Intel Merom CPU already works!

Posted:
in Future Apple Hardware edited January 2014
Translation from French site x86-secret (07/27/05):

Quote:

Champagne! Merom boots

Reliability level: 6/8

This week, many Intel engineers will congratulate themselves in front of big champagne magnums and caviar jars. Indeed, after being taped-out last week-end, the Merom A0 die has just been tested and it works! The announced planning of mass availability for mid-2006 should thus be respected. Remember that Merom, and Conroe the variation of the same core, respectively targeted for Mobile and Desktop, will be the two future CPUs replacing the actual Pentium 4 Netburst architecture. On paper, Merom is impressive. At first based on the Pentium M core, Merom now does not share many things with it, cause of very deep modifications. Alas, before that, we'll have to see poor stuff such as Cedar Mill and Presler, Pentium 4 65nm series, which will probably offer just good overclocking capabilities...



Merom is the dual core processor for mobile succeding Yonah. It will begin half-2006 on a 65 nm process at 2.5 Ghz, TDP 45 W, with 4 MB shared L2. It will also be fully 64-bit (Intel x86 IA32e EM64T), whereas Yonah will be only 32-bit.

Merom is part of the "Santa Rosa" platform (Centrino 4, 802.11a/b/g/n), succeeding the Centrino 3 "Napa" used with Yonah.



Steve said the first Intel-based Mac will be there June 2006. Sounds like "mid-2006" for me. At this time Yonah would have been there for months (target: early 2006. Yonah is even working nowadays: it is already being tested in some laptops). So in June it won't be such a cuting-edge innovation anymore.

Would Intel reserve Merom for Apple in a big June 2006 launch? It would be a huge step forward even compared to Yonah, but you can argue that June 2006 would be too much early for a mid-2006 target.
«1

Comments

  • Reply 1 of 31
    mi0immi0im Posts: 8member
    Can you believe such a foolish report?

    Quote:

    Originally posted by Cosmos 1999

    after being taped-out last week-end, the Merom A0 die has just been tested



    It takes 3-4 weeks
  • Reply 2 of 31
    rickagrickag Posts: 1,626member
    If true this is very good as I think, but am not sure, the Merom will be the first dual core chip from Intel in which the cores communicate directly with each other. I believe in the Yonah the cores must communicate over the external bus.



    Then again I could be totally wrong.
  • Reply 3 of 31
    hmurchisonhmurchison Posts: 12,341member
    Quote:

    Originally posted by rickag

    If true this is very good as I think, but am not sure, the Merom will be the first dual core chip from Intel in which the cores communicate directly with each other. I believe in the Yonah the cores must communicate over the external bus.



    Then again I could be totally wrong.




    The only real reason for cores to communicate with each other is cache coherency. Thus by Yonah having a shared cache you can easily say the Yonah cores communicate with each other in the most appropriate way.
  • Reply 4 of 31
    hirohiro Posts: 2,663member
    Quote:

    Originally posted by hmurchison

    The only real reason for cores to communicate with each other is cache coherency. Thus by Yonah having a shared cache you can easily say the Yonah cores communicate with each other in the most appropriate way.



    That's so 1980's, we didn't have integrated FPU's then either. It was considered most appropriate to save the transistor budget in the CPU for logic and farm out the calculations when they were necessary.



    Cores communicating directly can have bug wins for synchronization of threads and processes, should the OS support such things. Like order of magnitude wins.
  • Reply 5 of 31
    cubistcubist Posts: 954member
    Quote:

    Originally posted by Cosmos 1999

    ...

    Steve said the first Intel-based Mac will be there June 2006. Sounds like "mid-2006" for me. At this time Yonah would have been there for months (target: early 2006. Yonah is even working nowadays: it is already being tested in some laptops). So in June it won't be such a cuting-edge innovation anymore.





    Using Yonah 6 months after it comes out sounds about right for Apple. And again, Steve said they would be shipping by June 2006, so they could come earlier. Actually, I'd expect the first Mactels to use single-core Pentium Ms.



    Quote:

    Would Intel reserve Merom for Apple in a big June 2006 launch?



    No. Apple will not get special treatment from Intel in any way.



