ATI Premium Graphics Logo

Upcoming ATI Cards to be Built on 40nm

Add Your Comments

Image courtesy: AMD Corp.

Image courtesy: AMD Corp.

The planned next generation graphics parts from AMD/ATI, known as Northern Islands, was initially designed to be manufactured on Taiwan Semiconductor Manufacturing Company’s (TSMC) 32nm process. Since the late 2009 announcement that TSMC was facing difficulties with its 32nm manufacturing node and would not be developing the process, plans had to change at AMD. Sources close to both companies indicate that the revised upcoming generation of graphics cards will most likely be made for the Taiwanese manufacturer’s established 40nm process and will feature design elements from both Northern Islands and Evergreen architectures.

Problems encountered by the main manufacturer of AMD graphics chips, TSMC, are likely responsible for this decision. Since it has been decided by TSMC to pull its 32nm process, the company also had to make changes to their initial approach at 28nm, resulting in more delays and uncertainty in the manufacturing schedule. The company is currently developing its new 28nm process but it is unlikely to be ready in time for the HD6000-series launch window. Meanwhile, AMD spin-off manufacturer Global Foundries is working on their 28nm process but it is not scheduled to be ready for mass production before late 2010. In light of the information coming from the foundries regarding 28nm delays, Xbit Laboratories believes AMD decided to go for the safer option: the proven 40nm process from TSMC.

Writer Charlie Demerjian, from SemiAccurate, has advanced that the new hybrid architecture, dubbed Southern Islands, could feature the shaders design of the Evergreen HD5000-series combined with the uncore parts of the Northern Islands architecture. The new approach is likely due to the impossibility of implementing all the features of the new architecture without moving to a new, smaller process.  According to both sources, customers are looking at a launch scheduled for Q4 2010.

Sources:  Xbit Laboratories and SemiAccurate

-dostov

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>

Discussion
  1. dostov
    According to Xbit Labs, both manufacturer promise their 28nm processes will be ready for mass production in late 2010, but you have to take that kind of info with a (large) grain of salt.

    More at Xbit: http://www.xbitlabs.com/news/video/display/20100401171410_Forthcoming_ATI_GPUs_Different_Processes_Different_Suppliers_Hybrid_Architecture.html

    Also, to release new products in Q4, ATI have to start mass production in Q2 2010. TSMC takes no less than 10 weeks to manufacture chips. Add to that the time to put the chips on the boards, box them and ship... 28nm products won't come out before Q2 2011.


    Interesting stuff. Given how Fermi turned out ATI could afford to wait until then as it has the ability to reduce the prices of its existing cards and still turn a profit, while Nvidia can't do that.

    It will be very interesting to see what ATI does with Northern Islands.
    Neural Net
    Are there any articles that state how far along GF's 28nm process is? Is it no longer on track for Q4 2010/Q1 2011?


    According to Xbit Labs, both manufacturer promise their 28nm processes will be ready for mass production in late 2010, but you have to take that kind of info with a (large) grain of salt.

    Both Globalfoundries and TSMC promise that their 28nm high-performance (28HP) process technologies will be available for mass production already in late 2010. But there are chances that TSMC’s process technology will arrive somewhat later than expected, whereas Globalfoundries does not have a track record of quick and flawless ramp up of bulk fabrication processes.


    More at Xbit: http://www.xbitlabs.com/news/video/display/20100401171410_Forthcoming_ATI_GPUs_Different_Processes_Different_Suppliers_Hybrid_Architecture.html

    Also, to release new products in Q4, ATI have to start mass production in Q2 2010. TSMC takes no less than 10 weeks to manufacture chips. Add to that the time to put the chips on the boards, box them and ship... 28nm products won't come out before Q2 2011.
    Shiggity
    Global Foundries's 28nm node isn't ready for primetime yet.

    But architectural improvements on the same node isn't a bad thing.

    Just think of a 45nm core2duo wolfdale vs. 45nm Nehalem. Around the same transistor count, big performance increase.

    Work out your architectural bugs on a cheaper and safer node, then port it down.


    Interesting stuff, the only problem I see is that ATI's small chip philosophy seems to require a new process for each generation (adding features and power but reducing die size).

    Are there any articles that state how far along GF's 28nm process is? Is it no longer on track for Q4 2010/Q1 2011?
    killem2
    What does this mean for the normal person? I thought the point of a refresh is to get smaller chips to get better consumption and heat?


    The point of a new architecture is to bring better performance and new features. The point of a new, smaller process is to keep power consumption and heat at bay as the new architecture implements more transistors.

    This is why ATI's Northern Islands architecture was set for 32nm node. Unfortunately the manufacturing process isn't going to be ready so they have to adapt. It is believed they made changes to the architecture in order to implement as many new features and performance increases as possible, but not the complete new architecture because of the 40nm process being too big (too much power consumption --> too much heat).

    I hope this helps in understanding!

    Cheers! :)
    Global Foundries's 28nm node isn't ready for primetime yet.

    But architectural improvements on the same node isn't a bad thing.

    Just think of a 45nm core2duo wolfdale vs. 45nm Nehalem. Around the same transistor count, big performance increase.

    Work out your architectural bugs on a cheaper and safer node, then port it down.