Page 46 of 49 FirstFirst ... 364445464748 ... LastLast
Results 451 to 460 of 486

Thread: 28nm APU speculation.

  1. #451
    Quote Originally Posted by sirroman View Post
    People complaining about Brazos 2.0. saying that it should be replaced by a die-shrunk or whatever: AMD intended to release a Krishna/Wichita, on 28nm and this year, but it's very likely that the economics wouldn't be right, since GF's 28nm wasn't ready on time (at least in the volume side, they need to ramp up AND the new fab isn't working at 100% yet). On top of that, that market segment needs a lot of volume.

    So AMD cut out those products (which is a shame for I was excited about Wichita) and extended the shelf-life of Brazos, rebranding it as Brazos 2.0. AT BEST, AMD did a respin, which is much cheaper and faster to do. (but we don't know if there's any respin)

    Answering "why they didn't release Brazos 2.0 last year?" (or something along those lines), it's simple, because you prepare a product and launch it, after you have final silicon you learn more about that implementation and THEN you can tweak it. Also, clock increases during a chip shelf life is not unheard of, it's VERY common and means that AMD and the foundry partner (in this case TSMC) are doing better chips due to increased learning from actually making those chips.

    If you look at the history, it's actually amazing that AMD could churn out Z-series so fast yielding a power consumption reduction so big, all with the same silicon.

    OFC, those are all my opinions and conjectures. Feel free to think differently.
    I have no problem with Brazos at all. It was a wonderful product well executed and performed admirably against competition. The issue is not neccessarily about die-shrink or lack of it. I mostly look at it on how it helps AMD and less than truthful marketing...

    When Brazos family was introduced the Atom competition was intentionally crippled by Intel, fabbed in a 45nm process. Too many restrictions were put by Intel on the hardware & software stacks that need to be present. At this time Brazos with OOO, x64 capability and a VCE and mild gaming capability was heads and shoulders above Atom variants. Well the competitive world does not stop around early 2011 eh!

    Fast forward, Dec2011 Atom is now fabbed in 32nm HKMG. Also it got a Video Engine with Blue Ray 2.0, 1080p support an other things that previously were strong points of Brazos. Granted gaming still sucks compared to Brazos but who would buy netbooks at those price points for games anyway. Going forward Atom is going to have the benefit of fabbing under the best available process

    Silvermont 2013 - 22nm
    Airmont 2014 14nm

    But 1 thing undeniable that has happened with Atom was that what it failed in performance against Brazos, it had gains in power. Each shrink would produce better performance under the same power envelope as well as CPUs that have TDPs that will go lower than the previous generation.

    45nm Pineview had single & dual configs that range from 6.5W-13W, the 32nm Cedarview had all duals with 3.5W-10W. Compare that with Brazos 2.0 which looks like nothing significant at all, just a 100MHz CPU clock speed increase & some GPU core speed increase. So tell me how in scale is it different from the C0 stepping of Ontario or the E-450 zacate?

    I remember John Fruehe(who now very rarely comments here) commenting before that "Brazos is architected to scale below while Bulldozer is architected to scale above". Why did Brazos 2.0 not provide a 6W Ontario successor? The addition of USB3 in the SB, will it reduce the power characteristics of the platform? Does it help to expand the market or provide something that will cement it further in the market? These are the strategic things that I thought RR and team would deliver. And AMD has an age-old tradition. When there is a product that improves the prior product in a minimal fashion, it generally releases it without as much as a whimper. That is how the last few Opterons were released. Brazos 2.0 should have been like that in my opinion. While I have no problems with Trinity being highlighted at Computex(while I do worry about very little by its partners), Brazos 2 is a entirely different story...

    Whether AMD has done a spin or re-spin of the Brazos core, looks like there are some people in this forum spinning things more than AMD marketing really can.

    Cheers!

  2. #452
    Quote Originally Posted by gljvd View Post
    There isn't a 28nm bobcat because AMD screwed the pooch .
    Actually I think it is Global Foundries.
    a) It hurt AMD in Llano yields in 2011.
    b) AMD added its own hurt itself with the Bulldozer core which could overclock to 9GHz but ran at a official clock of < 4GHz. Again that falls to GF SOI that would fail to deliver the required clock speeds at the anticipated power envelopes.
    c) AMD still does not officially know what is the path its server MPUs will take after 32nm SOI. While its APUs (Steam Roller) are going to move to bulk HKMG its server roadmaps(Abudhabi, Delhi, Seoul) as well as desktop/WS MPUs(Vishera) still list only 32nm SOI as the process technology as far as 2013.

