Intel reveals details of new CPU design

IDF — On the heels of Intel’s announcement of a single, common CPU architecture intended to drive its mobile, desktop, and server platforms, the company has divulged additional details of that microarchitecture. This dual-core CPU design will, as we’ve reported, support an array of Intel technologies, including 64-bit EM64T compatibility, virtualization, enhanced security, and active management capabilities. Intel says the new chips will deliver big improvements in performance per watt, especially compared to its Netburst-based offerings.

At 14 stages, the main pipeline will be a little bit longer than current Pentium M processors. The cores will be a wider, more parallel design capable of issuing, executing, and retiring four instructions at once. (Current x86 processors are generally three-issue.) The CPU will, of course, feature out-of-order instruction execution and will also have deeper buffers than current Intel processors. These design changes should give the new architecture significantly more performance per clock, and somewhat consequently, higher performance per watt.

Unlike Intel’s current dual-core CPU designs, which don’t really share resources or communicate with one another except over the front-side bus, this new design looks to be a much more intentionally multicore design. The on-die L2 cache will be shared between the two cores, and Intel says the relative bandwidth per core will be higher than its current chips. L2 cache size is widely scalable to different sizes for different products. The L1 caches will remain separate and tied to a specific core, but the CPU will be able to transfer data directly from one core’s L1 cache to another. Naturally, these CPUs will thus have two cores on a single die.

The first implementation of the architecture will not include Hyper-Threading, but Intel (somewhat cryptically) says to expect additional threads over time. I don’t believe that means HT capability will be built into silicon but not initially made active, because Intel expressly cited transistor budget as a reason for excluding HT.

On the memory front, the new architecture is slated to have the ever-present “improved pre-fetch” of data into cache, and it will also include what Intel calls “memory disambiguation.” That sounds an awful lot like a NUMA arrangement similar to what’s found on AMD’s Opteron, but I don’t believe it is. This feature seems to be related to a speculative load capability instead..

The server version of the new Intel architecture, code-named Woodcrest, will feature two cores. Intel is also talking about Whitefield, which has as much as twice the L2 cache of Woodcrest and four execution cores.

The company has decided against assigning a codename to this new, common processor microarchitecture, curiously enough. As we’ve noted, the first CPUs based on this design will be available in the second half of 2006 and built using Intel’s 65nm fabrication process.

Comments closed
    • Soldier
    • 14 years ago

    So Intel has finally seen the light and will now copy AMD pretty much across the board…shorter pipelines….64 bit OS support for the desktop…lower power requirements than those power plant killing Prescotts…all thats left is the on die memory controller and Intel can make chips in the same fab as AMD….

    • jutta
    • 14 years ago

    r[

      • UberGerbil
      • 14 years ago

      I see tinfoil hats also impair the ability to type coherent sentences…

        • blastdoor
        • 14 years ago

        So do rose colored glasses…

      • d0g_p00p
      • 14 years ago

      haha.lol

    • SGWB
    • 14 years ago

    Previously, Intel flat out DENIED that the next generation Pentium would be based on the Pentium M. I have been having a hard time finding any news articles from earlier this year, but I did find this on Wikipedia.

    §[< http://en.wikipedia.org/wiki/Pentium_M#Merom<]§ Merom Intel expects to launch the Merom core in the third quarter of 2006. Merom will support the EM64T instruction set, and will serve as the basis for a new desktop core, *[

      • Pettytheft
      • 14 years ago

      I think they meant that it will not a Pentium M tweaked more for the desktop which is what everyone was speculating. This is a new core featuring technologies from both processors.

      • UberGerbil
      • 14 years ago

      It is a new design. It’s not a clean-sheet design (Intel already tried that with the Itanium); since it processes the x86 instruction set, and Intel has spent a lot of time, energy, and money figuring out how to do that very fast and very efficiently, it borrows a lot of ideas from both the P4 and the Pentium M. But it’s not a “tweaked” Pentium M (or Pentium 4). Just the shift from 3 issue to 4 issue (and all that that implies) makes it a significantly different microarchitecture.

      Yes, it’s pretty much what most sensible people expected and it’s not the OMG radical CPU-fu that some wild-eyed commentators predicted, so perhaps it is relatively unexciting. But that doesn’t mean it’s not a new design.

