space. Game developers have been calling on this since 2001, but all the work at DirectX AB went unanswered by both AMD and NVIDIA. However, with the HPC segment up for grabs there is no space for leniency and Sea Islands will be "the" architecture upon AMD will base its future. Canary Islands will be "wrapping up the job" in 2014, but it is still too early to talk about that particular architecture.

The Performance APU side shows that this year will see the arrival of Trinity APU, which according to Mark Papermaster, Senior VP and CTO of AMD – is already shipping to clients, a second generation APU. Trinity is a very interesting part, since it combines Piledriver x86 cores, Northern Islands GPU core and Southern Islands Video Processing Engine. Piledriver is the successor to somewhat controversial Bulldozer core, meaning AMD is speeding past discrete desktop parts and putting the new x86 core in an integrated part right away. Trinity will get succeeded by Kaveri, which is again bringing "Steamroller" cores, third generation x86 core combined not with Southern Islands, but rather Sea Islands GPU architecture from the get go. This part is being touted as the first 1TFLOPS APU, or the first 1TFLOPS CPU for the desktop and notebook designs.

Brazos is getting refreshed with a 2.0 version: read Turbo Core technology and USB 3.0 support, as AMD is buying time for Kabini, which will bring a second generation low power x86 core codenamed "Jaguar". Kabini will also utilize Sea Islands architecture and fully support key HSA features such as the unified memory, which will significantly reduce the power output and more importantly, internal on-die and external PCB complexity. On the tablet side of things, Hondo (in reality, a very low power Brazos, i.e. ULP version of Ontario core) will get succeeded by Temash, which is also based on the Jaguar x86 core.

Most important part about Kaveri, Kabini and Temash is that all of these parts are SoC’s. That’s right, AMD is going head to head against Intel’s Haswell SoC architecture but unlike Haswell, AMD will offer a performance SoC part as well, not just the Ultrabook/Ultrathin 17W version.

AMD 2012-2013 Server roadmap: Terramar, Sepang canceled, new product line is based on AMD's Desktop socket AM3+

On the server side, everything changes. AMD has decided not to proceed with 20-core Terramar and 10-core Sepang, just as the company canceled the 10-core desktop parts. Instead, the company will stick through 2012 and 2013 using the same 32nm SOI process at GlobalFoundries before completely new generation of processors based on "Excavator" cores takes over in 2014. For 2014, AMD is expected to migrate directly to 14nm Fully-Depleted SOI process.

Also, this is the first time we’re seeing that a company will stick to a same CPU socket for three generations in a row. Since the debut of DDR4 memory got delayed to 2014, AMD decided to keep the same sockets – AM3+ for 1P Web Hosting/Microserver, C32 for 1P and 2P and G34 for 2P and 4P Enterprise Solutions. Do note that 8P is no longer being mentioned, as the datacenter focus shifted from a CPU-centric to a CPU/GPU hybrid platform, with four CPU sockets being the maximum you want to put in a server. Given that we’re seeing a big uptake in number of 2P+2GPU designs, this is an understandable course of action.

In March of this year, AMD will debut its 1P Web Hosting/Web Serving/Microserver platform based on Zurich processor. Zurich is not much more than an Opteron branded, Enterprise-grade version of the Orochi silicon which debuted as FX-8000 series. Zurich Opteron will be available in 4- and 8-core versions. In 2013, Zurich will be replaced by Delhi with the same number of cores, this time based on the "Piledriver" x86 core.

C32 will be occupied by sexa- and octa-core Seoul, as 10-core Sepang is dead. Besides IPC improvements, don’t expect much until 2014. The big daddy, Socket G34 will now be occupied by Abu Dhabi processor, consisting out of four and eight highly clocked or 12/16 normally clocked "Piledriver" cores. In this move, we see a repeat of a neat trick Intel did with their sexa-core Xeon, which had four cores disabled and ran at 4.4GHz.

Cities, Racetracks… and again, Ferrari
From the looks of it, AMD is now naming their servers for cities, even though there is just one city that doesn’t have a racetrack, and that’s Zurich: Interlagos is a racetrack in Sao Paolo, Valencia has two race tracks with the same name, Abu Dhabi racetrack is bordering with Ferrari World, fascinating 5-star hotel and a marina, while Seoul has a racetrack which wants to take over the Korean GP. Delhi of course, was the place of spectacular Indian GP last year. All platforms are named on Ferrari facilities anyways (Maranello for Socket G34 etc.).
After all, Intel’s delays on Sandy Bridge-EP (March 2012, 7-month miss) and Ivy Bridge-EP (pushed to March 2013) are giving AMD much needed breathing space, as we won’t see a high-performance Haswell-based Xeon until 2014.

For the end, we leave you with an image of upcoming CPU architectures by AMD:

AMD Opteron Future Technology: 2012-2013-2014

All in all, we’ll be bringing much more in-depth look of the path AMD is now taking in the upcoming days. Stay tuned.