ADFS 2011 Wrapup & AMD Lynx Platform Tests



Company: AMD
Author: James Prior
Editor: Charles Oliver
Date: July 18th, 2011

AMD A-series APUs

AMD held their inaugural Fusion Developer Summit in Bellevue, Washington, from June 14th through 17th. The two and a half day event filled with keynotes, more than 90 technology sessions geared to developers, a product launch, and one or two next-generation hardware teases. As first conferences go it was extraordinary, with a smooth and polished feel reported elsewhere to be reminiscent of a certain FruityCo's media expertise. That's entirely right, and suitable for the introduction of what AMD considers their most important product - the mainstream accelerated processing unit (APU).

AMD Fusion Developer Summit, 2011

The summit's running theme was OpenCL, the open standard compute language first brought to light by Apple, and now stewarded by the Khronos Group. The members of the Khronos Group are numerous and resplendent, and include ARM, AMD, Intel, NVIDIA and other important industry leaders. OpenCL is an API used for harnessing the power of a graphics processing units (GPU) capabilities for parallel programming, under the banner of heterogeneous compute - hardware agnostic as long as the independent hardware vendor (IHV) offers support. Intel, NVIDIA and AMD all now provide support for OpenCL v1.1, the most recent version.

Despite the AMD Fusion Developer Summit title, the feel was very much about programming for heterogeneous compute platforms, using OpenCL as the common standard. This was very refreshing, with developers, speakers and companies alike feeling free to openly discuss how different hardware platforms performed and their strengths and weaknesses, even in the face of the AMD Fusion branding adorning the hallways, presentation slide decks and Fusion Experience Zone where meals were served. The whole shebang could very well have been the Heterogeneous Compute Developer Summit, powered by AMD.

Monday, June 13th was the first day of check-in for the summit, although the desk was open for the whole conference. Different color badges and tag lines denoted AMD employees, members of the press, speakers, and attendees. AMD had hoped to host around six hundred attendees, and well exceeded that number by the end of the week. Despite the great attendance, the sessions, keynotes, breakfasts, lunches and networking times proceeded like clockwork, a testament to the competent organization of the event and the staff at the Meydenbauer Convention Center. In the afternoon, there were several pre-summit tutorials about OpenCL and DirectCompute, and how to use them, including a look at the memory model on Fusion APUs. Fusion APUs allow the OpenGL and OpenCL APIs to leverage 'Zero Copy' where data is accessed by both the CPU and GPU in the APU, but not copied as traditionally seen in CPU with discrete GPU systems.

Manju Hegde & John Taylor

The Summit opened on a high note on Tuesday, June 14th, with the double act of Manju Hegde and John Taylor introducing former ATI executive and now AMD Senior Vice President and General Manager for the AMD Product Group, Rick Bergman. Overnight the Sabine platform for mobility APUs had launched, introducing the first mainstream accelerated processing unit with discrete performance level graphics.

Mr. Bergman gave us a more comprehensive look at AMD's strategy, focusing on the platform as a whole and using graphics products to differentiate from competitor products. AMD had planned to unveil of the Sabine platform at the morning commencement but instead moved it up to the previous evening. The reasoning for this became clear as Bergman pulled a demonstration of next year's APU out the hat, running in a notebook. With Llano only 12 hours old, AMD showed the world next year's model (codenamed Trinity) and promised a 50% increase in performance - aiming for around 1TFLOP of parallel compute power in 2012's APU. Trinity was first seen in hardware just a few short days before, at Computex on June 1st. Like Llano, it uses GlobalFoundries 32nm SOI process but everything else is different; both the CPU and GPU core architecture are replaced. Having working silicon in a notebook indicates that the platform is also ready, possibly indicating that Trinity APUs will use the same APU socket as Llano for faster platform development and perhaps drop in upgrades. This was the first running Bulldozer architecture product sighting at AFDS - but not the last.

Bergman's opening segued into the first keynote of the conference, The Programmers Guide to the APU Galaxy by AMD Fellow Phil Taylor. First, however, we'll talk about the mainstream APU platforms, Sabine and Lynx.