Penryn. Nehalem. Phenom. Fusion. Inside these four cryptic code names lies the future of computer desktop processing for 2008. Ultimately, however, it’s all about the epic, age-old battle between chip giant Intel Corp. and underdog Advanced Micro Devices Inc. for desktop dominance.
The harsh reality for AMD is that over the last two years, Intel has absolutely dominated in terms of performance. But performance is only half of the price-performance ratio, and AMD’s willingness to slash prices and aggressively pursue the low- and midrange tiers of the desktop computing market cannot be overrated, despite the company’s technological lag.
Read on for in-depth details about both Intel’s and AMD’s desktop processor plans for 2008, along with a quick glance at both companies’ mobile CPU strategies. Along the way, you’ll find the information you need to make the best purchasing decisions for your company or home.
Intel: Tick-tock
At this point in time, it appears that Intel is far ahead of its primary competition in the CPU performance race. Critics and consumers alike have unanimously recognized the chip maker’s Core 2 microarchitecture as vastly superior to AMD’s processors. These circumstances smell like bad news for AMD fans — particularly at the high end.
Intel’s current CPU road map is a continuation of the company’s “tick-tock” strategy. According to this approach, each year the company alternates its emphasis between shrinking its CPU fabrication process and implementing a new microarchitecture.
In odd years, Intel goes with die shrinkage. Hence the shrinkage of the Core 2 processor line to an efficient and speedy 45nm process in 2007.
In even-numbered years, Intel implements the “tock” in its tick-tock strategy and releases an entirely new CPU microarchitecture. Hence the planned release of Nehalem, detailed below, later this year.
But before we jump into the game-changing aspects of Nehalem, let’s take a closer look at the chip maker’s Core 2 CPU plans for the first half of 2008.
Penryn Core 2 processors: 45nm and more
Penryn, the 45nm version of the Core 2 architecture, was the darling of CPU enthusiasts in 2007 because it delivered on both Intel’s and consumers’ price-performance expectations. According to numerous independent tests from the likes of PC World and the usual assortment of hardcore enthusiast sites, Penryn Core 2 processors put forth about a 20 per cent increase in performance on average.
Additionally, the smaller size of these CPUs means that Intel has been able to mass produce more processors per wafer of silicon, resulting in lower prices and higher profit margins for Intel. Other new and improved features in Penryn chips include improved Virtualization Technology, multimedia-enhancing SSE4 instructions and improved power consumption.
For the majority of 2008, Intel will continue to exploit the substantial performance advantages Penryn processors currently enjoy over AMD processors.
The first consumer octo-coresAt the extreme high end of the performance spectrum, Intel has several new CPUs in store for the first half of 2008. Both the QX9770 and QX9775 processors are Extreme Edition processors clocked at 3.2 GHz, offering a slight clock-speed increase over the enthusiast-caliber 3.0-GHz QX9650 released at the end of 2007.
The kicker for these two CPUs, which will be released in coming weeks, is that both represent a leap in front-side bus (FSB) speeds from 1333 MHz to 1600 MHz. The QX9770 will operate on Intel’s pending X48 chip set, which supports the 1600 FSB bump in speed; it is also compatible with the slower X38 set.
The QX9775 is built for Intel’s new server-inspired LGA771-based Skulltrail platform, which will allow hardcore users to plug two of these quad-core processors into a single motherboard, offering the first octo-core application for consumers.
In addition to running at 3.2 GHz, both processors will feature astonishingly large 12MB L2 caches. At price ranges near $1,500, these CPUs will be far from cheap.
It is likely that Intel will roll out another iteration of Extreme Edition processors at 3.4 GHz and possibly up to 4.0 GHz before the year’s end.
Quad-cores (and dual-cores) for the rest of us
Intel will also be rolling out nonextreme Penryn Q- and E-class processors throughout the course of 2008. (Q stands for quad-core, while the E series is the standard Core 2 Duo line.) In weeks, we’ll begin to see the first of these made available to consumers.
At the midrange, Intel will release numerous quad- and dual-core CPUs under the desktop Core 2 and server-based Xeon labels. It’s worth noting that these quad-core processors are not “native” quad-core — instead they consist of two dual-core dies joined at the hip.
