Original Link: http://www.anandtech.com/show/1520
Intel's Dual Core Strategy Investigatedby Anand Lal Shimpi on October 22, 2004 3:09 PM EST
- Posted in
Been hearing conflicting dual core information lately? Here's a compilation of everything we have and know about Intel's dual core plans for the next two years.
Dual Core for Desktops in 2005
Intel has yet to determine what brand they will market their first desktop chips under, although we'd expect them to continue to use the Pentium 4 brand but with some sort of appendage like Extreme Edition or Lots of Cores Version. Intel has, however, already determined what the specifications and the model numbers of their dual core chips will be.
Currently set for release in Q3 2005, Intel has three dual core chips on their desktop roadmap: the x20, x30 and x40. The only difference between these three chips is clock speed, with the x20 running at 2.8GHz, the x30 running at 3GHz and the x40 running at 3.2GHz. All of the chips are LGA-775 compatible and run off of an 800MHz FSB. Hyper-Threading is not enabled with Intel's dual core chips.
As far as architecture goes, the x-series of dual core CPUs from Intel are built on the little talked-about Smithfield core. While many have speculated that Smithfield may be Banias or Dothan based, it's now clear that Smithfield is little more than two 90nm Prescott cores built on the same die. There is a requirement for a very small amount of arbitration logic that will balance bus transactions between the two CPUs, but for the most part Smithfield is basically two Prescotts.
But doesn't Prescott run too hot already? How could Intel possibly build their first dual core chip out of the 90nm beast that is Prescott? The issue with Prescott hitting higher clock speeds ends up being thermal density - too many transistors, generating too much heat, in too small of a space. Intel's automated layout tools do help reduce this burden a bit, but what's important is that the thermal density of Smithfield is no worse than Prescott. If you take two Prescotts and place them side by side, the areas of the die with the greatest thermal density will still be the same, there will simply be twice as many of them. So overall power consumption will obviously be increased by a factor of two and there will be much more heat dissipated, but the thermal density of Smithfield will remain the same as Prescott.
In order to deal with the fact that Smithfield needs to be able to run with conventional cooling, Intel dropped the clock speed of Smithfield down to the 2.8 - 3.2GHz range, from the fastest 3.8GHz Prescott that will be out at the time. The reducing in clock speed will make sure that temperatures and power consumption is more reasonable for Smithfield.
Smithfield will also feature EM64T (Intel's version of AMD's x86-64 extensions), EIST (Enhanced Intel SpeedStep Technology) and Intel's XD bit support. Chipset support for Smithfield will come from Glenwood and Lakeport, both of which support the 1066MHz FSB (as well as 800) and Dual Channel DDR-2 667 and 533. Glenwood (the successor to 925X) will support up to 8GB of memory, making it the perfect candidate for EM64T enabled processors that want to break the 4GB barrier.
Dual Core Mobility
A while ago I asked Pat Gelsinger what was in store for the future of Pentium M with regards to threading, and he responded with multi-core. Thus it's no surprise to finally see Intel giving more details about Yonah (or Jonah depending on what part of the world you're from), the 65nm dual core successor to Dothan.
Yonah's dual core setup will be much more power optimized than what Smithfield will bring to the desktop, and in an effect, much more efficient. There's little information available about Yonah, other than it will most likely have a power and thermal balancing dual core setup, with the individual cores powering down when they're not needed. The idea here is to switch between cores not based on performance needs, but based on thermal and power needs. If one core happens to be running too hot, it can be powered down and the active workload shifted to a different thread running on the remaining core, thus reducing the problem of thermal density by effectively spreading the thermal load across two cores.
While Dothan was more of a small set of fixes and updates to Banias, Yonah is going to be a significant set of improvements to what we've seen in the past. Yonah has already taped out and Intel is slated to release the chip in 2006. Yonah will begin sampling by the end of 2005 and Intel expects it to ramp up to 50% of the performance notebook segment by the end of 2006.
