A recent article by Tom’s Hardware discussed “Intel’s Top Secret Processor Plans.” I found it to be quite an intriguing article as Intel has been struggling recently. Intel’s market share has slipped and AMD has gained ground in recent months, due to a number of factors, but Intel’s new plans should make them competitive again.
The cause of Intel’s troubles can be traced back to two events in the past two and a half years. The first occurred in September 2003, when the AMD Athlon 64 came onto the scene. The Athlon 64 wasn’t perfect, and chipset support was spotty at first, but it offered performance on par with Northwood P4s in many applications, most notable games. The P4 was no longer dominant over AMD’s offerings (the Athlon XP had become an inferior performer as the P4 reached 2.4 GHz and higher speeds); it had real competition.
The second event occurred in February of 2004, when Intel introduced the new Prescott core to replace the Northwood. Prescott used Intel’s new 90nm process technology, which should have been an improvement from Northwood’s 130nm process, but because of the P4’s clock-speed-driven architecture, named NetBurst, Intel had to use high-speed but high-heat-leakage transistors. The result was that Prescott was an absolute furnace.
Additionally, in order to achieve higher clock speeds, Intel lengthened the length of the NetBurst pipeline from 20 stages to 31 stages (each stage executes certain types of instructions). This meant that, although theoretically Prescott had more headroom from a speed standpoint, Prescotts performed worse in certain benchmarks than their identically-clocked Northwood predecessors, and more importantly, gained no advantage over the steadily-improving Athlon 64.
AMD already had a loyal following, but they now had a competitive advantage over Intel in the performance-per-watt department. Many people decided they didn’t want a space heater and switched to AMD. Furthermore, Intel started having stability problems – the P4 560 (LGA775, 3.6 GHz) ran so hot that it throttled down so it wouldn’t burn up, thus impacting performance. The P4 570, clocked at 3.8 GHz, was even worse from a heat standpoint.
Intel had originally intended to hit 4 GHz by the end of 2004, but facing these problems, they canceled their 4.0 GHz and higher processors. The reason was ostensibly so that they could focus on dual-cores, but effectively the announcement was a public admission of failure. The NetBurst architecture that had replaced the P6 architecture of the Pentium Pro, PII, and PIII was no longer a viable option. This became even more evident when their dual-core lineup came out; they couldn’t clock their dual-core chips any higher than 3.2 GHz because they produced too much heat even at that speed.
On December 4, 2005, Tom’s Hardware published their article on Intel’s roadmaps over the next two years. Their future looks very promising. Intel had already announced that they would be producing desktop ships based on the Pentium-M architecture that underlies their Centrino laptop platform. First, though, they’ll be coming out with Yonah, a new dual-core version of the Pentium-M that shares the L2 cache between the two cores. This means the cores are tied much more closely together than they are on the Pentium D NetBurst-based dual cores, which essentially communicate over the front-side bus. The result of this tighter integration is better performance. Yonah is slated to come out in January 2006 and will move away from the Pentium name; instead, it will be called Core Solo for the budget single-core versions and Core Duo for the dual-core ones.
Next, in the second half of 2006, comes Yonah’s successor in the mobile space, Merom. Merom will be based on Intel’s new 65nm process technology, and will feature more L2 cache because it’s easier to fit more on the die now that everything is crammed in closer together. More noticeably, though, there will be a desktop version of Merom, codenamed Conroe. Futhermore, there will be a server-class version of Conroe named Woodcrest. According to Tom’s Hardware, “The introduction of the Merom design will be a turning point in Intel’s product policy, because it will be the backbone for all processor families that go into the desktop, the mobile or the enterprise space.” In other words, the NetBurst P4 architecture will no longer be used for desktop and server CPUs. Prescott is the end of the line.
The roadmap continues on without any sign of NetBurst in 2007. Kentsfield will be the first Intel quad-core CPU, also based on Merom/Conroe. By 2008, Intel will be coming out with 45nm chips based on the same architecture. For Intel’s complete roadmaps, both 65nm and 45nm, see the following two pages, go this page and this page.
I’m really glad that Pentium-M is replacing NetBurst. It has good performance per watt, and is similar to the Athlon 64 in its performance-to-speed ratio and in terms of performance strengths and weaknesses. Intel chipsets are better from a stability standpoint than the VIA and nForce on AMD anyway, so Intel will once again become my platform of choice for new computers. Bring it on, Intel!
