BETA
This is a BETA experience. You may opt-out by clicking here

More From Forbes

Edit Story

AMD Woos Developers With Promises Of Efficiency

This article is more than 10 years old.

For clarity, Fusion is the name that AMD has given to its new semiconductor system architecture.  Chips based on this architecture AMD calls Accelerated Processing Units or APUs from the old idea that graphics processors are hardware accelerators for rendering visual data.  Fusion blends (fuses) traditional x86 computing cores, which compute things serially or one after the other, with graphics processors, which compute them in parallel or all at the same time.  Perhaps oversimplified, that explanation should hold for the duration of this article, but if the reader is truly interested in the details, further enlightenment is available here.

The theory behind Fusion, borne out in real-world cases, is that traditional processors (CPUs or Central Processing Units) are good at certain things and graphics processors (GPUs or Graphics Processing Units) are good at others, and that if these two types are harnessed in tandem, they are capable of “balanced” computing, producing better results faster and at lower power.  This balanced approach is also called “heterogeneous computing.”  More on that in a bit.

Fusion is gaining a degree of traction, as AMD was able to draw hundreds, if not thousands, of developers during a week when a lot else was going on.  I asked several developers that I ran into at various informal moments how they were getting on with Fusion and whether they were finding it useful.  Those I spoke with tended to be in areas like medical imaging and videogame development, which already make plenty of use of graphics processing.  They seemed enthusiastic, and I suspect they were a little more so by the end of the conference, as AMD painted a convincing portrait of the benefits of Fusion during the keynote speeches and breakout sessions.

It was interesting to see how AMD related to its partners.  At most events of this type, the host tends to keep the vast majority of speaking slots to itself, all the better to bandy about its good works.  In contrast, AMD shared many of the keynote slots with partners and gave most of the breakout sessions to a variety of other firms.

One of the better keynotes was given by Amr Awadallah, chief technology officer at Cloudera, who first cut his teeth on Big-Data meaning extraction at Yahoo.  He and Jeff Hammerbacher, who was doing the same thing at Facebook, threw their lots in together, becoming early proponents of Hadoop, which everybody likes to say, but almost nobody groks.  Like any great professor, Awadallah, with simple language, gave a complete and panoramic explanation of Hadoop that even a baby could understand.

Two announcements made during the conference gave a sense of the direction AMD and its partners intend to go.

One concerned the formation of the Heterogeneous System Architecture (HSA) Foundation, initial founders of which include AMD, ARM Holdings, Imagination Technologies, MediaTek, and Texas Instruments (TI).  The foundation’s mandate is to standardize hardware specifications for these mixed CPU/GPU platforms and nurture an ecosystem of developers and partners around it.

Of interest here is that AMD — a silicon maker known primarily for creating processors based on the x86 design pioneered by Intel and used in most PCs (even Apple’s) — is matching up in this consortium with a crew from what could be argued is the opposing camp.

As high-mobility devices like smartphones and tablets continue their rapid rise in prominence, representing an increasing proportion of all Internet access devices, many processor sales are going to ARM licensees these days rather than to Intel and AMD.  So, what’s AMD doing up on stage with ARM, TI, and MediaTek, the last two being ARM licensees, and, while we’re at it, with Imagination, which competes with AMD in graphics?

Some of the answer comes with the second announcement, which trumpeted the fact that AMD is licensing ARM’s Cortex-A5 processor design for a security co-processor that will be integrated into future APUs — a sort of trusted chip within a chip.

The rest of it can be inferred from statements that Rory Read, AMD’s new CEO, has been making about how AMD has to break out of its perennial role as an also-ran in a race with Intel and pursue products in the interests of its customers alone.  If one takes the ideas that heterogeneous computing is primarily about efficiency (read: low power usage), and that ARM, with its legacy in mobile phones, is known for low power usage, then a future that includes AMD parts with actual ARM processor cores running them suddenly seems likely.

When asked directly about this possibility, AMD was a bit coy and continued to insist that x86 is alive, healthy, and slated for a magnificent future, but nonetheless left the door open a crack with respect to the potential that ARM cores could ultimately replace x86 cores in some future products.

Meanwhile, Mark Papermaster, AMD’s chief technology officer, opened up a whole horizon of potential products that might make use of heterogeneous computing, including servers (remember the SeaMicro acquisition?), desktops, notebooks, tablets, automotive infotainment systems, game consoles, set-top boxes, smart TVs — and, yes, potentially even smartphones.

Disclosure: Endpoint has a consulting relationship with AMD.

© 2012 Endpoint Technologies Associates, Inc.  All rights reserved.

Twitter: RogerKay