    • Ricardo Dawkins
    • 14 years ago

    I forgot to ask: intel…umm new socket for this ?

      • indeego
      • 14 years ago

      Need you even askg{

      • tfp
      • 14 years ago

      Does it matter? They will probably require a new northbridge anyways or the old boards will not support something that is needed. (power ect…)

    • doctorwho
    • 14 years ago

    Hey guys just thought I’d chime in here. Intel’s metioning of “memory disambiguation” is probably a reference to perhaps the work Intel has done with regard to FB-DIMM technology? Just a thought….

    • wof
    • 14 years ago

    If this is as much better than the P4 as the P4 was better than the P3 AMD will have a field day.

    This better be better than that or we’ll have to live with the boring AMD64’s for a long time.

      • astrotech66
      • 14 years ago

      What’s so boring about the A64? I find them a lot more exciting than anything Intel has ever done. Integrated memory controller, 64-bit capability, excellent overall performance … what’s wrong with that?

        • wof
        • 14 years ago

        They work,perform well, are reliable, pretty much stay the same and consistently spank Intel cpu’s = boring 🙂

          • ScythedBlade
          • 14 years ago

          And a P-M can spank a AMD 64 FX clocked @ the same rate without an integrated memory controller …. (which was intended for games) … wow …

            • Namarrgon
            • 14 years ago

            No surprise that the shorter-pipelined P-M can beat a longer-pipelined A64 at the same clockrate – but the longer pipeline allows faster clocking, so A64 still has the performance advantage.

            • leor
            • 14 years ago

            that statement is simply not true

            • ScythedBlade
            • 14 years ago

            Look up stuff, its true. You need more cooling to overclock an AMD64. With a P-M, overclocking doesn’t even require water cooling…

            Post results where an AMD64 @ the same clock kicks intel’s p-m @ the same clock… (I got my results from one of these affliates somewhere)

            • indeego
            • 14 years ago

            Yeah I was wondering where you couldn’t look up this information. Anand and Tom’s also came to the same conclusiong{<.<}g

            • swaaye
            • 14 years ago

            Yeah and A64 has 2x the L1 cache of Pentium M. The caches of these two CPUs aren’t really comparable, btw. They are very different.

            The gist of it is that PM can compete with an Athlon64 at nearly equivalent clocks while putting out a mere 27W max heat. The heatsink in my notebook on its 2.13GHz dothan is roughly half the size of a deck of cards, and this keeps the CPU around 65C when prime95’ing and around 30C when it’s clocked down idling. When it’s clocked down and idle it doesn’t need a fan on at all. The temp will just sit at around 35C or so. Uhh, that’s fairly amazing.

            • BobbinThreadbare
            • 14 years ago

            AMD has made 25 watt TDP turions at 2.2 ghz.

            • Soldier
            • 14 years ago

            Yeah right, and Netburst is the wave of the future. Pentium Mobiles are an also ran compaired to any high end A64. Sure they run some benches “ok” but if you think a 2 Ghz PM will touch a A64 4000, FX-55 or FX57 you better check that crap thats in your pipe…

    • albundy
    • 14 years ago

    I hope amd responds with quad core!

      • tfp
      • 14 years ago

      I hope VIA responds with an Octal corel[

        • indeego
        • 14 years ago

        OMG Trademark infringementg{

          • tfp
          • 14 years ago

          prove itr[

    • swaaye
    • 14 years ago

    OMG it’s the P7 that never was….and it’s dual core.

    • rgreen83
    • 14 years ago

    And herein lies exactly why I feel even staunch intel supporters should back AMD, competition made these processors, and they should be damn good because of it.

      • absinthexl
      • 14 years ago

      That doesn’t make any sense. Support the company with the best product, so the other will be forced to catch up, regardless of who they are.

    • astrotech66
    • 14 years ago

    So all Intel has to do is say what their plans are and people are already declaring them the victor and relegating AMD to the bargain bin? C’mon … there’s a big difference between announcing a product and actually launching it. Wait till it comes out and the benchmarks are in before you pass judgement. I remember when people were saying that Prescott was going to destroy AMD and then Tejas was going to be even better. We all know how that turned out.

      • Joshvar
      • 14 years ago

      Agreed. Paper launches be damned!

