Monday, February 13, 2012

AMD's plans for future: A new AMD? (Part 1)

Last week AMD (Advanced Micro Devices) held its Financial Analyst Day (FAD) which is basically an annual event to assess, analyze and provide information about the company and its strategy in various market sectors. Generally there are lots of financial tidbits presented and discussed in such an event which are of no concern to the common user, unless of-course one is a stockholder or investor. It's the technical side that attracts people like us because this is where we get glimpses of future products and technologies. Lots of eyes were on this year's FAD and how AMD tackles issues like desktop performance, graphics integration and mobile computing. Finally we've got some very interesting information in our hands and we'll try to analyze those.

AMD Now 

But before we go into details about future road-maps and directions of AMD, it'll be wiser to have a look on the current scenario. Those familiar with computer technologies from last couple of decades always give credit to AMD for being competitive and innovative. AMD is the 2nd largest provider of microprocessors and GPUs in the world. AMD has a long history of competition and rivalry with Intel, which is a much bigger company than AMD, in the field of x86 computing. Also we all know about the famous rivalry between AMD/ATI and Nvidia, which has been a highlight of PC graphics industry for decades now often resulting into great products for the consumers. On a modern day PC, the most important two components are arguably its CPU and GPU. Historically AMD has performed well in both of these markets and that's not an easy feat to achieve considering the kind of competition it had to face from the likes of Intel and Nvidia! But things have been rough for the company recently, specially in the CPU front.

As we all know, Intel has been the dominating force when it comes to the x86 CPU market and though AMD had its moment of glory with the Athlon processors and 64bit computing, it didn't take long for Intel to take back the performance crown with its Core2 (Conroe) series of CPU in 2006/2007. From then on, the performance gap between Intel and AMD chips has only widened, specially in the servers and high-end desktops. While some may point to Intel's reported anti-competitive and monopolistic practices as a reason, couple of delayed and poorly executed product launch from AMD's part didn't help them either. The original Phenom (Agena) processors from AMD were disappointing and while 45nm Phenom II (Deneb/Thuban) had much better price/ performance competitiveness against Intel's 2nd generation Core2-Quad (Wolfdale) CPUs, when Intel launched chips based on new Nehalem micro-architecture in late 2008, the performance gap was restored once again. At that point Intel was well ahead of its perennial rival by a good margin but at least AMD competed well on price, providing excellent value for money chips such as the Phenom II x4 955BE or Phenom II x6 1065T. But it was obvious that AMD wouldn't be able to keep up with Intel's relentless "Tick-Tock" cadence- where Intel continues to deliver new micro-architectures and smaller process technologies, for much longer. And that was exactly what happened in 2011 with Intel introducing the new Sandybridge architecture while AMD had no answers to it. AMD's reply came in form of Bulldozer (BD) micro-architecture based FX chips but while novel and innovative in concept, these CPUs were late, premature and under-performing! No, BD is not the epic fail some reviewers try to claim but at the same time it's no Intel killer either. The story in the server space is somewhat similar, where the Buldozer based new Opteron processors did a tad bit better than their desktop variants but that's about it.

AMD's execution was rather poor in 2011 in the CPU front and that was not entirely its own doing. As you can remember, AMD is a fab-less company right now meaning it has to depend on others for manufacturing the chips. This is one of the biggest reasons why AMD fails to deliver in a new process node, 32nm in case of Bulldozer. Intel's manufacturing department, which is arguably the best in the business, is one big difference between these two rival chipmakers. AMD depends heavily on manufacturing partners like GLOBALFOUNDRIES (Glo-Fo) and TSMC, for the success of its CPU/GPU designs. This means if there is an issue with Glo-Fo's process node, it will affect AMD too and is exactly what happened last year. Globalfoundries' transaction from 45nm to 32nm process node was not very smooth due to some serious yield issues and as a result AMD suffered badly from short supply of products (mainly Llano and BD based CPUs) throughout the year. This is one problem Intel didn't have to face when it launched Sandybridge.