    And as for the off-chip connection between cores on Intel's dual-core chips, Hiro is right. To put it simply, that is a lame design. It's a stinking kludge. Don't even bother making excuses for it. AMD is at least a year ahead of Intel on multi-core processors, and any honest person knows it.
  • Reply 6 of 31
    salmonstksalmonstk Posts: 568member
    Quote:

    Originally posted by cubist

    [B]Using Yonah 6 months after it comes out sounds about right for Apple. And again, Steve said they would be shipping by June 2006, so they could come earlier. Actually, I'd expect the first Mactels to use single-core Pentium Ms.





    No. Apple will not get special treatment from Intel in any way.



    And as for the off-chip connection between cores on Intel's dual-core chips, Hiro is right. To put it simply, that is a lame design. It's a stinking kludge. Don't even bother making excuses for it. AMD is at least a year ahead of Intel on multi-core processors, and any honest person knows it.




    I actually disagree. I think apple will be the first to get new CPU's from Intel. Partially because I bet Steve asked, but I bet Intel agreed not because of the RDF but because of availability.



    Today if intel introduces a new chip they need to have millions ready just for Dell to sell a machine with it. By giving Apple the chips first as production ramps Intel does not have to built up inventory and there is no shortage because Apple sells fewer computers.



    Mind you I am not talking a huge lead on a new chip, but a couple of months may be likely as Intel ramps production.



    An a side but supporting note, Intel gains the Apple publicity machine if they do this. Sure they announce a chip months ahead. But if it "debues" on a Mac they get all the press that goes with a Mac announcement. When was the last time the press covered the intoduction of a new Dell- even with a new chip in it.
  • Reply 7 of 31
    Quote:

    Originally posted by salmonstk

    I actually disagree. I think apple will be the first to get new CPU's from Intel. Partially because I bet Steve asked, but I bet Intel agreed not because of the RDF but because of availability.



    Today if intel introduces a new chip they need to have millions ready just for Dell to sell a machine with it. By giving Apple the chips first as production ramps Intel does not have to built up inventory and there is no shortage because Apple sells fewer computers.




    Ha, that is turning on old problem on its head. I remember when Apple had a hard time competing with the clones. PowerComputing would get the latest and greatest chips from Moto first because they didn't need as many as Apple. So Apple was losing all the cream at the top.
  • Reply 8 of 31
    cwestphacwestpha Posts: 48member
    Ok time for some corrections from a Intel Insider.



    Merom is a 65nm Pentium M deritivitive. It does not appear to be netburst (pentium 4) based. It includes SSE3 extensions but merom (this is NOT the dektop version of the processor) does NOT include 64 bit extensions as of April 2005. However the desktop version DOES have 64 bit extensions.

    The Santa Rosa platform will launch in the second half of 2006 along with a high end dual core desktop version of it (with EMT64) and a clock around 3 Ghz.



    Just so everyone knows Napa is supposed to have a possible A/B/G/WiMax option (coming at or shortly after launch) and Santa Rosa is supposed to have a A/B/G/N/WiMax chip option. Intel has already shown a working A/B/G/WiMax 1.0 chip, but it is unknown if they will make this chip widely available as WiMax 1.0 is a stationary technology. WiMax 1.0 is more likely to show up in the Merom desktop derivitive. However with Santa Rosa it is possible that Intel will be finished with a mobile version of WiMax and as such could be included with N (N of course pending 802.11 board approval). Intel is not in the market to release pre standards wireless chips as they don't want to risk some of the functionality suddenly becoming un-usable... that and pre-N devices vary wildly already.



    P.S. I am currently under some NDAs for Napa. However my NDAs for Santa Rosa and Merom does not go into effect until fall ICC 05'. So ask away about Santa Rosa and Merom.



    Quote:

    Steve said the first Intel-based Mac will be there June 2006. Sounds like "mid-2006" for me. At this time Yonah would have been there for months (target: early 2006. Yonah is even working nowadays: it is already being tested in some laptops). So in June it won't be such a cuting-edge innovation anymore.

    Would Intel reserve Merom for Apple in a big June 2006 launch? It would be a huge step forward even compared to Yonah, but you can argue that June 2006 would be too much early for a mid-2006 target.



