Anand and I got particularly antsy this month to get an Intel roadmap up - this month's changes look nothing short of spectacular. When we look at some of our other Intel roadmaps in retrospect, there are very few new developments outside of the Smithfield and Yonah announcements. This month looks completely different however; new processor announcements and details in Q1'06, chipset information and - for the first time in a very long time - most of Intel's processor roadmap has moved up, ahead of schedule.

It takes time to realign a huge corporation such as Intel, and we can guess that the recent roadmaps have been the proverbial "calm before the storm". Intel doesn't normally make a lot of noise about major changes in the public as that can lead to reduced sales of current products. However, with AMD making some inroads against Intel and the lackluster retail reception of current 915/925 chipsets, that may not be as much of concern right now. Another possibility is that Intel was working feverishly on some new products and they are now confident enough of their release dates to add them to their roadmaps.

The recent corporate shuffles in Intel must have made the company more aware of their consumer position or more lean to deal with it. Either way Intel is still the 800 pound gorilla; we don't need to look much past their last quarters earnings in relation to AMD's to verify that. If you thought Intel was aggressive before their regrouping last year this year ought to be impressive - to say the least.

Chipsets

First let's take a look at the chipset side of things. It was no surprise that the first generation Socket 775, DDR2, PCIe chipsets Alderwood and Grantsdale faced delays, production problems and poor saturation. Unfortunately such is the life of a first generation chipset. The second generation usually does better, and it looks like Lakeport and Glenwood should be no exception. Actually we no longer need to refer to the next generation DDR2 chipsets by their code names as Intel has cheerfully dubbed the two core logics as 945P and 955X respectively. Even though the launch is yet another month away i945 and i955 news will flood headlines in the upcoming weeks without question.

Before we go under NDA for the launch, here are a few tidbits about 945 that we already know:

• First platforms for dual core support (915, 925 won't support dual core)
• Both platforms support 1066MHz FSB
• 945G will have Intel GMA 950 graphics
• Both platforms support 667MHz DDR2
• 955X will support 8GB of ECC DDR2

For much further details you will probably have to wait for the launch next month.

We also have the upcoming launch of the 915PL and 915GL chipsets, but there's nothing exciting there. 915PL is the new budget 915P, and it drops HD Audio and DDR2 support, as well as limiting the chipset to 1 DIMM per channel with a maximum of 2GB of RAM. The 915GL is similar and falls roughly between the 915GV and 910GL in terms of features. DDR2 support is dropped, but both 533 and 800 FSB support remains. Performance enthusiasts will want to stay away from any of the GL/GV platforms, as usual.

The latest iteration of the roadmap also paid a peculiar amount of attention on Vanderpool Technology or VT. Intel simply refers to this first generation of VT as "the first step in Intel's long term Virtualization roadmap." VT is supposed to take virtual machine applications and allow them to run simultaneously on the same hardware with the same processor - if we are to believe Intel's IDF keynote. Rather than setup two different machines for Linux and Windows, VT aims to unify them both in the same computer. However, the catch seems to be that the processor, chipset, BIOS and software all have to be aware of this process and it isn't a transparent, free upgrade.

Vanderpool won't show up right away however. Intel claims the technology will start showing up in Itanium configurations by the second half of this year, with the mass production server launch date by Q1'06. This almost implies that we will not see any steps forward with this technology until the next processor launch for Xeon, but that's another story in itself. Desktop processors, starting with the Prescott 2M, will get the feature sometime in 2H'05.

Single Core Desktop

Now for desktop processing; we have good news and better news. The good news is almost all desktop Prescotts (including the Celerons) will get a 64-bit makeover real soon. Unfortunately, you'll still need to buy a new processor but the roadmaps indicate there will be virtually no price premium on the 64-bit versions. With Windows XP 64-bit release less than a few months away, it makes sense that Intel's 64-bit push comes strong and hard in the 9^th inning. We are particularly interested in how fully committed the roadmap details EM64T; even the puny Celerons get the instructions. You may recall that the Socket 754 Sempron processors are nearly identical to AMD's Athlon 64 processors with half the cache and the 64-bit instructions removed. It will be interesting to see how the two companies play this against each other as AMD will be the 32-bit SKU on the desktop when WinXP Pro x64 launches.

