Beyond Dual Core: 2007 Desktop CPU Road Map

Article Index

by George Jones @ ComputerWorld

January 01, 2007 (Computerworld) -- What a difference a year makes. One year ago, we were dazed, dazzled and beguiled by the arrival of dual-core processors. Offerings from Intel Corp. and Advanced Micro Devices Inc. had analysts, journalists, IT professionals and enthusiasts all gushing with praise for a bright new multitasking future.

Amazingly, both Intel and AMD were able to deliver on the potential of dual-core processing. Throughout 2006, desktop PCs played host to a series of processors that, while slower at the clock-speed level, were faster in real-life usage, allowing for unprecedented amounts of multitasking. (For more about both companies' current lineups of desktop CPUs, see our CPU Buyer's Guide.)

As the calendar flips to 2007, we are firmly entrenched in the world of multicore processors. And based upon the confidential road maps of both Intel and AMD, it is clear that dual-core CPUs are only the launching point for the future of the microprocessor. In 2007, quad cores and even eight-core CPUs will be available. By 2009, there's a good chance that sixteen-core processors will be on the market.

As we enter 2007, five key questions regarding the pending year's CPU battle are on our minds:

  1. Will AMD be able to continue its dominance in the U.S. desktop market?
  2. How will Intel capitalize upon the success of Core 2?
  3. Will AMD be able to match the success of Intel's Core 2 processors?
  4. When will the market see true quad-core and even eight-core processors?
  5. What surprises do the chip makers have up their sleeves?

With all this in mind, we're taking an extended look at the processors and processor trends you can expect to see in 2007. Not surprisingly, neither AMD nor Intel was willing to divulge many specifics regarding their CPU releases for the coming year. So we scoured the Net, pored over statements from both companies and dug into reports from the host of analysts and experts who cover them.

It's worth noting that much of the information in this road map is preliminary and code-name-level information. As such, the specifics of the processors could change in coming months.

All secrets are revealed within.


Intel advances


Extensive digging has revealed a good portion of Intel's plan for increasing desktop market share in the coming year. Not surprisingly, the bulk of the company's processor road map revolves around the Core microprocessor architecture, formerly code-named "Merom." One of the smashing success stories of 2006, Core 2 processors offer unparalleled levels of performance per watt of energy consumed and may allow Intel to recapture market share lost to AMD over the past three years. (Core 2 processors are based on the Core architecture; so-called Core processors were based on the company's previous Pentium 4/M architecture.)

In attempt to round out its desktop CPU portfolio in the first half of 2007, Intel will focus on several new processor families based on the Core 2 architecture at all performance levels, including a new value line that uses Core 2 at the Celeron level. Here are the details.

Early 2007 brings new Core 2 processors

At the high-end performance level, Intel will release three new quad-core CPUs at the beginning of the year, dubbed the Core 2 Quad Q6600, Q6400 and Q6300. These three processors will be dual-core, dual-die processors, meaning that they will essentially be two Core 2 processors joined together.

Scheduled for release in the first week of January, the Q6600 will have a clock speed of 2.4 GHz, the Q6400 will have a clock speed of 2.13 GHz, and the Q6300 will operate at 1.86 GHz. Each processor will operate on a 1,066-MHz front-side bus and have 8MB of total Level 2 cache, with 4MB of shared cache on each die. (A large L2 cache allows for faster retrieval of frequently accessed data, thereby speeding up overall system performance.)

In the first half of 2007, Intel will also release a new series of Core 2 Duo processors aimed at the midrange market. These dual-core, single-die processors will reside in the newly introduced Core 2 Duo E4000 series, and the initial release will consist of three CPUs: the 2-GHz E4400, the 1.8-GHz E4300 and the 1.6-GHz E4200.

This category of CPUs will operate on an 800-MHz front-side bus and will likely come with a 2MB shared L2 cache. The E4300 will be the first processor in this family released and could be in desktop PCs as soon as February. It is widely expected that E4000 processors will come with virtualization and 64-bit support.

Finally, in an attempt to make significant inroads in the value CPU sector -- one that has traditionally been dominated by AMD -- Intel is trickling its Core 2 CPU line down to the low-cost market. Intel has not yet made it clear whether these processors will be single-core versions of the Core 2 Duo or dual-core chips with one core disabled.

In the second quarter, Intel plans to release a number of processors in this value category. Around this same time, the chipmaker will probably phase out the Pentium 600 series, specifically the Pentium 4 651, 641 and 631.