    This is baseless speculation without merit. I am amazed how many Apple people don't actually LISTEN to what Jobs says, they just hear for the summery and don't mine for facts in what he says.



    Jobs said clearly that he would have products shipping and in customer hands BY that time next year. How everyone assumes that it will be weeks or days before that next developers conference is beyond me. (was IBM really that bad with product delivery?)

    Beyond that Apple has all but confirmed the use of Yonah and given many signs it wont use the Pentium 4 in a desktop solution (outside of the development kits). As such it is likely that the Yonah will be used for laptop and perhaps the mini. However Jobs DID say that he was going to keep the PowerPC platform around for a while in the high end systems and then gradually transition over the course of two years or so.



    This makes sense since a laptop replacement will arrive MONTHS before the next developers conference with the early available Yonah chip (launch in December or January with immediate product availability) and Napa platform (which the CPU and chipset has already taped out). Santa Rosa gives Apple a new platform for which to totally replace the PowerPC in its laptops and it comes with the added bonus of a desktop platform based off of it that. So now they can replace Power in all laptop lines, the mini, and the iMac. My guess is that they want to keep the G5 in servers and high end power macs for a bit longer for about another 6 months beyond this at least. Centrino has not had the best record with content creation, though Yonah and Merom will mostly take care of that.



    Quote:

    It takes 3-4 weeks



    Ironicly the Pentium M has always had amazingly short periods from the first test chip to final silicon. In part this is thanks to the advanced computerized testing done in the Israeli labs (home of the 8088, MMX, and Centrino). Centrino is mostly designed using advanced computerized testing and design software. This allows them to optimally lay out components to cut down on power usage, heat, die size, and cut down interference between on die components. Once they are done with this then the only major potential problem with making the processors would be the a change in manufacturing process, which is negated since 65nm is mostly a simple die shrink from 90nm. From 130nm to 90nm electrons and components start behaving slightly differently. This is true of every couple of manufacturing transitions.



    However it is vary believable that Intel made a almost perfect die, used their ion gun to fix a few little problems, re-test, find everything worked correctly, make the proper changes in in the computer controlled fab, then spit out a perfect new processor in about a week after a day of testing or so.

    When you are getting ready to tap out a new die you usually have a few extra wafers around just ready to quickly be ?thrown in? to the machine and spat out a couple of days latter.



    Quote:

    If true this is very good as I think, but am not sure, the Merom will be the first dual core chip from Intel in which the cores communicate directly with each other. I believe in the Yonah the cores must communicate over the external bus.



    Kinda, Yonah is like two Sonoma cores glued together with a little bit more tweaking and some new power management features. It wont be as bad as the dual core P4, but its not nearly as ideal as the AMD solution (then again the AMD solution is power hungry and generates more heat then Yonah's solution). For example it looks like Yonah MAY include limited cache to cache transfers. If not Merom will support this feature in some form (it may even share one big cache).



    Quote:

    Using Yonah 6 months after it comes out sounds about right for Apple. And again, Steve said they would be shipping by June 2006, so they could come earlier. Actually, I'd expect the first Mactels to use single-core Pentium Ms.



    No Jobs said he would have mactels shipping and in the customers hands by June 2006. They already have Yonah in their labs. Apple has taped out Yonah about three months ago.

    If Apple waits till June 06' that would mean it would have taken them almost a YEAR to make a product out of final silicon.

    Intel is vary aggressively ramping Centrino development. Israel labs got brand new equipment about a year ago to help speed up their development cycles dramatically and start work on desktop solutions.



    Quote:

    No. Apple will not get special treatment from Intel in any way.



    Actually Intel is giving Apple more help then any other partner in Intel history. Intel looks like it may also be helping design the new motherboard and may manufacture it for them, they have NEVER made a custom one use motherboard and manufactured for only one customer.



    Quote:

    And as for the off-chip connection between cores on Intel's dual-core chips, Hiro is right. To put it simply, that is a lame design. It's a stinking kludge. Don't even bother making excuses for it. AMD is at least a year ahead of Intel on multi-core processors, and any honest person knows it.



    AMD is both ahead and behind in multi-core design... though the net effect is they are behind. On one hand AMD has integrated memory controllers that allow vary low latency memory calls. Also AMD is ahead on the desktop with hyper transport.