Yonah's platform is codenamed Napa, which brings support for DDR-2 667 as well as the 667MHz FSB to help keep the dual core Yonah fed with data. Given that Dothan will ramp to 2.26GHz by the end of 2005, we can expect Yonah's clock speeds to be around there upon its launch.
Dual Core Servers
Most recently Intel announced that their only multi core enterprise product shipping in 2005 would be Montecito, the dual core version of Itanium 2. This is where the rumor of Intel not shipping any dual core chips until 2006 came from, as it seems that the first dual core Xeons won't ship until Q1 2006. Now as long as Intel's desktop chips don't face any further delays, there will still be dual core desktop CPUs based on Smithfield available by the end of 2005. Given Intel's track record lately, it would not be surprising to see these chips slip into 2006 as well though.
The first dual core Xeons appear to be nothing more than Xeon versions of Smithfield,
but Intel does list that more power efficient versions of the first dual core
Xeons will appear in the second half of 2006 - potentially including
some of the power management features that will be included in Yonah.
An interesting inclusion on Intel's dual core Enterprise roadmaps is the mention of a Dual Independent Bus. The term Dual Independent Bus hasn't been used since the days of the Pentium III to indicate the separation of the internal L2 cache bus from the external Front Side Bus. However, we wonder if the dual core Xeon's Dual Independent Bus may in fact be individual 64-bit datapaths to each socket on a dual socket server. This way a pair of dual core Xeons would have just as much FSB bandwidth per pair of cores as the present day dual processor Xeons, which will end up improving performance tremendously.
The platforms that will enable dual core Xeon support will be Blackford and Greencreek for 2S (two Socket) systems, and the next generation Twincastle for 4S systems.
The Problem with Intel's Approach
The major issue with Intel's approach to dual core designs is that the dual cores must contest with one another for bandwidth across Intel's 64-bit NetBurst FSB. To make matters worse, the x-series line of dual core CPUs are currently only slated for use with an 800MHz FSB, instead of Intel's soon to be announced 1066MHz FSB. The reduction in bandwidth will hurt performance scalability and we continue to wonder why Intel is reluctant to transition more of their CPUs to the 1066MHz FSB, especially the dual core chips that definitely need it.
With only a 64-bit FSB running at 800MHz, a single x40 processor will only have 6.4GB/s of bandwidth to the rest of the system. Now that 6.4GB/s is fine for a single CPU, but an x40 with two cores the bandwidth requirements go up significantly.
While Intel's current roadmap appears to place dual core on the desktop before it makes its way to the enterprise (other than with Itanium), AMD's strategy is reversed - with dual core appearing in workstations and on servers before making a splash on the desktop.
Overall, AMD's approach simply makes more sense, since the overall performance benefit to dual core on the desktop will be minimal at best but strong in very specific applications and usage patterns. With most desktop applications continuing to be single threaded, dual core will still have to wait until there is more application support before truly being useful on the desktop. Heavy multitaskers and those running workstation applications will appreciate the benefits of dual core, but gamers and most other users will find higher clocked single core chips to be better suited for their needs.
The scenario is exactly the opposite in the workstation and server space, with the applications already seeing huge benefits from going to multiple processors thanks to their multithreaded nature.
When AMD mentions that their K8 architecture was designed for multicore operation from the start, they weren't lying. Each Socket-939 or Socket-940 K8 chip, whether it's an Athlon 64, Athlon 64 FX or Opteron, features three Hyper Transport links (whether they are all operational is another question). In order to create a dual core version of a K8 based chip, you simply remove a single pair of Hyper Transport PHYs, one from each chip, and fuse the two Hyper Transport links together - thus creating a direct path of communication between the two cores, capable of transmitting data at up to 8GB/s (at 1GHz) between the two chips. Update: There is some debate as to how AMD implements dual core in their K8 architecture. The above description was provided by AMD from an earlier discussion but many readers have emailed to point out that the two cores are connected at the SRQ level. We are awaiting official confirmation from AMD as to exactly how their dual core technology is implemented. Update 2:While AMD never got back to us with an official response, unofficially they did confirm that the two cores on a single dual core Opteron die do communicate at full speed and are not connected at the HT level. We apologize for the error.