In the first quarter of 2008 alone, we’ll see the release of the Core 2 Quad Q9300 (2.5-GHz clock speed with a 6MB L2 cache), the Q9450 (2.7-GHz, 12MB L2 cache) and the Q9550 (2.8-GHz, 12MB L2 cache). All three processors will run on a 1333-MHz front-side bus and Socket LGA775. Near the third quarter of 2008, we’ll see the Q9400 (2.7 GHz, 6MB L2) and the Q9650 (3.0 GHz, 12MB L2). Both of these processors will run on the LGA775 socket and a 1333 FSB. (If you’re keeping track at home, the “50” designator on Q-series CPUs indicates a 12MB L2 cache.)
A sure sign that market dynamics in the CPU biz have greatly changed in the past few years is the fact that 45nm dual-core 2.6- to 3.3-GHz Core 2 Duo processors are now considered the bottom rung of mid-range processors.
In the first half of 2008, Intel will release a series of 45nm Core 2 Duos, including the E8190 (2.7 GHz), E8200 (2.7 GHz), E8300 (2.8 GHz), E8400 (3.0 GHz), and E8500 (3.2 GHz). Sometime in the third quarter, the E8600 (3.3 GHz) will debut. All of these CPUs are Socket LGA775 and will feature 6MB L2 caches, run on a 1333 MHz front-side bus and support Intel’s Virtualization Technology.
The lowest end of the Core 2 Duo spectrum will feature a new 65nm Core 2 Duo CPU: the 2.6-GHz E4700, which sports a 2MB L2 cache and runs on an 800-MHz FSB. Also on tap is the Penryn-based E7200, a 2.5-GHz chip with a 3MB L2 cache and 1066-MHz FSB speeds. Neither of these processors will be capable of virtualization.
Nehalem, Intel’s code name for its next big leap in CPU technology, is named after a small town near the northwest corner of Oregon. The name originally refers to a Northwestern tribe of Native Americans known more commonly as the Tillamook.
In keeping with the company’s tradition of premiering new CPU microarchitectures in even years, Nehalem will represent a fairly significant enhancement over current Core 2 processor technology. In fact, the buzz around this new processor class has indicated that it will represent the biggest set of changes since Intel released the Pentium Pro in 1995.
We’ve outlined the most important new features below:
Integrated memory controller: This is perhaps the biggest news and the biggest philosophical/structural change in Nehalem. Intel has confirmed that this processor will mark the demise of the Northbridge memory controller. By integrating the memory controller — the logic chip used to handle the input and output of data moving to and from memory — onto the CPU core, Intel will circumvent the throughput limitations imposed by the front-side bus.
The result will be data transfer rates up to 32Gbit/sec. This is an important consideration, given Intel’s ambitious multicore plans for this new chip. Intel’s name for its new memory controller is the Intel QuickPath Interconnect.
The decision to move to an integrated memory controller has stirred some controversy amongst AMD fans, who have derisively commented that Intel finally understands what AMD did many years ago. Beyond reducing memory access speeds and latencies, the integrated memory controller should also offer reduced power consumption.
DDR3 memory: Another big architectural change is the shift to faster DD3 SDRAM.
Native quad-, octo-core and beyond: Nehalem is being constructed from the ground up to permit native (meaning all cores on a single die) quad- and octo-core functionality. Given this, it’s easy to assume that we’ll see dual-die 16-core processors in 2009.
Integrated graphics: In September 2007, Intel announced that Nehalem would also have the capacity for integrated graphics, meaning that 3-D graphics processors would be located on the same chip as the CPU. The important distinction between Intel’s integrated graphics and AMD’s approach is that in Nehalem, the graphics processor will not be integrated at the die level. Instead, it will be on the same piece of silicon, but on a separate die.
Other enhancements in the Nehalem microarchitecture include an improved version of Hyper-Threading and dramatically improved simultaneous multithreading, which will allow multiple processor cores to pool available cycles and memory to better handle CPU-intensive applications. Improved shared caching at the L2 and L3 levels will further help these processors perform more effectively.
It’s important to note that Nehalem does not represent a reduction in die size. This shift will come in 2009 when Intel shrinks the Nehalem architecture to 32nm in the processor family currently code-named Westmere. Assuming Intel stays on its tick-tock routine, in 2010 we’ll see another brand new processor microarchitecture, currently code-named Gesher.
In terms of specific CPUs, the first wave of Nehalem desktop chips (currently code-named Bloomfield) will be released in late 2008. Internet speculation has indicated that the quad-core variants of these Nehalem-based Bloomfield processors will likely feature 8MB of shared L3 cache and sport three separate DDR3 memory channels. It’s likely that we’ll also see octo-core variants of Bloomfield in the fourth quarter of 2008 as well.
On the server side, Intel has two chips in development code-named Beckton and Gainestown. Internet rumors have indicated that Beckton will be a native octo-core CPU (with 16 cores available in a dual-socket configuration), while Gainestown will be more similar to Bloomfield.
