Vol 20, Issue 13

March 27, 2006

Is Intel Back (and Should AMD be Worried)?

By Kevin Krewell

There must a sense of relief at Intel these days. After the Pentium 4/NetBurst microarchitecture took Intel into microprocessor hell with a mix of power, heat, and performance problems, compared with AMD, the new Core microarchitecture looks to be the company's savior. Which once again proves it's good to have a Plan B.

Back after Intel's Israeli Design Center's Timna processor turned into a bust, it would be hard to imagine that this same processor could spawn a processor-design direction that would eventually save Intel's processor business from becoming a second-class processor supplier. Whoever within Intel decided to let/have the Intel Israeli Design Center work on evolving Timna into a new mobile processor (Banias) now looks like a genius. Banias then led to Yonah and to the Intel Next Generation Microarchitecture, now officially called the Intel Core Microarchitecture. NetBurst turned into Net-Bust as the "hyper" deep pipeline developed power problems and never overcame its low relative instruction per cycle (IPC) performance (compared with AMD and with the P6 microarchitecture). Maybe in a few years, NetBurst will be considered just a bad dream, like the infamous season of the Dallas TV show.

In another article in the March issue of Microprocessor Report, I look at the changes Intel made to the Banias microarchitecture to build the Core microarchitecture, and the changes are dramatic. Compared with Intel's dual-core NetBurst Pentium D processor, the forthcoming Conroe desktop processor (based on Core) is more than 40% faster and will ship with a thermal design point (TDP) about half that of the Pentium D.

In some ways, the core microarchitecture is a return to the Pentium III design and a rejection of the Pentium 4 design, with no trace cache, no double-pumped ALUs, and a merely superpipelined core. But Core has four decoders (up from three for NetBurst and Pentium III) and some seriously fast execution resources compared with Pentium III. Core's out-of-order execution resources can support more instructions in flight and greater peak execution.

Yet, All Is Not Rosy at the New Intel

While Intel looks to get back into the performance game for its bread-and-butter market—volume server, desktop, and notebook processors—its efforts in some other areas could use some more work. In particular, Intel doesn't look like it's getting much more traction in the cellphone baseband-processor business beyond Blackberry and some smaller regional Asia suppliers. Here's a market in which Intel met its match: it tried to jump into a market with entrenched competitors and made limited progress.

Another weakness lies in the company's continual efforts to push the big Itanium rock up a hill. Even with the announcements of a $10 billion program with Itanium partners to promote Itanium and help port more applications, there is little enthusiasm for Itanium's future by many in the industry and by us at Microprocessor Report. In retrospect, given the tremendous improvements in x64 processors (formerly referred to as IA-32 processors with EM64T extensions); the continued efforts of IBM with its Power server processors; and the rise of thread-level parallelism for performance scaling, the case for continued development of the EPIC microarchitecture (Itanium) looks weak.

Intel does show a processor roadmap taking Itanium to many-core designs, and we expect the Itanium processor to be Intel's first processor with an integrated memory controller since Timna. It is clear that Intel continues to fund new Itanium developments, but once the Itanium processors and the Xeon processors share the same platform, the unique advantages of EPIC microarchitecture will continue to be whittled down to vectorized number crunching. Even for that application, a multicore Xeon processor with on-chip vector processing (something like an x64 version of the Cell processor) could prove a suitable replacement.

One other more recent development that I believe is a stumble is Intel's Viiv program. The problem with Viiv is that Intel is not doing a good enough job of clearly articulating what exactly Viiv is. To the best of my understanding, the Viiv platform is a combination of Intel processors, chip sets, and software that makes home PCs work better with other networked devices for the purpose of sharing media around the home. What Intel is reluctant to say is that Viiv is a platform and software solution to fix the ease-of-use problems of Microsoft's Windows XP. At Spring IDF, Don MacDonald, vice president and general manager of Intel's Digital Home Group, attempted to demonstrate that Viiv can eventually be even simpler than anything from the gurus of ease-of-use—Apple Computer. MacDonald's demo attempted to use voice recognition to control a Viiv PC, and it went as well as most voice-recognition demos do: that is to say, it was glitchy. While he may have studied one of the masters of executive demos in action—Steve Jobs—MacDonald still has quite a way to go before he reaches that level. Apple is likely part of the inspiration for Viiv—to make Microsoft Window's Media Center Edition as easy to connect to other PCs and connected devices as Apple's Bonjour network device discovery and connection technology is—and as easy to navigate as Apple's Front Row interface. In that context, Viiv can be a good thing for consumers, even if, like Centrino, it locks the OEMs and customers into Intel's chips because it makes the consumers' interaction with technology more intuitive and appliancelike.

Meanwhile, in Austin, Texas (the Real Home of AMD)...

It looks remarkably like AMD was caught flat-footed by the improvements Intel made in the Core microarchitecture. AMD was probably expecting performance parity from Intel's new microarchitecture, with a slight Intel advantage on (lower) power. But Intel's Core microarchitecture looks a lot more capable, with a wider-issue core, faster SSE hardware, wider internal buses, more prefetchers, 4MB of L2 cache, and a deeper pipeline. It now looks as if AMD will go from being king of the hill in desktop and volume server processors to being just competitive—at best.

AMD will be introducing a new processor socket design to support DDR2 memory, and that should give AMD a few percentage points of improvement in performance, but Intel was showing Conroe systems at IDF with 20% or better performance over AMD's fastest processor on CPU-challenging computer games. Even with DDR2 memory, AMD will very likely lose its position as top dog in the gaming and enthusiast markets that it enjoys today. AMD appears to have focused its near-term engineering efforts on lowering the power of its Turion mobile processor in order to be more competitive with Intel's Centrino/Core Duo processors. AMD's Opteron processor will have a much tougher time competing with the Woodcrest processor compared with the clear advantage it had over the NetBurst Xeon processors. AMD will still have an integration advantage over Intel with its large external north bridge (memory-controller hub) chip. AMD is also lagging behind Intel by about a year in the transition to 65nm. Intel's faster transition to 65nm allows it to add performance-enhancing features like additional L2 cache memory while still maintaining a reasonable die size.

AMD will have to step up its microprocessor designs with larger caches and either push for higher clock speeds or redesign its core with wider, faster execution. After two years on top, AMD was getting comfortable with its performance and performance/watt leadership role. Intel is about to launch a major challenge to regain the lead. AMD must now move to respond. We hope to see more about AMD's roadmap over the next two months, including a keynote address by Chuck Moore at our Spring Processor Forum on May 16.