    It screwed AMD on getting its 28nm bulk HKMG ready thus resulting in the cancellation of Wichita/Krishna APUs and AMD scrambling to run towards Brazos 2 as a stop gap measure. Also GF had only 1 32nm process running which was SHP or SOI. Only in 28nm node, did they have an idea of including both bulk and SOI when I looked at their marketing presentation a while back. Now they are totally silent on 28nm SOI.

    http://www.globalfoundries.com/technology/32-28nm.aspx
    SLP, LPH, HPP refer to their bulk wafer technologies

  3. #453
    Quote Originally Posted by Third Eye View Post
    Actually I think it is Global Foundries.
    a) It hurt AMD in Llano yields in 2011.
    b) AMD added its own hurt itself with the Bulldozer core which could overclock to 9GHz but ran at a official clock of < 4GHz. Again that falls to GF SOI that would fail to deliver the required clock speeds at the anticipated power envelopes.
    c) AMD still does not officially know what is the path its server MPUs will take after 32nm SOI. While its APUs (Steam Roller) are going to move to bulk HKMG its server roadmaps(Abudhabi, Delhi, Seoul) as well as desktop/WS MPUs(Vishera) still list only 32nm SOI as the process technology as far as 2013.
    Brazos was fabbed at TSMC so a new verison could have been fabbed on 28nm or they could have still droped from 40nm to 32nm at GF

  4. #454
    8-bit overflow
    Join Date
    Apr 2012
    Posts
    483
    Quote Originally Posted by gljvd View Post
    Brazos was fabbed at TSMC so a new verison could have been fabbed on 28nm or they could have still droped from 40nm to 32nm at GF
    please think before you talk. None of that is possible without money and time and neither of with would be worth the investment. That is all there is to it. By the time they have a die shrink ready from TSMC or 32nm GLOFO, Kabini could have been released instead.

    Brazo 2.0 was a stop-gap product because glofo could not come out with its 28nm production. AMD cannot just switch processes on the fly. Process shrinks could take 9-12 months. Considering the plans were revised about jan 2012, why would AMD bother to invest time and money into something that won't be ready by until 2013 anyways it would also give AMD nothing for 2012 in that segment. AMD would be better served to plan for a new chip for mid 2013 like kabini is now planned.

    This is basic economics and planning. Processors aren't made in a day. You can whine all you want but it won't magically give AMD enough money to throw at a die shrink just for you.

  5. #455
    8-bit overflow
    Join Date
    Nov 2010
    Posts
    343
    Its not as if Brazos on 28nm will take a year to launch from here on.

    Brazos 2.0 is to gap the remainder of 2012, which is around 6months.

  6. #456
    Quote Originally Posted by esrever View Post
    please think before you talk. None of that is possible without money and time and neither of with would be worth the investment. That is all there is to it. By the time they have a die shrink ready from TSMC or 32nm GLOFO, Kabini could have been released instead.

    Brazo 2.0 was a stop-gap product because glofo could not come out with its 28nm production. AMD cannot just switch processes on the fly. Process shrinks could take 9-12 months. Considering the plans were revised about jan 2012, why would AMD bother to invest time and money into something that won't be ready by until 2013 anyways it would also give AMD nothing for 2012 in that segment. AMD would be better served to plan for a new chip for mid 2013 like kabini is now planned.

    This is basic economics and planning. Processors aren't made in a day. You can whine all you want but it won't magically give AMD enough money to throw at a die shrink just for you.
    January ? I think not and charlie thinks not , he has known since Nov 2011
    http://semiaccurate.com/2011/11/15/e...a-and-krishna/

    Amd would have known before charlie . AMD could have been working on A Brazo shrink on TSMC 28nm or GF 32nm From Nov 2011 . June would be 8 months from that. From what I can tell the last leak that had witchita/krishina was Sept that of course would put us at 10 months. The plans could have changed in the summer giving them even more time .

    Now that would work with your sugestion of 9-12 months for a process shrink. Perhaps process shrinks take less time than that . I don't know


    Perhaps AMD thoguth they could pull in a 2013 part but aren't going to make it at this point ? Who knows , I can just tell you looking at this weeks annoucnements AMD was no where to be seen both trinity and brazos apeared only in Compal prototypes from AMD .