    For Intel they are head when it comes to performance per watt (for laptops) and being able to make low powered chips. Hyper transport is a pig when it comes to power (not as much as PCI-express though), and AMD just cant make a Pentium-m like chip.

    Intel could catch up and surpass if they integrated a memory controller on die and then fix the last of their content creation problems. Of course they would also need a new interconnect technology if they integrate a memory controller too, but I think they already have an idea for that.



    Quote:

    An a side but supporting note, Intel gains the Apple publicity machine if they do this. Sure they announce a chip months ahead. But if it "debues" on a Mac they get all the press that goes with a Mac announcement. When was the last time the press covered the intoduction of a new Dell- even with a new chip in it.



    Well Intel would risk going out of business if they did that. The Channel members (which I am apart of) is Intel's growth and the hammer that keeps beating down AMD. Simply put the Channel is how Intel is keeping growing and kills off all of it's competition.

    Intel gives us first dibs on new technologies and gives us the tools to take on the big boys, in return we use more Intel solutions and they keep growing. Heck in a year or two a whitebook (third party notebooks) may be a more upgradeable and serviceable product then even a Dell or HP notebook.

    If Intel started giving us second or third dabs on technology then Intel would lose some IC members to AMD. Granted AMD cant match the ?next day replacement parts shipped overnight, no questions asked? policy that we channel members have. Heck not even Dell gets that nice of treatment and support. If you laptop broke and you bought it from me I could fix your laptop faster then Dell could with in home service. With Dell ya have to schedual out a day or two where as with me you can say X isnt working, I order new parts for free (Intel pays shipping both ways and the cost of the replacement), and I come out next day around noon with the new part and quickly swap it out.

    Now do you understand why people sell more Intel then AMD products? Only Apple comes close with their in store tech support (even then it takes a day or two for hardware repairs usually).

    Intel also gives you all kinds of free selling stuff like window clings, posters, coasters, pens, and you actually get a bonus for selling Intel products that you can use to buy things like store floor models and upgrades to your own computers. Last time Intel even gave us demo stuff to put on demo models like slide shows, interactive kiosk stuff, and test software. Who cares if AMD has a price and performance advantage on desktops when you can get THIS kind of backing. It just makes to much sense from the point of a small business owner to have this kind of backing.

    AMD has some of this too, but you don't have a sales person you can call up at 3 AM in the morning I am running out of stock and say, ?please send me a hundred more units next day I am running out of stock fast and have a big contract I am starting.? They even know you by first name and send it same day instead of next sometimes for free. Every once in a while on slow days I will just call up my rep and ask whats new in Intel land. Granted I get most of my good non-NDA information from a few engineers and restricted forms.
  • Reply 9 of 31
    telomartelomar Posts: 1,804member
    Quote:

    Originally posted by cwestpha

    When you are getting ready to tap out a new die you usually have a few extra wafers around just ready to quickly be ?thrown in? to the machine and spat out a couple of days latter.



    You've never been anywhere near a manufacturing plant or even spoken to someone who has. The process to manufacture a CPU doesn't just take a couple of days. From tape out, you need to make a mask of the design to begin with. This is a slow and painful process due to the precision required. When the final design is actually completed to manufacture the silicon is a slow process with a lot of repetition (a few hundred steps) that takes considerable time (think over a month).



    Intel doesn't just throw the design in and have the machines spit out silicon a couple days later. Don't you wonder why a chip is taped out a full year in advance of mass production?



    If Intel has only just taped out Merom they aren't testing real silicon yet. That said Intel has definitely taped out Merom and in fact did it well ahead of schedule, hence why they now likely have silicon. The original article was incorrect in saying it had only just been taped out last weekend and they probably meant silicon has just been produced.
  • Reply 10 of 31
    wmfwmf Posts: 1,164member
    Quote:

    Originally posted by cwestpha

    Yonah is like two Sonoma cores glued together with a little bit more tweaking and some new power management features. It wont be as bad as the dual core P4, but its not nearly as ideal as the AMD solution (then again the AMD solution is power hungry and generates more heat then Yonah's solution). For example it looks like Yonah MAY include limited cache to cache transfers. If not Merom will support this feature in some form (it may even share one big cache).