      That said, I look forward to 2006…Intel having its act together is much more formidable than Intel being driven by marketing lunatics. I think AMD is in the best position they’ve ever been in to make a CPU war more interesting than ever. Yes, the original Athlon was a beast at the time, but their financial resources and mindshare have only built off of that foundation.

      • Ricardo Dawkins
      • 14 years ago

      yeah..intel sux for now…but if..and just if..these claims are for real..then give me my X2 for 80 dollars…

      • btb
      • 14 years ago

      Intel have a whole year to bring those products to market. Considering they already have samples up and running, it seems reasonable to expect them to be able to deliver(wouldnt surprise me if they actually moved up their timetable).

      Wether it will be better than AMDs 2H06 offerings remains to be seen of course. Either way, the consumers wins.

    • SpotTheCat
    • 14 years ago

    makes me want to hold off my laptop purchase even longer :-/

    • Joshvar
    • 14 years ago

    “…support an array of Intel technologies, including 64-bit EM64T compatibility…”

    I guess inventing a name for it made it an Intel technology?

      • Ricardo Dawkins
      • 14 years ago

      and AMD invented x86…so they can name the tech x86-64 …what a fanboy !!!

        • indeego
        • 14 years ago

        Did AMD ever claim to invent x86g{

          • Ricardo Dawkins
          • 14 years ago

          yeah…when u took work off someone else…and rename it..they claim they invented x86 renaming it to x86-64…its just “x86” with the so-great 64 bit extensions… 😀

            • Joshvar
            • 14 years ago

            They didn’t rename x86 to x86-64, they developed a 64 bit extension to the x86 architecture. That’s similar* to claiming that Intel renamed x86 to SSE, MMX, or various other extensions to the architecture that have been added (mostly by Intel themselves) over the years. Also, they changed the name to AMD64 anyhow, so I don’t see your point.

            *yes, I understand the complexity of the 64bit extension to the ISA was significantly higher than all of the SIMD Intel has thrown at x86 combined.

            • SGWB
            • 14 years ago

            Yes, x86-64 is an extension to x86-32 which was an extension to x86-16. And Intel is trying to rebrand x86-32 to IA-32 (which runs on x86 processors), but I still calls it x86. Intel’s IA-64 is the instruction set that runs on Itanium and Itanium2 and is not at all related to IA-32 (aka. x86-32). Ive never seen mention of IA-16. MMX, 3DNOW, SSE, SSE2, and SSE3 are all more or less stapled onto the x86 instruction set and can not run a processor on their own. For that matter, x86 processors have not natively run x86 code since the Pentium MMX. All modern x86 CPU’s translate x86 code into RISK-like instructions that can more easily be manipulated for parrallel execution.

            Blah-Blah-Blah.

            Who invented it? Who cares? What have AMD and Intel done for me lately? Real innovations are few and far between. Every advance builds on the work of an earlier technology. The real people who invented the technologies move around, retire, get fired. AMD was founded by former Intel engineers who probably worked on the very first microprocessors. AMD and Intel have cross-licesing deals so they can take advantage of each other’s work.

            Who created what is unknowable because these companies take credit for the brialliance of their engineers who do all the work and rarely get the glory.

            • RyanVM
            • 14 years ago

            RISK? errrr…..

            • SGWB
            • 14 years ago

            Oops… that should read RISC. I never liked that board game anyway.

            • Krogoth
            • 14 years ago

            Software library and intrastructure is the only reason the ancient x86 standard still exists today. Intel tried to pull away from the x86 with IA-64 platform(Itanium family). We all know how that turn out. IA-64’s x86 emulator performance was utterly lackluster, althought it has improved a bit with the latest designs.

            You also fall to see how difficult it really is to design a working CPU design. Any of the architechs out therein the market had *[

    • Ragnar Dan
    • 14 years ago

    Someone in these forums or news comments, perhaps *[http://www.intel4004.com/<]§ .

    • Joshvar
    • 14 years ago

    Um…uh…what? All I was saying was Intel renamed x86-64/AMD64/yetanotherhackonanundeadarchitecture, and now it’s “their technology.” Where did I say AMD claimed x86 as their own? Fanboy my ass!

