Intel Server Processors

In the server and workstation segment - or "Enterprise Processors" if you prefer - Intel continues to develop both their Xeon line as well as the IA-64 based Itanium line. We'll start with the more common Xeon roadmap. We have further broken down the Xeon lineup into the uni-processor (UP), dual-processor (DP) and multi-processor (MP) segments.


Pentium 4 Xeon Lineup
Xeon Uni Processor
Processor Speed L2 Cache L3 Cache FSB Launch Date
Xeon UP 4.0F 4.00 GHz 1 MB 800 MHz Cancelled
P4XE 3.73 3.73 GHz 2 MB 1066 MHz * Desktop Only
Xeon UP 3.8F 3.80 GHz 2 MB 800 MHz Q2'05
Xeon UP 3.8F 3.80 GHz 1 MB 800 MHz Q4'04
Xeon UP 3.6F 3.60 GHz 2 MB 800 MHz Q1'05
Xeon UP 3.6F 3.60 GHz 1 MB 800 MHz Already Available
Xeon UP 3.4F 3.40 GHz 2 MB 800 MHz Q1'05
Xeon UP 3.4F 3.40 GHz 1 MB 800 MHz Already Available
Xeon UP 3.2F 3.20 GHz 2 MB 800 MHz Q1'05
Xeon Dual Processor
Processor Speed L2 Cache L3 Cache FSB Launch Date
Xeon DP 3.8F 3.80 GHz 2 MB 800 MHz Cancelled
Xeon DP >= 3.6F >= 3.60 GHz 2 MB 800 MHz Q3'05
Xeon DP 3.8F 3.80 GHz 1 MB 800 MHz Q4'04
Xeon DP 3.6F 3.60 GHz 2 MB 800 MHz Q1'05
Xeon DP 3.6F 3.60 GHz 1 MB 800 MHz Already Available
Xeon DP 3.4F 3.40 GHz 2 MB 800 MHz Q1'05
Xeon DP 3.4F 3.40 GHz 1 MB 800 MHz Already Available
Xeon DP 3.2F 3.20 GHz 2 MB 800 MHz Q1'05
Xeon DP 3.2F 3.20 GHz 1 MB 800 MHz Already Available
Xeon DP 3.0F 3.00 GHz 2 MB 800 MHz Q1'05
Xeon DP 3.0F 3.00 GHz 1 MB 800 MHz Already Available
Xeon DP 2.8F 2.80 GHz 1 MB 800 MHz Already Available
Xeon Multi Processor
Processor Speed L2 Cache L3 Cache FSB Launch Date
Xeon MP 3.66 3.66 GHz 1 MB 667 MHz Q1'05
Xeon MP >= 3.50 >= 3.50 GHz 1 MB 8 MB 667 MHz Q2'05
Xeon MP >= 3.50 >= 3.50 GHz 1 MB 4 MB 667 MHz Q2'05
Xeon MP 3.16 3.16 GHz 1 MB 4 MB 667 MHz Q2'05
Xeon MP 3.16 3.16 GHz 1 MB 667 MHz Q1'05
Xeon MP 3.0 3.00 GHz 512 KB 4 MB 400 MHz Already Available
Xeon MP 2.7 2.70 GHz 512 KB 2 MB 400 MHz Already Available
Xeon MP 2.2 2.20 GHz 512 KB 2 MB 400 MHz Already Available
Xeon MP 2.0 2.00 GHz 512 KB 1 MB 400 MHz Already Available

Changes in the Xeon lineup in many ways mirror what we saw on the desktop front, with Intel being even more conservative. Unlike the desktop parts, all of the Xeon chips based off the Prescott NetBurst core (i.e. Nocona, Cranford, and Potomac) have EM64T support enabled. There may still be some older pre-E-0 stepping 1MB cache cores around, but they are no longer in production.

On the UP parts, we again see the cancellation of the 4.0 GHz Nocona, while in the DP group we see the cancellation of the 3.8 GHz Irwindale. The reason for the removal of the P4XE 3.73 as a workstation part is not quite the same: the 925XE chipset does not support ECC RAM, which Intel requires for their workstation processors. The P4XE with 1066 FSB will thus be relegated to the enthusiast sector until Intel can come out with a newer chipset. A similar occurrence happened with the initial launch of the 925X chipset, as only later revisions have allowed for DDR2 to function with ECC memory.

