Well here's one story that managed to slip behind our desk for a few days: Samsung has finally begun mass production of their 900MHz(1.1ns) GDDR3 chips. As the primary high-end memory supplier in the video industry, ATI and NVIDIA effectively live and die by what and how soon Samsung can offer bigger and faster memory chips, so the ramifications of this are going to be pretty straightforward.

The timing on this is fairly consistent with what little information we know about ATI and NVIDIA's winter-refresh products due some time in Q1 2006, so mass production of the necessary memory a couple of months ahead of time is fairly normal. These are all 32bit-wide 512Mb chips, so the sweet spot for any cards using them will need to be at least 512MB, since 8 chips are necessary to achieve a proper 256bit memory bus (so those of you looking for a 7900 GTX 256 may be out of luck). For those users worried about the power consumption of the latest video cards, don't look for things to get any better with these memory modules; it's no fluke that the 7800GTX 512 and X1800XT ran hotter than previous cards, as these are 2.0V chips, so if there's a drop in power usage with the refresh cards, it won't be due to the memory.

The more interesting factor at this point in time however is what this means for NVIDIA's oft-lauded and hard to find GeForce 7800GTX 512. It's no secret that while NVIDIA managed keep with their press-favorable policy of hard-launching the card with the announcement, they haven't been able to keep the card in any regular supply (much to the chagrin of eVGA owners in the step-up queue), and its rumored that the supply of this memory had a great deal of involvement in this problem. While NVIDIA certainly is not apt to talk about the problem, given that this memory only entered mass production a month after the GTX 512 launched, we're inclined to believe that this is indeed the case.

Now if NVIDIA and its partners intend to use some of these mass produced parts to pick up GTX 512 production or horde it all for the winter refresh remains to be seen, although it is more likely they'll horde it given the difficulties and amount of stock required to hard launch a high-end video card. So consumers still looking for a GTX 512 may be out of luck and better off waiting for the winter refresh. We can't say we would disagree with the decision to use the current stock of memory to build up the refresh if this is the case, though it doesn't excuse our disappointment in NVIDIA for being unable to regularly supply the GTX 512. Certainly some initial exhaustions of stock are expected for a new card, but a hard launch is only as good as the card supply, and this is a case where NVIDIA has failed.

With the mass production of their 900MHz modules, Samsung is nearly done with GDDR3 as their headliner graphics memory. One last refresh is due in 2006, 1GHz modules which Samsung is still working on, after which they will move to GDDR4. Samsung expects to start mass production of GDDR4 in Q2 2006 with speeds of at least 1.25GHz, so everyone can get out their calendars now and start dreaming about what the summer 2006 lineup will look like.

POST A COMMENT

11 Comments

View All Comments

  • ShadowVlican - Thursday, December 22, 2005 - link

    7800GTX 512MB hard to find? i can go out right NOW, drive 5 minutes to a local computer store, and come back less $1000cdn with a brand new spanking ASUS 7800GTX 512MB... and i live in Canada btw lol Reply
  • ViRGE - Friday, December 23, 2005 - link

    Then you're fortunate. AnandTech's RTPE shows that the 512 is http://labs.anandtech.com/search.php?q=GTX+512">out of stock everywhere, as it has been for weeks. Reply
  • mat128 - Saturday, December 24, 2005 - link

    A major supplier in Canada reported an available quantity of 22 as of dec. 22 2005 for the 7800GTX/2DHTV/512M (that's the asus one).

    Enjoy =)
    Reply
  • bobsmith1492 - Thursday, December 22, 2005 - link

    "ram"ifications.... very punny. :P Reply
  • Ryan Smith - Friday, December 23, 2005 - link

    It wasn't intentional, I swear. It just happened. Reply
  • stephenbrooks - Thursday, December 22, 2005 - link

    The first thing that fell out of Google
    http://www.xbitlabs.com/news/memory/display/200405...">http://www.xbitlabs.com/news/memory/display/200405...
    seems to imply that there's no additional "redoubling" of the data - it's still a sort of DDR (and not QDR or anything like that), just with some tweaks to let it hit in the region of 1.4GHz.
    Reply
  • Brian23 - Thursday, December 22, 2005 - link

    So then what is it about GDDR4 that lets it get up to 1.25GHz when GDDR3 is stuck at 900ish MHz? Reply
  • Furen - Thursday, December 22, 2005 - link

    There's 1GHz GDDR3 parts, but they're not in mass production. I would guess that GDDR3 will be able to hit around 1.1-1.2 GHz but it may not be worth the effort (lousy yields, high power consumption, etc) to mass produce them. Reply
  • stephenbrooks - Thursday, December 22, 2005 - link

    When they say 1.25GHz GDDR4, do they mean 625MHz "double-pumped GDDR3"? I'm a little confused by which of these RAM types have 1x 2x and 4x the standard bandwidth. Can anyone here explain? And what will the latencies be like comparatively? Reply
  • Anton74 - Thursday, December 22, 2005 - link

    To my understanding, the frequencies mentioned here are the base clocks, which means that their effective frequency is twice that (it is DDR - Double Data Rate).

    How much actual bandwidth you have at the end of the day also depends on the width of the data path of course - 32 bit per chip in this case.
    Reply

Log in

Don't have an account? Sign up now