Just to detail the whole outline for 64-bit Socket 775 processors, here is a quick roadmap of what we have to look forward to:

As you can see, the EM64T enabled CPUs have incremented their model numbers by 1 relative to their non-EM64T counterparts. We're glad that Intel is making a clear distinction between the two variants, rather than simply adding a new suffix. Where there is no earlier part, like the 3.33GHz Celeron D, the model numbers do not have the +1.

The 3.73GHz Pentium 4 EE will show up real soon, if that's your thing, and it will become the second processor to support 1066FSB. We haven't been real impressed with the 1066FSB launch thus far, and a 300MHz bump in clock speed doesn't strike us as something that will revolutionize the performance desktop anytime soon either. However, keep in mind this new P4EE is very different from the previous 3.46GHz revision, and with a different core we may see a very different performance curve on the 1066MHz front side bus. The rest of the Intel roadmap neglects to mention any other 1066FSB processors, including the dual core behemoths, so the technology is either a little bit ahead of its time or simply a temporary dead end.

Next we have the great news. Not only will we see the launch of four Prescott 2M/Iriwindale processors next month, but soon after we will also get our first taste of Smithfield - several quarters ahead of what the previous roadmap had anticipated! Prescott 2M will launch with four SKUs listed below, along with a "670" model clocked at 3.8GHz sometime shortly after.

Like the other Prescott processors, Prescott 2M will launch with EM64T and XD, but it adds Enhanced Speed Step (EIST) as well. EIST is very similar to AMD's Cool n' Quiet as it dynamically ramps the clock speed of the processor to conserve thermals and power. However, the big difference between CnQ and EIST is maturity - EIST has existed in some form or another since the earliest days of the P6 architecture. How EIST will affect performance on everyday desktop processing - particularly on a processor with such a high clock speed - we leave for the actual launch date sometime next month.

Dual Core on the Horizon

So we lied originally - we have even better news. Dual core Smithfield processors, which are really nothing more than two Prescotts slapped together with independent caches, are scheduled to launch a little earlier than we originally claimed in previous roadmap articles. In fact, part of the push to launch so early seems to be to coincide with the 945/955 launch as those chipsets are the only ones to support the multiple core processors. Recall AMD's dual core launch strategy is to enable existing hardware (nForce4, K8T890, 8xxx) to run multiple cores. So while you can't plug a Smithfield into your existing 925X motherboard, it may be for the better. DDR2 has plenty of bandwidth to offer, but as we have seen in server benchmarks, multiple Pentium 4's competing on the memory bus can be quite slow. Dual core Pentium 4's might be horribly inefficient without DDR2-667, however that is another theory we can put to the test on launch day. If you look carefully, you'll see the Smithfields launching only at 800FSB. We find it slightly unusual that the entire 945/955 platform supports a front side bus speed that two $1000 SKUs utilize.

Also note that the dual core processors on the desktop do not support HyperThreading. The server implementation of Smithfield, "Dempsey," has HyperThreading enabled. For database applications, this makes sense - although we have known for a long time that single threaded applications take a performance hit when a HyperThreading processor exclusively runs that program. Interestingly enough the Smithfield lineup has some very competitive price points according to the launch data. The 820, 830 and 840 models will launch at $241, $316 and $530 respectively - compare that to the Pentium 4 lineup today [RTPE: Pentium 4 775]. At today's prices that's only an $80 premium on the second core.

Mobility

Although we get real annoyed with paper launches and PR vaporware, the Centrino lineup never struck us as a set of hardware that didn't show up when promised (at least in the OEM sector). In fact, Sonoma based notebooks with Dothan 2.1GHz processors are already shipping from at least two of the Tier 1 notebook manufacturers although the launch was just last week. This gets us particularly excited since dual core, 65nm Yonah processors are scheduled to launch with the third generation Centrino platform, Napa. And given how reliable Centrino's track record has been, it could be very possible for us to review a 65nm, dual core notebook 365 days from now.