To avoid confusing CPU buyers, Intel will use the Pentium and Celeron brand names for these new CPUs, even though they are based on the Core architecture.

In the Pentium bracket, we'll see releases of the E1060, E1040 and E1020. The E1060 will have a clock speed of 1.8 GHz, the E1040 will run at 1.6 GHz, and the E1040 will run at 1.4 GHz. Each will have 1MB of L2 cache with a front-side bus speed of 800 MHz. While these processors will support Intel's 64-bit extensions, none of the E1000 line will support virtualization or hyperthreading, a technology that allows single-core CPUs to behave as if they were dual-core ones.

In the Celeron bracket, CPU buyers will likely see a wide range of clock speeds. At press time, no specific model numbers or clock speeds were available, but it appears that the name of this series of processors will be the Celeron 400 series and that these processors will have 512KB of L2 cache. It is not clear whether or not these processors will support 64-bit extensions, virtualization or HyperThreading.

Intel advances (continued)

'Bearlake' chip set boosts front-side bus speeds

As Intel shifts to multicore processing, the bus speed becomes a more pressing concern because of the increased volume of data traffic generated by separate CPU cores. The front-side bus (FSB) is the primary channel of data communication between the CPU and other devices on the system, such as RAM and hard drives. It's essentially a single-lane highway with limited bandwidth (see Wikipedia's excellent explainer for details). As CPU manufacturers stack more processing cores onto a single processor, the risk that this data channel will become full increases, hence the need for faster FSB speeds.

Thus, one of the most significant releases Intel will make in 2007 is a brand-new chip set foundation code-named "Bearlake." This chip set is the successor to the 975X chip set and will feature a number of upgrades and improvements. The P35 Express will be released first in the second quarter of 2007 and will feature two key upgrades: an all-new 1,333-MHz FSB and support for DDR2-800 and DDR3-1066 memory.

Intel recently announced official names for the first wave of Bearlake chip sets. The G35 and G33 monikers will be attached to mainstream consumer desktop chip sets. The G35 chip set will feature an integrated DirectX 10-compatible graphics processor.

The P35 Express and X38 Express will be Intel's performance-oriented, high-end versions of Bearlake. The X38 will feature the same 1,333-MHz FSB and DDR2-800/DDR3-1066 memory support found in the P35 Express, and it will also feature two PCI-Express x16 slots and PCI Express 2.0, which is twice as fast as PCI-Express 1.0 (5 GHz, compared with 2.5 GHz).

1,333-MHz front-side bus CPUs by midyear

At the same time it releases the Bearlake chip set described above, Intel will also release three speedy new Core 2 processors that are compatible with the Bearlake chip set's 1,333-MHz FSB and other new features. The model numbers of these processors are the E6850, E6750 and E6650. (The "50" designator in the model number indicates a FSB speed of 1,333-MHz).

The clock speeds of the E6850, E6750 and E6650 will be ratcheted up to 3 GHz, 2.66 GHz, and 2.33 GHz, respectively. For reference's sake, 3 GHz is the current high mark for Intel's Core 2 CPUs and can currently be found in only one Core 2 processor -- the Extreme X6800. Each of these new Core 2 CPUs is a dual-core, single-die processor that utilizes 4MB of shared L2 cache.

At the same time it releases the E6850, E6750 and E6650, Intel will also release a non-Bearlake CPU -- the E6800. The E6800 will have a clock speed of 3 GHz and a 4MB shared cache, but will run at a bus speed of only 1,066 MHz.


Intel advances (continued)

The road to 45nm begins here

Initially, based upon statements Intel released in the middle of 2006, it appeared that Intel would be fairly conservative in terms of CPU releases in 2007. Goals for 2007, as stated by the chipmaker, were to release several new processors based upon the Core 2 architecture. But the company's technology and design priorities would be on improving its fabrication process with the aim of producing 45nm-process CPUs by 2008.

However, toward the end of 2006, Intel indicated that it was ahead of schedule for reaching a 45nm fabrication process. In late November, the company stated that it had already produced a prototype of a 45nm processor, and that it was now hoping to release 45nm processors by the second half of 2007.

Why the emphasis on the shift to 45nm? Beyond the simple metrics of cost -- a 45nm process is smaller than today's state-of-the-art 65nm process, which allows for more CPUs to be manufactured per wafer of silicon -- smaller fabrication processes allow for performance boosts via shorter distances for electrons to travel, faster clock speeds, larger cache sizes and reduced energy consumption.