There are once more several "greater than or equal to" entries in the roadmap, which in this case we assume to be there pending testing and validation of the parts in an enterprise environment. One thing that Intel has always prided itself on was rock solid stability in the server segment, and other than a few minor slips they tend to deliver. Intel would rather cancel parts or lower clock speeds than risk any instability in the server market. The desire for stability over performance is also reflected in the use of the slower 400 and 667 MHz FSB on the Xeon MP parts. The 4M and 8M L3 cache Potomac cores will be available in Q2'05, but final clock speeds remain an unknown at present, with a target of somewhere around 3.5 GHz.


Itanium 2 Lineup
Itanium Dual Processor
Processor Speed L3 Cache FSB Launch Date
Itanium 2 Dual Core TBD 667 FSB Q4'05
Itanium 2 Dual Core TBD 533 FSB Q4'05
Itanium 2 1.6 1.60 GHz 3 MB 533 FSB Q4'04
Itanium 2 1.6 1.60 GHz 3 MB 400 FSB Already Available
Itanium 2 1.3 LV 1.30 GHz 3 MB 400 FSB Nov '04
Itanium Multi Processor
Processor Speed L3 Cache FSB Launch Date
Itanium 2 Dual Core TBD 667 FSB Q4'05
Itanium 2 Dual Core TBD 533 FSB Q4'05
Itanium 2 TBD 9 MB 667 FSB Q3'05
Itanium 2 1.6 1.60 GHz 9 MB 400 FSB Nov '04
Itanium 2 1.6 1.60 GHz 6 MB 400 FSB Nov '04
Itanium 2 1.5 1.50 GHz 4 MB 400 FSB Nov '04

On the Itanium side of the roadmap, we again see the conservative nature of the market. 533 and 667 FSB designs have been pushed back, with the note that they are "dependent on OEM platform capability and validation." The chipsets and processors are available to Intel's partners for testing right now, but it does not look like any will be ready for public consumption just yet. The Madison 9M part that was originally supposed to launch in Q1'05 has been pushed to Q3'05 with clock speed likely to be in the 1.6 GHz range, based on information in past roadmaps. Q4'05 marks the availability of dual core Millington processors, with clock speeds again to be determined. They could very easily slip into 2006, although testing and validation of the parts will begin sometime in 2005.

Most of us may never actually see let alone use an IA-64 platform, but the technologies used in such high-end parts invariably trickle down into the mainstream processors. What sort of technologies are we talking about? One example is the redundant L3 cache design that is used in the Itanium. Due to the large die size of the chips, a flaw in the silicon matrix has a chance of drastically reducing the yields. Since the probability is high that flaws will occur in the large L3 cache rather than in the processor core itself, Intel has designed in some redundancy so that individual L3 banks can be deactivated and the work can be relocated to backup banks. With a die size that is more than four times as large as current Prescott chips, such precautions make sense. Another feature that will likely one day make it's way into desktop processors is the higher dispatch rate of the Itanium cores. Where NetBurst and K8 cores make do with a dispatch rate of up to 3 instructions per clock cycle, current Itanium 2 chips can dispatch up to 8 instructions per core per clock, and future revisions of the architecture could increase this further. It may take a while before we see these features in our home computers, but sooner or later they will come.

That wraps it up for this update of the Intel roadmap. Please remember that all of the dates and information in this roadmap is of a preliminary nature and could easily change. The cancellation of several parts in this update is clear evidence of that fact. Also, this is only a roadmap and not a comparison of processor performance, and not an endorsement per se of any of the parts. We are looking at what Intel is planning relative to what they already have available, and we'll worry about detailed comparisons with AMD when and if the parts actually launch.