  7. #457
    Quote Originally Posted by gljvd View Post
    January ? I think not and charlie thinks not , he has known since Nov 2011
    http://semiaccurate.com/2011/11/15/e...a-and-krishna/

    Amd would have known before charlie . AMD could have been working on A Brazo shrink on TSMC 28nm or GF 32nm From Nov 2011 . June would be 8 months from that. From what I can tell the last leak that had witchita/krishina was Sept that of course would put us at 10 months. The plans could have changed in the summer giving them even more time .

    Now that would work with your sugestion of 9-12 months for a process shrink. Perhaps process shrinks take less time than that . I don't know


    Perhaps AMD thoguth they could pull in a 2013 part but aren't going to make it at this point ? Who knows , I can just tell you looking at this weeks annoucnements AMD was no where to be seen both trinity and brazos apeared only in Compal prototypes from AMD .
    Sorry, but as stated before, it takes a lot (and I mean a lot) of time and money to bring out a chip on a certain process. A design of a chip is completely tuned to that process. If you want to change a certain chip to an other process (so like you would like to do from 28 nm to 32 nm) you almost have to completely start from scratch with you layout. This is something you just do not do.

  8. #458
    Quote Originally Posted by MonBl71 View Post
    Sorry, but as stated before, it takes a lot (and I mean a lot) of time and money to bring out a chip on a certain process. A design of a chip is completely tuned to that process. If you want to change a certain chip to an other process (so like you would like to do from 28 nm to 32 nm) you almost have to completely start from scratch with you layout. This is something you just do not do.
    Oh, not that is really that important, I have been following this forum for a few years now. Thought about creating an account a lot over this period and recently finally did so. So, including this message I now have written 2 messages .

  9. #459
    itanic
    Join Date
    Nov 2011
    Posts
    139
    Quote Originally Posted by Third Eye View Post
    Now they are totally silent on 28nm SOI.

    http://www.globalfoundries.com/technology/32-28nm.aspx
    SLP, LPH, HPP refer to their bulk wafer technologies
    The current 32nm SOI is actually PD-SOI (Partially Depleted). During the Technology Forum 2012, the Common Platform decided to go bulk at 28nm and 20nm. Anna Hunter, VP Foundry Business, Samsung said as follows:

    The problem with the current partially-depleted SOI (PD-SOI) technology is that the pressure needed for SOI insulation to take place is decreasing yields due to pressure on the already strained silicon. When you pressure the strained silicon transistors, they tend to break. This is also the main reason why Common Platform Manufacturing Alliance, manufacturing arm of Common Platform Technology Alliance decided to go “SOI less” and go bulk with the 28nm and 20nm processes.
    http://www.brightsideofnews.com/news...m-process.aspx

    It seems that only STMicroelectronics and a foundry partner (probably IBM because they have a three-way alliance with STMicroelectronics and Leti) are working on FD-SOI (Fully Depleted or Extremely Thin) at 28nm:

    http://www.advancedsubstratenews.com...s-from-soitec/

    It is true that GlobalFoundries had problems with the 32nm SOI process. One should, however, also consider the cost of 28nm FD-SOI R&D process. Who is going to cover the cost of the 28nm SHP process? Given the market penetration as well as performance of the AMD MPUs, is it worth it?

    The AMD CEO said that they will move ahead on 28nm with GlobalFoundries.

    http://semimd.com/blog/2012/03/22/gl...1;-at-dresden/

    The future of AMD MPUs will probably not be clear until the Financial Analyst Day in 2013.

    On the other hand, nothing is set in stone though. It seems Intel's FinFETs are not without problems:

    http://www.eetimes.com/electronics-n...SOI-shrink-GSS

  10. #460
    Quote Originally Posted by gljvd View Post
    Amd would have known before charlie . AMD could have been working on A Brazo shrink on TSMC 28nm or GF 32nm From Nov 2011 . June would be 8 months from that.
    Building new mask costs money. And such kind of products requires high volume. They can shrink Ontario to 28nm, but is it still profitable after deducting the additional cost from the new 28nm mask and the cancelled 28nm mask? Given that it has to launch before mid 2012 while the Kabini APU with new core microarchitecture is still scheduled to hit the market in early 2013, I personally think it isn't worthy.

    Cancelling masks costs no money, but the development cost is pretended to be covered by later products.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  
WordPress Appliance - Powered by TurnKey Linux