    There is no Sonoma core; I guess you meant Yonah is two Dothan cores. Since both cores share one cache in Yonah, there is no need for cache-to-cache transfers. (I don't know whether it can do L1 interventions or whether the line would have to be written back to L2 and then reloaded, but either way is fast.)
  • Reply 11 of 31
    hirohiro Posts: 2,663member
    Some folks are just sloppy with the terminology and munge taped-out with the first fully working pre-production version, prototype silicon. It's hard to tell nowadays. I don't know whether that is the case with these prototype reports or if they are just fiction.
  • Reply 12 of 31
    Quote:

    You've never been anywhere near a manufacturing plant or even spoken to someone who has. The process to manufacture a CPU doesn't just take a couple of days. From tape out, you need to make a mask of the design to begin with. This is a slow and painful process due to the precision required. When the final design is actually completed to manufacture the silicon is a slow process with a lot of repetition (a few hundred steps) that takes considerable time (think over a month).



    This is correct but I am talking about the steps pre-taped out. When Intel makes a new chip they start by using a proprietary suite of software applications that will assist in the design of processors. These applications help negate as many problems with the initial tap-out as possible along with optimizing die usage and component placement.

    Usually they have test silicon that they will sometimes test ideas out in real life before declaring something has taped out. As for that to happen they have to solve all of the problems in simulation before going on to trouble shooting ?release candidate? silicon.

    This test silicon is usually done in such a way allowing quick alteration (using their ion gun) or quick alteration to a mask. Also Intel will often use mask-less techniques that they are testing at this point.

    They then take the results of the test silicon, make final alterations with the ion gun, and spit out a taped out silicon.

    Next they do what you speak of. They test, make any corrections if possible using the ion gun or make a new temporary mask. This is the point where they test the chip as a whole with a mass production fab. Once they have the bugs worked out they make a few final masks, ship them out to any other lines or fabs that will assist in manufacture of the chip. And there ya go. Granted this is simplified but currently (based on TSMC and Intel Centrino avarages) it will take about 6 to 10 months or so just to get the manufacturing down correctly. Its about a year for the recent steppings of centrino to get from design to taping out.



    Quote:

    Intel doesn't just throw the design in and have the machines spit out silicon a couple days later. Don't you wonder why a chip is taped out a full year in advance of mass production?



    This is partially correct. However Intel DOES make test silicon before this point to see how different components will behave. While it is not ?feature complete? or as complex as the silicon made during the taping out process it still exists.

    However I would like to point out that Sonoma is estimated to have been a 6 to 8 month tap out (depending on who you ask since some include the a/b/g chip and chipset with it). The G70 from Nvidia was a 8 month tap out I believe. I don't have any numbers for the R500 or R520 though (since the latter is not finished yet and the earlier is closely protected by Microsoft still)



    Quote:

    If Intel has only just taped out Merom they aren't testing real silicon yet. That said Intel has definitely taped out Merom and in fact did it well ahead of schedule, hence why they now likely have silicon. The original article was incorrect in saying it had only just been taped out last weekend and they probably meant silicon has just been produced.



    It is important to note that most news sites often report early test silicon as tap out silicon. The difference is once again that the tap out silicon is supposed to get all of the problems out using the final manufacturing process going to be used in mass production. An example would be Sonoma test silicon at 130nm manufacturing during the middle of it's design stage to test new gate throttling techniques using proven manufacturing techniques so they could tell the difference between a real problem that would make it into a final product or a problem with the still early manufacturing technique.



    Quote:

    There is no Sonoma core; I guess you meant Yonah is two Dothan cores. Since both cores share one cache in Yonah, there is no need for cache-to-cache transfers. (I don't know whether it can do L1 interventions or whether the line would have to be written back to L2 and then reloaded, but either way is fast.)



    Yah reading to many tech review sites after a product release tends to blur the lines on code names. They cant keep track of everything and when you read 30 reviews calling Dothan Sonoma it sometimes gets to ya.



    I cant say anything else about Yonah as I am under Intel Channel Conference NDA on some of that stuff still.



    Quote:

    Some folks are just sloppy with the terminology and munge taped-out with the first fully working pre-production version, prototype silicon. It's hard to tell nowadays. I don't know whether that is the case with these prototype reports or if they are just fiction.