AMD's graphics wing though did exceptionally well in recent years. When AMD acquired ATI back in 2006, many were skeptical about what kind of an impact it will have on both of the companies. Initially things didn't look great as ATI GPUs were under constant pressure from Nvidia's offerings. But as time went on, the combined forces of AMD and ATI proved to be a strong challenge for Nvidia. With AMD releasing its now legendary Radeon HD48xx (RV770) series of graphics cards in 2008, intense competition came back to the graphics market. Since then it has been a seemingly endless battle for supremacy between AMD and Nvidia, with both companies delivering great products in every market segment, much to the joy of the end users! AMD has enjoyed good growth in the graphics market in recent times and while Nvidia had the performance crown last year with its Fermi series of GPUs, AMD's own Northern Islands (Cayman, Barts) GPUs were very competitive. By the end of 2011, AMD took back the performance crown with the introduction of Radeon HD7970 which is based on a completely new GCN (Graphics Core Next) architecture and it is the fastest single GPU graphics card as at the time of this writing.

And finally we have the AMD APUs, the only saving grace for the company last year as some may call it. APU stand for Accelerated Processing Units, which is essentially a CPU with an integrated GPU on the same die. This is one area that AMD has been working hard for years now under marketing names like Fusion and more recently Vision. The idea is very obvious- many of today's computing workloads depend heavily on graphics processing power of the system and it's not only gaming but HD entertainment, video trance-coding, image editing and even web browsing can be benefited from more powerful graphics. Now, AMD has deep understandings of both CPU and GPU and it only makes sense for the company to combine them in a single package. The end result is APU. AMD's first effort came in form of its Brazos line-up of APUs using low power Bobcat cores and integrated radeon graphics. Brazos based E/C/G series chips were impressive and delivered better performance and efficiency than Intel's low power Atom CPUs and became very popular amongst the netbook/ultra-portable OEMs. In 2011, we saw AMD launching Llano, based on existing Stars (the same architecture found in previous generation Phenom II/Athlon II CPU) cores but built in a modern 32nm process this chips are capable of delivering some never seen before IGP performance, thanks to the mainstream (Evergreen/HD5xxx based) Radeon HD parts. While Llano could not match the raw processing power of Intel's latest Sandybridge CPUs, its thermal and power efficiency combined with strong graphics performance made it a good product for entry-level/mid-range desktops and laptops. Unfortunately for AMD, Llano suffered serious yield issues at Glo-Fo and was unavailable most of time, specially the quad-core A8 parts. So even if AMD had a good product in its hands, it could not supply the market and this was pretty much the picture in 2011. However AMD also made it clear that APUs are an important part of their future silicon investment and it has long term plans with APUs and Fusion System Architecture (FSA). This so called "Fusion" is AMD's effort to combine the processing power of both the CPU and GPU more effectively, distributing the workload between them. In essence this is AMD's GPGPU push. As we said it earlier, this is not a surprising move from AMD considering its graphics expertise, but combined with AMD's relatively weak showings in pure x86 performance (Bulldozer being the big disappointment) in recent times, this sparked some well-spread rumors and speculations throughout the web. While some interpreted this to be clear sign of AMD leaving the x86 arena and perhaps embracing another ISA (Instruction Set Architecture) like ARM - which has been the dominating force in the smartphones and tablets, others rejected the idea of AMD deserting x86 citing some sound technical and business reasons. If you are an enthusiast and like to speculate and contemplate on technical tidbits such as these, we encourage you to click on the links provided- both of them are good reads. But for the rest, this is all that we need to fill in the context while we try to analyze what AMD has stored in for the upcoming future. And that will be NEXT...

(you can find the second part of this article here)

1 comment:

iFLAME said...

This is the Part 1 of two part article, the second part will be following soon. As usual, comments and feedback are most welcome :)