Another important reason behind the frenzied pursuit of 45nm CPUs: Intel wields a significant advantage over AMD when it comes to fabrication process technology. AMD just released its first 65nm processors in December 2006, and it is not expected to move to a 45nm process until mid- to late 2008. Given the theoretical performance-per-watt advantages that 45nm-process CPUs will possess, Intel's rapid transition to 45nm could place considerable pressure on AMD to catch up.

Enter the 'Penryn' processors

Intel's 45nm process will manifest itself in a microprocessor architecture known only by the code name "Penryn." Not surprisingly, Intel has kept a fairly tight lid on Penryn, but based on rumors and speculation by analysts and experts, it appears that these processors will be based on the Core 2 architecture, but will take advantage of the 45nm processor to provide larger L2 caches and increased performance. (It's worth noting that Penryn will also serve as Intel's mobile processor architecture, with laptop CPUs scheduled for release in early 2008.)

In terms of specific processor releases, Computerworld has heard of a few Penryn-based CPUs that should be released in late 2007. Two dual-core, single-die processors known as "Ridgefield" and "Wolfdale," respectively, could be released as early as the third quarter of 2007. There has been no concrete information regarding the clock speeds of these two processors, but reliable early information has indicated that the Ridgefield processor will have 3MB of shared L2 cache, while the Wolfdale variant will have 6MB of shared L2 cache.

One of Intel's most potentially exciting desktop CPUs is code-named "Yorkfield" and appears to be a 45nm-process quad-core processor that uses a single die (referred to as "native" quad-core) and has an astonishing 12MB of shared L2 cache. When combined with the performance-per-watt advantages of the 45nm processor, this could be Intel's extreme high-end CPU of the year if it is released on schedule at the end of Q3 or the beginning of Q4 2007.

Finally, although the company has not confirmed this in any way, it's entirely plausible that Intel could combine two Yorkfield processors at the end of 2007 to create an octo-core, dual-die, 24MB L2 cache monster.

All the Penryn processors described above will be compatible with Intel's new Bearlake chip set.

As a teaser to what may come beyond 2007, rumors have swirled around a future-gen Intel microprocessor architecture code-named "Nehalem" that will be released in 2008. No details on this architecture have been revealed to date.


AMD battles back (and takes the eight-core lead)

At the beginning of last year, AMD was the CPU darling in terms of performance and the all-important price-performance ratio. This run of dominance ended with a thud in the summer of 2006 when Intel released its stunning new Core microprocessor architecture. Based on a highly efficient 65nm fabrication process -- a process AMD just reached at the end of 2006 -- this new architecture produced results that swiftly relegated AMD CPUs to also-ran status. Much to AMD's chagrin, benchmark result after benchmark result declared Core 2 processors the winners.

Interestingly, while AMD was left scrambling to keep up with Intel on the performance and performance-per-watt fronts for both desktop and laptop CPUs, the company experienced one of its best years ever. In January 2006, reports indicated that AMD CPUs were dominating market share on PC desktops in the United States to the tune of 85% to Intel's 15%. Even longtime Intel stalwart Dell Inc. got into the movement, inking a deal to use AMD CPUs in some Dell PCs.

But this success was largely fueled by price-performance advantages that existed prior to the release of the Core 2 Duo line. How will AMD respond in the coming year to what appears to be a clear technology advantage on Intel's part?

Part of the answer to this question appears to reside outside of the realm of CPUs. In July, AMD announced a whopper of an acquisition as it took over venerable graphics and chip set manufacturer ATI Technologies Inc. It's not likely that this acquisition will have a significant impact upon AMD's 2007 CPU forecast above and beyond the growing pains and distractions that a large acquisition can create.

Read on to find out the details about AMD's push to a smaller fabrication process, an all-new Socket AM3 and the alluring potential of eight- and sixteen-core processors.

The push to 65nm

One of the chief advantages Intel wields over AMD is the ability to deploy new technology at a more rapid pace. This was made clear in the early part of 2006, when Intel pushed out a 65nm series of processors many months ahead of AMD.

In December, AMD finally caught up with the release of four new 65nm dual-core processors in its X2 line: the Athlon 64 X2 5000+, 4800+, 4400+ and 4000+. These processors operate at clock speeds of 2.6 GHz, 2.5 GHz, 2.3 GHz and 2.1 GHz respectively. Each has 1MB of shared L2 cache and support for AMD's virtualization and 64-bit technologies.

In the second quarter of 2007, AMD will release two more 65nm processors at the high end of this product series. The X2 5200+ will run at 2.7 GHz, while the 5400+ will operate at 2.8 GHz. Like the rest of the processors in this lineup, the 5400+ and 5200+ will support virtualization and 64-bit technology.