    Exactly. I get bits and pieces from Intel people about products a year or so out. Then I read stuff on some sites and suddenly I fully understand exactly what the Intel rep or engineer was telling but couldn't say flat out. Problem is while some of these sites do give some ok information their naming conventions don't fallow the same ones the engineers are always fallowing (like some people think Yonah is actually the name for the new Centrino platform). It gets so bad around release time and after it (news sites just wildly using code names to sound like they know what they are talking about with these new products) and by then I just give up remembering the codenames and start classifying things in my head by product, features, sockets, manufacturing technology, and benchmarks.

    For example I now sometimes mix Sonoma (platform) and Dothan (processor) since most people now ask me for a Sonoma laptop then a Dothan equiped laptop.

    So I just think Sonoma with a 90 nm processor, SSE 2, enhanced FPU, and then approximate performance for the models. Platform has support for PCI express and the new buscard system. HD audio, etc, etc.

    You kinda have to think this when when you sell these products. Once the consumers and press get their hands on all of these new terms classifying things by code names becomes tedious.



    Anyways you have prototype silicon that they will use for testing stuff before taping out (alpha). Then you have taped out silicon where the design is pretty much done and you are just getting it ready for mass production and consumer use (beta and Release Candidate). At least that is kinda the software equivalent.
  • Reply 13 of 31
    telomartelomar Posts: 1,804member
    The report is rather specific in that it is A0 silicon that was tested. That silicon is first run following the tape out of a design and not design process test silicon or CAD testing. I can't imagine any source, who had reasonable knowledge, that would muddle those stages.



    Reference to it taping out last weekend was more likely the site taking an initial production run as completion of the tape out, which is a relatively common misinterpretation. Taping out refers to the "hand over" from the design team to the production team so to speak. It may also occur several times. For instance the latest ATi card is said to have been taped out a couple times to try and improve the design after issues came up following initial manufacture.



    Timescales vary depending on the problems you find during manufacture but test silicon or maskless silicon isn't A0 silicon.
  • Reply 14 of 31
    vinney57vinney57 Posts: 1,162member
    Thanks for the info cwestpha, it rings straight and true. I've always thought that we will see MacIntel announcements acroos the consumer range at SF in January; hence the conservative tweaking of the iBook and Mini line recently.
  • Reply 15 of 31
    mjteixmjteix Posts: 563member
    Quote:

    Originally posted by vinney57

    Thanks for the info cwestpha



    ditto.



    According to the Inquirer, Yonah will be available end of february/early march (instead of january).



    I found another article about the Apple/Intel relations that you may find interesting: Apple Matters
  • Reply 16 of 31
    vinney57vinney57 Posts: 1,162member
    Quote:

    Originally posted by mjteix

    [B]ditto.



    According to the Inquirer, Yonah will be available end of february/early march (instead of january).




    Announce in January, ship in March; 'tis the Apple way.
  • Reply 17 of 31
    smalmsmalm Posts: 670member
    Quote:

    Originally posted by Cosmos 1999

    Translation from French site x86-secret.com

    quote:

    Champagne! Merom boots

    Reliability level: 6/8

    [...] Indeed, after being taped-out last week-end, the Merom A0 die has just been tested and it works! [...]




    Wasn't the tape out of Merom already in june ?
  • Reply 18 of 31
    Quote:

    Originally posted by smalM

    Wasn't the tape out of Merom already in june ?



    No. Again, a misleading information. Merom seems to have been taped-in (design finished and sent to the fab) in June, precisely first or second week of June 2005. This step is before creation of the mask (tape-out). Is this really important anyway?
  • Reply 19 of 31
    smalmsmalm Posts: 670member
    No, but thanks for clearing that up.
  • Reply 20 of 31
    telomartelomar Posts: 1,804member
    Quote:

    Originally posted by Cosmos 1999

    No. Again, a misleading information. Merom seems to have been taped-in (design finished and sent to the fab) in June, precisely first or second week of June 2005. This step is before creation of the mask (tape-out). Is this really important anyway?



    Actually tape out refers back to the days when they used to store designs on magnetic tape. The tape out was when the design was completed and written to tape to be handed over to the next team.
Sign In or Register to comment.