To quote Bender from Futurama, "... the X makes it sound cool." When Yonah hits the retail market it seems fairly likely that Intel will drop the X in favor of a number - although it seems like all the good ones thus far are used up. Well, X is the Roman numeral for 10...

One of the first things we knew about Yonah when we first saw the preliminary work in Taipei for Computex 2004 was that it uses a shared L2 cache between the two cores. While the Smithfield dual core processors separate their cache per core, Yonah is unique as arbitrating the cache for both processors. We also know a bit about the further enhancements on EIST utilized in the processor, including a technology that dynamically throttles power from unused portions of the cache. Feel free to check out or IDF 2004 coverage for that announcement. Another interesting tidbit on the roadmap reveals that the Napa platform will utilize both the 945 and the 955 chipsets in production models. Why a notebook would need ECC support is beyond me, but perhaps there are some features in 955X that haven't been fully leaked yet.

Unfortunately thermals are a huge piece of the pie, and we don't have any confirmed behind the scenes details on that yet. It looks like you'll just have to wait until launch dates for Prescott 2M, Dempsey and Yonah for that scoop.

Though Intel's roadmap reveals a gung-ho attitude for EM64T, the mobility platform looks completely devoid of any such notion. Even the much famed Yonah doesn't explicitly mention any 64-bit support, which may give AMD the upper hand in that match up. AMD's Turion platform will have 64-bit support - will their competition?

Unannounced Secret Stuff

The very forward looking stuff on Intel's roadmap, Q1'06, shows more promise than we had also originally anticipated. First off - get used to the names Presler and Cedar Mill. We had mentioned Cedar Mill before as a single core Pentium 4 evolution. While probably not a direct NetBurst revision, don't be surprised if some of those wonderful projects scrapped with Tejas show up in Cedar Mill instead. Cedar Mill utilizes 2MB of L2 cache, Socket 775 architecture and a 65nm process. On the enterprise portion of the roadmaps Intel is very careful to separate Cedar Mill from the rest of the Prescott 2M SKUs so perhaps there is more to meets the eye for this little processor.

Presler is a whole different animal. On the roadmaps Intel marks Presler as the eventual dual core replacement for Smithfield albeit with an extra megabyte of cache per core. Since this is the first we have heard of the processor in official circles, details were pretty light.

We talked real casually in the past about Dempsey - the dual core Xeon. From the roadmaps Dempsey doesn't look similar to any Pentium 4 or Xeon processor we know about. For starters, expect 1066FSB, dual core, and HyperThreading. If four logical processors per socket didn't seem to catch your attention the addition of Fully Buffered DIMM (FB-DIMM) and iAMT will also show up on the chip platform. Intel also refers quite often to its "Diamond Peak" technology - which they loosely describe as:

Platform level LAN acceleration based on improvements in processor, MCH and ESB-2

Like the desktop platforms, the next generation server core logic will also feature Vanderpool Technology. This leads to a real lot of promise for those who rely on User Mode Linux or VMWare for their enterprise solutions. Rather than placing separate operating systems in different machines, VT opens the door to putting different operating systems on the same processor. The roadmap stresses this sort of configuration makes sense for high availability; if one OS crashes, its OK because we have 2 running.

"Blackford," the E7520 replacement for Xeon, will utilize ESB-2 and will become Dempsey's chariot in the server market. A cheaper, stripped down version of Blackford dubbed Bensley will perform the task of Dempsey's value platform, Greencreek is the workstation variant.

Final Thoughts

Generally we do not get to write this much about the Intel world, so today's opportunity to look at everything Intel has to offer is quite refreshing. While a lot of people have looked at Intel's recent reorganization as a sign of weakness, we feel the company is just trimming itself back to the main arteries it does best. We are at least optimistic that 2005 will be a better year for consumers than the last one - now we just need to see if AMD has enough up their sleeves to make things interesting.