It's highly likely that AMD will release more 65nm processors into this lineup throughout the year. In the second half of 2007 and possibly sooner, buyers and systems integrators will likely see 65nm X2 5600+, 5800+ and 6000+ parts as well as conversions of lower-run CPUs such as the 3800+.

As soon as January or February, AMD will also release a single-core 65nm Athlon processor. Code-named "Lima," these processors will be introduced as the Athlon 64 3800+ and the 3500+. In the second quarter of 2007, AMD will release an Athlon 64 4000+ CPU on this same 65nm process. All of these processors will have 512K of L2 cache.

Also in the second quarter, AMD plans to release four single-core Sempron processors fabricated on the new 65nm process: the 2.2-GHz Sempron 3800+, the 2-GHz 3600+, the 1.8-GHz 3500+ and the 1.8-GHz 3400+. These processors will have 256K of L2 cache, with the exception of the Sempron 3500+, which will have only a 128K cache.


AMD battles back (continued)

The high end: Quad FX rumbles in

In early December of 2006, AMD released three new performance-oriented processors -- the 3-GHz Athlon 64 FX-74, the 2.8-GHz FX-72 and the 2.6-GHz FX-70 -- under the chipmaker's newly introduced Quad FX line. Based upon a new dual-socket Socket 1207 motherboard and AMD's enterprise-class Opteron CPU architecture, Quad FX processors are purchased in pairs, one per socket. Early performance benchmarks have indicated that these CPUs are indeed suitable for the "megatasking" environments AMD has constructed them for.

In 2007, AMD will continue to build out this Quad FX line with the Q2 release of the Athlon 64 FX-76. The FX-76 will have a clock speed of 3.2 GHz and 1MB of L2 cache per core. This, like the FX-74, FX-72 and FX-70, will be fabricated on the older 90nm process. AMD has indicated that these will be the last of the FX CPUs to be built on a 90nm process.

65nm 'Agena' makes its debut

In early Q3 2007, AMD is planning to release a brand-new performance-oriented 65nm CPU architecture code-named "Agena," and it sounds like a high-performance dream. This new processor line will be the first "native" quad-core processor released by either AMD or Intel. (When used with a multicore processor, the term native refers to a processor with all the individual CPU cores integrated on a single die. To date, all previous quad-core processors have essentially been two dual-core processors attached together.)

One other impressive attribute of the Agena FX processor is that it will operate at a bus speed of 4 GHz, thanks to the 3.0 iteration of AMD's HyperTransport link that will debut at the same time. This doubles the bus speed of previous FX and other Athlon 64 processors. The Agena FX quad core will feature 2MB of shared L2 cache and 2MB of L3 cache. (L3 cache functions in a similar manner to L2 cache, but it's a little slower and is consequentially less expensive.)

Preliminary information has revealed that Agena FX processors will run at clock speeds between 2.7 GHz and 2.9 GHz. It's likely that we'll see two or three different Agena FX processors when they're released, possibly under Quad FX-8x model numbers.

It's too early to say for sure, but the native single-die nature of these CPUs and the shift to 65nm should result in a massive performance boost. One other interesting attribute of the Agena FX processors is that using them with the Quad FX platform, which uses two CPU sockets, will likely allow AMD to be the first chipmaker to release an eight-core platform. By the end of the summer, high-end enthusiasts will be able to run two Agena FX processors at the same time.

CPUs for the masses: Socket AM2+ and Kuma

AMD's rapid embrace of quad-core processing at the high end of CPU performance does not mean that the chipmaker is leaving mainstream dual-core computing out in the cold.

In the middle of 2007, AMD will revise Socket AM2 to increase energy efficiency and bus speed. Currently scheduled for release at the end of Q2 2007, this revision will be named Socket AM2+.

Finally, in Q3 2007, AMD will release a new series of 65nm native dual-core processors aimed squarely at the mainstream consumer market. Currently code named "Kuma," these processors -- which emphasize power consumption and high performance-per-watt yields -- will operate at clock speeds from 2 GHz to 2.9 GHz and will contain 1MB of shared L2 cache and 2MB of shared L3 cache.

These processors will be compatible with the all-new Socket AM2+ and as such will feature bus speeds of 4 GHz. At press time, AMD had yet to reveal model names or numbers for Kuma-based CPUs.

Don't miss the rest of Forecast 2007.

George Jones is the editorial director of IDG Entertainment.