Intel’s mobile CPU highlightsAround March or April, Intel will release the mobile version of its Penryn processor under the Core 2 brand. Various mobile CPUs will be released throughout 2008, and it’s highly likely that consumers will see quad-core mobile processors from Intel in the second half of the year, possibly as early as late spring. It is not clear what clock speed these processors will run at, but a version that Intel showed in October 2007 sported two shared 6MB L2 caches on the die. In the same time frame, Intel will launch an all-new Centrino mobile platform code-named Montevina that includes a new CPU, a new chip set and a new wireless adapter. Montevina’s new chip set is code-named Cantiga and will mark the debut of a 1066-MHz front-side bus at the mobile level. Shiloh, the code name for the new wireless adapter, has been rumored to support WiMax (also known as 802.16), a long-range protocol that theoretically permits the transmission of wireless data signals over ranges of up to 30 miles. (Note, however, that WiMax has never been put into practice in any consistent way.) Montevina is also rumored to support the new DisplayPort standard, which specifies a new methodology for connecting a PC to displays and external audio sources. In the first half of 2008, Intel will release a new ultramobile processor code-named Silverthorne. This 45nm CPU is an integral component of Intel’s future CPU plans, particularly in the expanding universe of smart phones and other Internet-ready handheld devices. The key to Silverthorne is its extremely low-power consumption. Intel has reported that these processors’ power needs are about 15 times lower than the chip maker’s lowest-power dual-core processor.
AMD makes its move: Phenom and then some
For AMD, 2007 was not exactly the company’s finest year. Missed ship dates and inferior processor performance blemished the spunky chip maker’s fairly favorable track record of providing solid performance at affordable prices. Industry analysts were quick to point the finger at AMD’s acquisition of graphics manufacturer ATI Technologies Inc., claiming that the company had bitten off more than it could chew.
For its part, AMD has been surprisingly candid about its missteps, and the company is hopeful that 2008 will tell a different story.
At the end of 2007, AMD bid its Athlon series of CPUs adieu and finally released its next-generation Phenom processors, a product line that included the intriguing presence of a triple-core line of processors. One of the biggest advantages AMD has been able to claim over Intel is that Phenom processors are the first native quad-core CPUs on the market — meaning all cores are integrated onto a single die. A shared L3 cache and improved power management were also touted as key new features in this new microarchitecture.
Unfortunately, two setbacks marred the highly anticipated release of Phenom. First, right before the processor’s debut, a bug was discovered in the L3 cache that could cause system lock-ups in rare instances. The software patch that AMD subsequently released fixed this problem, but system performance took a slight hit. (Buyers beware: Avoid systems bearing Phenom 9500 and 9600 processors. AMD will release new CPUs with this bug fixed under the 9550 and 9650 model names.)
The second problem was that processor performance of the initial wave of high-end Phenoms trailed that of Intel’s already-entrenched Core 2 Duo line of CPUs in head-to-head testing. There’s an important caveat here: Benchmarking tests between Phenom and Core 2 were waged at a fairly high range of CPU ranks. We’ll know more about relative performance at the low- and midrange in coming weeks.
Interestingly, the current state of the market is not unlike the conditions AMD faced 10 years ago, when Intel was a consensus favorite at the performance and compatibility levels. In 2008, system buyers can expect AMD to attempt to make up ground in the price-performance ground war by aggressively pricing Phenom and old-gen Athlon processors. The end result should be highly affordable basic-use PCs.
For 2008, AMD
Over the course of 2008, AMD will release a number of Phenom processors at all price levels. At the performance level, we’ll see the release of three processors in the quad-core Phenom FX line: the 2.6-GHz FX-82, the 2.4-GHz FX-90 and the 2.6-GHz FX-91. All three processors will have 2MB of L2 cache and 2MB of shared L3 cache. The FX-82 will be compatible with Socket AM2+, while the two FX-9x will utilize AMD’s server-oriented Socket F+, which features the speedier memory controller found in HyperTransport 3.0. It’s a fair bet that we’ll see more FX processors later in the year — we expect to see this series of processors top at 3.4 GHz by year’s end.
At the high- and midrange CPU calibers, we’ll see numerous Phenom chips. In the first quarter of 2008, AMD is scheduled to release the fixed versions of the flawed Phenom 9500 and 9600 processors released in 2007. As mentioned, they will be named the Phenom 9550 and 9650 and will run at clock speeds of 2.2 GHz and 2.3 GHz, respectively.