Mobile Processors
Comments Locked

22 Comments

View All Comments

  • RotoSequence - Tuesday, November 23, 2004 - link

    I think I have gotten a hold of the true reason behind the cancellation of the 4 Ghz Pentium 4. For at least now, there are some serious issues resulting after extended use of Pentium 4s that are clocked at 4 Ghz+. I E-mailed an Intel Engineer and asked why they dont use FSB increases instead of higher multipliers (as the Pentium 4s tend to like that). When you drive the processor that hard, there is electron polarization that occurs in the processor. Over the course of a few years, this permanent electron polarization (As a result of the electron density in such a small area) results in an open circuit-and an un-bootable CPU. Evidently, this is the average net effect, and in the Intel Guy's own words, "We would rather sell an underclocked processor than one that will fail in a couple of years." So potentially, there you have it people; the reason behind Intel's (possibly temporary?) reason for the cancellation of the Pentium 4 4Ghz.
  • JarredWalton - Saturday, November 6, 2004 - link

    #20: Personally, I won't be surprised to see the dual-core "x" chips slip into Q4 or possibly even '06. Actually, Intel will probably ship limited samples in Q3'05, which may allow them to claim that they have "shipped". Yonah/Jonah is not yet on the roadmaps, which is what this article is about. See Kristopher's recent look at code names, chipsets, and features for an overview of other stuff coming down the pipe from Intel.
  • knitecrow - Thursday, November 4, 2004 - link

    Does anyone seriously expect intel to keep the q3 2005 timeline for x** chips?

    Do they even have working engineering samples?

    ---
    why isn't there anything said about dual-cored pentium M chips?

    I guess we won't seem then untill 2006
  • JarredWalton - Thursday, November 4, 2004 - link

    Oops.... my bad. I thought that the 5xxJ actually included EDB as well as EIST, but on closer inspection this is not the case. Urg... So I removed it from the 5xx chips. A new inspection of the roadmap reveals that EIST is apparently enabled on all "Prescott" core Xeons (really Nocona, Cranford and Potomac), and any 775-LGA chips with EM64T also have it. So for now that means the 5xx series is without speedstep. The x.xxF revision Xeons are the same as the 5xxJ revision Prescotts, but EIST and EM64T are not enables for the desktop parts (yet?).
  • jarthel - Wednesday, November 3, 2004 - link

    A member of an Australian computer forums mentioned that the Intel channel roadmap does not mentioned EIST in the J CPUs.

    Also this publicly available page (http://www.intel.com/products/processor_number/inf... does not discuss EIST with J CPUs.

    Can you clarify?
  • JarredWalton - Saturday, October 30, 2004 - link

    For official 1066 FSB support, you will need the 925XE chipset, which is not yet available. EM64T support should still work in older chipsets - a BIOS update might be required, however.

    As for EIST and unlocked lower multipliers, I read something somewhere (yeah, I know - highly specific information) about some mod that could be done to unlock lower multipliers on new P4 chips. I don't even recall if it was a software mod or a hardware mod, or maybe just something that needed to be enabled in the BIOS. If I find out more information, I'll let you know, but for now that's about as specific as I can get. :|
  • danidentity - Saturday, October 30, 2004 - link

    Thanks Jarrad...about my question in post 8, the 5xxJ CPUs include EIST. Does that mean they'll have unlocked downward multipliers like Athlon 64's?
  • Foxbat121 - Friday, October 29, 2004 - link

    Current Intel chipset (915 and 925) does not support 1066 FSB, IIRC.
  • MIDIman - Friday, October 29, 2004 - link

    Is it safe to say that everything here will be compatible with current LGA775-based chipsets and motherboards, including 1066mhz FSB?

    Does 925x and 915p support EM64T and dual-core CPUs?

    Just curious if its too early to buy into LGA775.
  • JarredWalton - Friday, October 29, 2004 - link

    Pumpkin (happy halloween!): The roadmap I have has no mention of any new socket 478 parts other than a Celeron D 345. There have been rumors on and off that Intel may introduce a faster socket 478 part, but to my knowledge Intel has said nothing official about this.

    Danidentity (no relation to Bourne?): "Q1" is as specific as the roadmap gets in relation to the 6xx series right now. Typically, Intel will narrow that down to an actual month (and day even) a couple months prior to launch. We'll update the roadmaps with this information when we get it. I'll take the middle ground right now and guess at February - that way I can't be more than 30 days off. :)

Log in

Don't have an account? Sign up now