    • Wulvor
    • 14 years ago

    Foot in Mouth

      • Ricardo Dawkins
      • 14 years ago

      nah…just I don’t understand..how AMD fanboys (same boat as Creative haters) curse just on a PR where some intel tech..get crossed with the cross-license tech from AMD….read the headline…

      just one tech is for sure from AMD: read
      ” support an array of Intel technologies, including 64-bit EM64T compatibility, virtualization, enhanced security, and active management capabilities”

      plus I dont understand why EM64T is first…I care more about virtualization…if I can get a system with that in my lifetime.

        • Joshvar
        • 14 years ago

        Right, but the other technologies listed there weren’t licensed and renamed. That was my point. I’m not a fanboy of AMD any more than I’m a fanboy of Intel. I’ve spent more money and time on Intel boxes than AMD boxes, but I don’t respect the PR games that any tech companies play. It just sucks that they’re right to do it because marketing makes the US consumer economy go round, rather than seeking what you need and researching it on your own.

    • Spotpuff
    • 14 years ago

    So no integrated memory controller?

    I heard rumors AMD was going to integrate the PCI-E stuff onto the die; Intel still isn’t going for this type of approach?

      • t-readyroc
      • 14 years ago
      • Joshvar
      • 14 years ago

      I presume that profits from chipsets are delicious, and Intel doesn’t want to give them up. AMD doesn’t make any profits, so cramming more stuff into the CPU core makes a whole lot more sense (aside from the performance perks associated with it).

    • liquidsquid
    • 14 years ago

    I wonder what AMD will have out by them, probably make this Intel offering look lame I would bet, but I doubt AMD will switch to the smaller processes as quickly so performance per watt will not be as good.

    Wow, 65nm, that is tiny.

    -LS

      • rgreen83
      • 14 years ago

      AMD may get a little out of the socket change, but I would guess k10 to be at least a year to two after these come out, just because k8 is a lot newer than netburst by like what 3 years? Would be nice to see AMD get k10 out at 65nm though.

    • indeego
    • 14 years ago

    So Intel is basically telling us wait like 12 months to buy any of their PC’sg{

      • rgreen83
      • 14 years ago

      Thats what i hear also, but I have no problems recommeding dothan or yonah in the meantime.

        • Joshvar
        • 14 years ago

        What about for desktops?

          • tfp
          • 14 years ago

          what about AMD?

            • indeego
            • 14 years ago

            Would love a small form factor OEM AMD systemg{<...<}g

            • tfp
            • 14 years ago

            so would I but my barton is still a live and kicking

    • blastdoor
    • 14 years ago

    Could somebody who knows what they’re talking about say a few words (or provide a few links) regarding the efficiency of trying to excecute four instructions in parallel? Can that much parallelism be extracted from code that was not explicitly written to be parallel? Also, what’s the difference in transistor count between a four-issue processor and a three-issue processor? Is it large enough for there to be a tradeoff between thread level parallelsim instead of instruction level parallelism? That is, could AMD make a three core chip that issues three instructions at a time in the same space that Intel can make a dual core with 4 instructions at a time, or is the tradeoff not that severe?

      • Tairc
      • 14 years ago

      Cores are much more expensive than execution pathways. When you think about an execution pathway, you add some transistors to the scheduler, add some issue queue complexity, another port to the register file, the execution units in question, and the commit/reorder buffer. That’s it.

      When you add a core, you /duplicate/ the fetch, decode, schedule, issue, register file, execution stages (3 of them, in your example), reorder/commit, etc.

      Essentially, even ignoring things like fetch/decode/register file replication, your example already notes the reason wider=simpler. That going from 3wide to 4wide results in some components being replicated, for a 33% increase in some components. Going from a 2core to a 3core system means a 50% increase in all components. Since all > some, and 50% > 33%, more cores is more expensive.

      • Stefan
      • 14 years ago

      I also expect them to have difficulties saturating a fully four-wide architecture – no matter how much they spend on the whole reordering part.
      However, this would leave yummy resources left for implementing HT into this architecture sometime like 12 months (or whatever) later on. (As I understand it, HT is not going to be integrated into the first iteration of this architecture!)

    • Decelerate
    • 14 years ago

    q[

    • Chryx
    • 14 years ago

    I wonder how much general purpose code would actually be able to utilise a 4-issue chip with any degree of usefulness

    guess it falls to the compiler :/

      • Rottis
      • 14 years ago

      If you remember AMD not adding more ALUs than AthlonXP has to A64, only added more decoding and modified existing ALUs a bit.
      They had good reason, optimized generic 32bit x86 code has ipc of only little over 2 in best case. This is limitation of instruction-set architecture.
      AMD64 instruction-set extension introduces more registers (16), which helps some, but probably won’t give cases with more than 3 parallel instructions except very rarely.

