Back in January of this year when I talked to Phil Hester, AMD's CTO, I remember him telling me about the company's undivided commitment to the x86 architecture. “We're pretty religious about the x86,” he declared. Today, with the imminent merger of the company with GPU powerhouse ATI Technologies, it appears the AMD execs have become a bit more secular. At some point, they must have realized that x86 cores have their limits, no matter how many you can stuff into a processor.
As I wrote in my commentary a couple of weeks ago, GPUs, besides providing traditional visualization capabilities, can offer an extra dimension to general-purpose computing by providing high performance vector-type data parallelism in a commodity package. Applications as diverse as genomics research, seismic analysis, options pricing, and image and signal processing represent just some of the applications that could take advantage of graphics engines. And with the AMD-ATI merger, the interest in GPUs in high performance computing has taken off.
GPUs' ascent to power was driven largely by game enthusiasts, who continually demanded more realism and increased performance from these devices. At the same time, visualization became ubiquitous across almost all personal electronic devices, creating a broader need for graphics processing. As GPUs were asked to do more and fill more roles, their capabilities became more generalized.
David Orton, currently the CEO of ATI, has expressed the vision that GPUs will become even more capable in the next few years. Orton, who will probably continue to manage the ATI division after the merger, has said that they are currently working to improve data bandwidth performance and add double-precision (64-bit) capabilities to these devices. He believes this can be done without sacrificing the efficiency of traditional graphics processing. By using the same semiconductor technologies that are driving price/performance in CPUs, GPUs will continue to improve. A 500-gigaflop processor is apparently already in the works.
Supercomputing users are envisioning CPU-GPU hybrid systems offering boatloads of vector performance at a reasonable price. Vendors like PANTA Systems are busy introducing such systems. PANTA's new platform, which can combine traditional Opteron modules with NVIDIA GPU modules, is profiled in this week's issue of HPCwire. On the software side, startups like PeakStream and RapidMind are building development platforms for general-purpose computing with GPUs (GPGPU), giving application programmers access to the raw computing power on these devices.
At this year's Supercomputing Conference (SC06) in November, there are a number of presentations that focus on using GPUs for high performance computing (at SC05, they were barely mentioned). GPGPU now vies with the Cell BE, FPGAs and multi-core processors as one of the hottest topics in HPC. We'll be reporting on GPGPU developments during our upcoming LIVEwire coverage of SC06, next month.
AMD seems to be out in front of the GPGPU wave. Whatever the company's plans are for the ATI devices, one can assume it involves making these devices even more commonplace across computing platforms. Whether they migrate from external devices to co-processors to on-chip cores remains to be seen. But however it plays out, AMD's plans to bring the GPU to mainstream computing is certainly a bold move.
And where does that leave Intel? The rumors a fortnight ago of an Intel-NVIDIA merger turned out to be false (at least for now). But the biggest chipmaker in the world must be at least considering the possibility of getting into the GPU game themselves. If these devices are destined to become as important to software as, for example, the FPU, Intel needs to be involved. Having played catch-up with AMD over the past few years, the execs at Intel must realize that they're not infallible.
On the other hand, AMD has plenty of challenges to contend with as well. For one thing, a lot of time and energy is going to be spent swallowing ATI — a whole new company with a different technology, a different culture and a different market. And AMD itself is now playing catch-up with Intel in the realm of process technology. Intel has already shipped 40 million 65nm chips before AMD has shipped a single one. And next month Intel will begin shipping their first quad-core x86 processor, beating AMD by at least six months. Finally, with this year's introduction of the new Intel Core architecture, AMD's processor performance lead has evaporated.
Because of the resurgence of Intel, AMD's financial position has taken somewhat of a beating lately. Over the past year, its stock has tanked; the current price is about half what it was in January. Also, as a result of renewed price competition from Intel, AMD's profit margin has been slipping, down to 51.4 percent this quarter, a drop of 4 points from last year.
In general though, AMD is sound financially and its strong product lineup will continue to vex Intel for the foreseeable future. The ATI merger is an additional annoyance, presenting Intel with an unusual, asymmetric challenge. It will be interesting to see how they respond. The fun has just begun.
As always, comments about HPCwire are welcomed and encouraged. Write to me, Michael Feldman, at [email protected].