The second quarter of 2008 will bring the speedier quad-core Phenom 9700 (2.4 GHz) and 9900 (2.6 GHz) as well as a slightly slower 1.8-GHz 9150 model. All of the aforementioned 9000 series CPUs will feature 2MB of L2 cache and 2MB of L3 cache. Later in the year, we expect to see speed jumps up to 3.4 GHz in this series.
Triple-core — really?
A unique and intriguing addition to AMD’s CPU lineup for 2008 is its 8000 series of triple-core processors. The asymmetry of triple-core processing has apparently thrown some Internet pundits off their game: AMD has been subjected to a fair amount of criticism for selling what are essentially quad-core CPUs with one core disabled or unable to function. The reality is that Intel, AMD and all chip manufacturers sell CPUs with dysfunctional or underperforming cores.
These triple-core processors are intriguing because theoretically they offer a price-performance range between dual-core and quad-core CPUs. In January, AMD will release the 2.1-GHz 8400 and 2.3-GHz 8600 chips. In the second quarter of 2008, expect to see faster variants, including the 2.4-GHz Phenom 8700, the 2.3-GHz 8650 and the 2.1-GHz 8450. All of these processors will feature 1.5MB of L2 cache and 2MB of shared L3 cache. Expect clock speeds of these triple-core processors to reach 3.0 GHz by year’s end.
It’s important to note that, although AMD is currently creating large waves of publicity about the Phenom quad- and triple-core CPUs, the chip manufacturer will not be abandoning its dual-core lineup at the low- and midrange tiers of the market. In fact, it is quite likely that sometime in the second quarter of 2008, AMD will release two dual-core CPUs, model numbers 6050 and 6250. Clock speeds have yet to be confirmed, but each should have 1MB of L2 cache and a 2MB shared L3 cache. It’s likely that AMD will release several more dual-core models in Q3 and Q4.
From Peruses to Fusion
One of AMD’s big pushes for the first half of 2008 will be a new desktop platform initiative named Peruses. The desktop equivalent of Intel’s Centrino notebook platform, Peruses is a combination of a Phenom processor, an ATI graphics adapter and a new AMD chip set.
Unfortunately, it does not appear likely that AMD will achieve a 45nm fabrication process for its CPUs until the very end of 2008 and possibly not until early 2009. This forces the company to cede economic and power consumption efficiencies to Intel for this current generation of processors.
Right now, it appears that AMD is placing a considerable amount of effort and hope in an integrated line of processors known by the code name Fusion. Scheduled for release in 2009, these CPUs will combine CPU and GPU cores onto the same piece of silicon and the same die. This collocation of the central processing and graphics processing chips differs from Intel’s architecture in that the CPU and GPU cores are literally integrated at the die level.
AMD hopes this approach will provide increased performance because of faster access to shared memory and other resources. Beyond improved memory allocations and thermal/power efficiencies, which will probably benefit portable computers more than desktop PCs, it is not yet clear what advantages this new design holds.
AMD’s mobile CPU highlightsIn the first half of 2008, AMD will focus on improving the power-performance ratio of its new mobile CPU. Code-named Griffin, this next-gen processor will represent the debut of an interesting new power management scheme. Griffin chips will feature a split power plane, which grants each core on a dual- or quad-core processor its own power source and should allow for some substantially increased energy efficiencies, as well as improved automated control and management over the various cores on Puma-platform CPUs. Griffin CPUs will be paired with a brand new AMD mobile platform code-named Puma, which will mark the debut of a hybrid form of integrated graphics named Power Xpress. Also scheduled for release in the first half of 2008, this new platform will feature both a motherboard-level integrated graphics processor (GPU) and a separate, more powerful 3-D GPU. Theoretically, this new scheme will allow for increased power savings and fairly high-powered graphics capabilities. Puma will also mandate support for the finally approved 802.11n wireless networking standard, as well as Microsoft’s DirectX 10 graphics technology. As noted above, AMD is strongly looking ahead to 2009 when it plans to launch its new Fusion processors, which integrate the CPU and GPU cores. The first dual-core Fusion CPUs for notebooks should begin rolling out in the second half of 2009, followed by quad-core CPUs for notebooks at an undisclosed time. Regardless of the company’s success this year, AMD’s novel approach to chip design will certainly intrigue CPU enthusiasts. At the very least it gives the company some much-needed differentiation from Intel as it attempts to recapture the price-performance magic that allowed it to make huge inroads earlier this decade.
George Jones is senior vice president of creative services and editorial director for IDG Entertainment.