      So I don’t think having 4 generic ALUs helps much. Of course, Intels approach could be having slightly specialized ALUs to reduce complexity and get lower power consumption and higher clockrate.
      Or they just count ALUs differently.

    • Shining Arcanine
    • 14 years ago

    Its been two years since the last time Intel had anything worthwhile, and it wasn’t this good. I’ll build my new PC with this next year. 🙂

      • Lord.Blue
      • 14 years ago

      don’t count the chickens yet. Wait for the counter from AMD next year (they will most likely announce the capabilities of the K10 at next years(or maybe this falls) Tech Tour and set to release it around the same time as the Pentium 5).

        • Shining Arcanine
        • 14 years ago

        They never released the K9…

          • dragondestroyer
          • 14 years ago

          <bad joke> Cause it would have been a dog. <bad joke>

            • jobodaho
            • 14 years ago

            nice, it could have been worse.

    • Decelerate
    • 14 years ago

    I hope AMD will introduce something competitive.

    Things will look interesting indeed. Finally time Intel came back to the fight with something better than a shout and pink glasses.

      • Shining Arcanine
      • 14 years ago

      Since Intel’s marketing department isn’t in charge of processor design anymore, AMD might find that it can’t compete with Intel. It might be back to the value segment for AMD if they can’t match this.

        • continuum
        • 14 years ago

        I’m not holding my breath… this is a whiles off, I suspect I’ll do at least one Athlon64 X2 upgrade in the meantime, since my Barton 2500+ is getting very long in the tooth.

        • Decelerate
        • 14 years ago

        Didn’t AMD state that they’d have quad-cores by ’07?

        • theAntipop
        • 14 years ago

        AMD has plans to match Intel step for step. Virtualization, security (read copy protection), power management. The beauty for AMD is that Intel is playing catchup at this point. AMD already has all those architectural advantages such as direct die linking that Intel is having to redesign into their chips. AMD will probably make a die shrink around the time these new chips show up and not long after we’ll start hearing about the K9. It’s more than a bit brash to say AMD is going back to the value segment.

          • Lord.Blue
          • 14 years ago

          Actually we probobly won’t EVER hear about the K9, as that has be deemed an ‘internal’ step that will never be released to the public. The next one in line for public consumption is currently the K10. It will most likely be on socket 1207/M/S (I think). At least this is what I gathered from the AMD tech tour.

      • WaltC
      • 14 years ago

      q[

    • castlevanity
    • 14 years ago

    if is better than pentium M performance per watt + em64t technology then it will be better than the current athlon 64

      • GodsMadClown
      • 14 years ago

      Coming out in the 2H of ’06, it’d darn well better spank current A64s. If by that time AMD can execute the shift to a 65nm process, then the target will have moved (considerably?) by that time.

        • rgreen83
        • 14 years ago

        I doubt AMD will move the target considerably, they didnt get alot out of the 90nm switch as far as more clock speed. A64s should be pretty small by then though, so higher clocks will probably move down the price scale quite a bit by then.

          • Joshvar
          • 14 years ago

          Judging by the overclocking people have done with the 90nm Venices, there’s plenty of headroom that doesn’t need to be tapped yet for AMD to keep its lead. Much like the leader does when it has a technological advantage that translates into an easily-scalable performance lead, AMD hasn’t kept the rapid pace of releases we saw during the original Athon days.

            • Chryx
            • 14 years ago

            I’ve yet to see any venices overclock much above the clock of chips AMD are actually selling (FX57) ?

            2.8Ghz clocks aren’t unheard of, but that’s an FX57, which is the same process

            • LordFoul
            • 14 years ago

            ya, i have a 3000+ venice (1.8 ghz) overclocked to just under 2.8 ghz. the most i could get out of the larger cores was 2.4 ghz, so they got a 400 mhz gane with the smaller process.

            • Soldier
            • 14 years ago

            Hmmm my 130nm core is running at 2.6 with no problems….

  • Pin It on Pinterest

    Share This