AMD's performance limitation here will be memory bandwidth, with the two K8 cores sharing the 128-bit DDR memory bus. While we currently don't see a huge performance increase from going to a 128-bit memory bus from a single channel 64-bit interface, the move to dual core will definitely make greater use of memory bandwidth.
AMD continues to list the second half of 2005 as the introduction timeframe for their dual core CPUs, with Opteron coming first and then Athlon 64 FX. Once again, as with all release dates, nothing is set in stone, but right now it looks like that both AMD and Intel are planning on having dual core on the desktop in the same general timeframe.
AMD has yet to reveal what the official specifications of their upcoming dual core desktop products are, but based on roadmaps and what we've seen, it would seem that the first dual core desktop parts will be based on two 90nm Athlon 64 FX cores with a shared memory controller. Interally AMD is referring to this CPU as "Toledo" as we've already published.
The best way to evaluate the impact of dual core CPUs on the desktop is to look at the impact by moving to a multiprocessor setup on the desktop. The vast majority of applications on the desktop are still single threaded, thus garnering no real performance benefit from moving to dual core. The areas that we saw improvements in thanks to Hyper Threading will see further performance improvements due to dual core on both AMD and Intel platforms, but in most cases buying a single processor running at a higher clock speed will end up yielding higher overall performance.
For the most part, it would seem that the dual core releases of 2005 are mostly to establish a foundation for future dual core CPU releases that will provide functionality such as power and thermal balancing across multiple cores. Next year Intel will be releasing a number of new processors, including the new 2MB L2 Prescott parts as well as the dual core x-series, but despite all of the new product launches, clock speeds will only increase by 200MHz in the next 14 months. If anything, the release of larger cache and dual core desktop processors is a way to continue to promote the "newer, faster, better" upgrades without necessarily improving performance all that much.
Today the slowest Prescott based Pentium 4s run at 2.8GHz and 3.0GHz - and a full year from now the slowest Prescott based Pentium 4s will run at 3GHz. This is the first time in recent history that the predicted roadmap for CPUs will remain relatively flat. It will take continued maturity in 90nm manufacturing, a smooth transition to 65nm as well as improvements in multi core designs to truly make the migration worth it.
The future of dual core doesn't lie in taking two identical cores and throwing them on the same die. The future and true potential is in the use of multiple cores with different abilities to help improve performance while keeping power consumption and thermal density at a minimum. The idea of putting two cores, one fast and one slow, in a CPU has already been proposed numerous times as a method of keeping power consumption low while continuing to improve performance.
Right now dual core is more of a manufacturing hurdle than anything else. Putting that many logic transistors on a single die without reducing yield is a tough goal. Intel will have a slightly harder time with the migration to dual core since their chips simply put our more heat, but in theory Intel has the superior manufacturing (although it's been very difficult to compare success at 90nm between AMD and Intel thanks to all of the variables Prescott introduced). Needless to say that we'd be very surprised if both companies met the current ship dates for dual core desktop chips simply based on how things have progressed in the past.
That being said, despite the end of 2005 being the time for dual core, the desktop world will be largely unchanged by its introduction. It will take application support more than anything to truly bring about performance improvements, but with an aggressive CPU ramp developers may be more inclined to invest in making their applications multithreaded as more users have dual core systems. The more we look at roadmaps, the more it seems like while 2005 will be the year of anticipation for dual core, 2006 may be when dual core actually gets interesting. Until then, we view dual core on the desktop as a nice way of getting attention away the fact that clock speeds aren't rising. It's a necessary move in order to gain more traction and support for multithreaded desktop applications but its immediate benefit to the end user will be limited. But then again, so has every other major architectural shift.