Final Words

Without a doubt, there will be a time when 512MB cards will be necessary. As we mentioned earlier, more local GPU memory is never a bad thing, it just has to be taken advantage of, and that time is not now. 

Performance across the majority of the games that we tested remained unchanged, even at their highest detail settings.  We were a bit surprised that there was such a tangible benefit to the 512MB card running Half Life 2, but in the end, we attained better performance out of a similarly priced X850 XT with only 256MB of memory, even under Half Life 2. 

ATI's desire to make their first 512MB part based on the X800 XL doesn't appear to make much sense either.  The large amount of on-board memory would seem best fit for a GPU that was capable of running at resolutions and detail settings that would see some performance benefit from the additional memory. 

We are worried that ATI doesn't have their users as Priority One, given the level of confusion that the X800 XL 512MB will insert into their product line.  Without a significantly lower price tag, the new 512MB board inadvertently competes with ATI's X850 XT, which is clearly the better performer.  With a 30% fill rate advantage, even in future games that do eat up more memory, we'd expect the X850 XT to still outperform the X800 XL 512MB.  The fact of the matter is that developers aren't going to leave the majority of the gamers in the dark when it comes to implementing features and improving image quality. We wouldn't expect 512MB graphics cards to be the target for any development until the next generation of GPUs actually begins getting into the hands of gamers later this year.  At that point, you're better off picking up a new GPU with more memory rather than one based on the year-old R42x architecture. 

We can't help but feel a little puzzled by today's launch, but at the same time, very curious.  Not so much about the X800 XL 512MB itself, but whether or not ATI will hold true to their word this time around.  The X800 XL 512MB had better arrive later this month; otherwise, ATI will have just dug themselves a deeper hole and this time, it will have been for no actual benefit. 

Unreal Tournament 2004 Performance
Comments Locked

70 Comments

View All Comments

  • Fricardo - Wednesday, May 4, 2005 - link

    *terrible idea*
  • Fricardo - Wednesday, May 4, 2005 - link

    I knew that card was a terrible as soon as I read the title. How could ATI be so retarded? Doesn't even make sense...
  • Patman2099 - Wednesday, May 4, 2005 - link

    I could be mistaken, but isnt the Default FSAA on cards now using Multi sampling or mix of multi and super sampling?

    Perhaps ATI intends to bring back pure Super Sampling?
  • mbhame - Wednesday, May 4, 2005 - link

    Anand,
    There are more sites supporting your findings than contradicting your findings (ExtremeTech) - which wouldn't be a first for ExtremeTech.
    Your final comment in Post #21 couldn't have made me any happier - I'm tired of arguing with people who think slapping tons of RAM on anything dictates tons more performance, etc.
    Could you let us know how much is local and how much is shared system RAM in the recently-announced 512MB 6200's?
    Hey, when are you going to reply to my email? :P
  • Cygni - Wednesday, May 4, 2005 - link

    The card companies pushed 64meg cards when 32meg cards werent even close to maxed out. Then 128 when 64 was perfectly find. Then 256. Then 512. Its what they do, and the consumers have SHOWED the board makers again and again that they will pay out the nose for that extra RAM just to be "future proof", regardless of actual performance. They arent evil, but they have learned what sells. A 64meg card is JUST NOW becoming "needed" in most games.

    The two monitors i have in my house, Sony 200sx's (about 5 of em, all from around 1996 :D ) and early release 17in flatpanels (Dell and Mitsu) both are most useable at 1024x768, so i never even touch reso's higher than that in any games. Its not worth it to me. Personally, the gaganess for high reso, high AA, high AF gaming confuses me... especially at the increadible cost in both FPS and dollars to run that high. The visual difference is less noticeable than simple environ bump maps to me.
  • fishbits - Wednesday, May 4, 2005 - link

    Won't someone think of the poor starving gamers that don't have 256MB DDR3, much less an extra thus far-useless, overpriced 256MB of DDR3?

    Perhaps it could come in handy though in MOGs, where you're going back and forth through different zones and coming across the same characters multiple times, and it isn't as clearly defined as loading the next FPS level/map. "Hey, Jerx0r the Brave is still in video memory, don't have to reload him!" Of course it may not be a big hit if its already in system RAM, but with high-end systems running 1GB + 256MB, it's a decent bump up in total RAM.
  • AlexWade - Wednesday, May 4, 2005 - link

    Someone will buy it. Someone with deep pockets. If you are thinking about buying this card, can you give me some of your extra money?
  • Anand Lal Shimpi - Wednesday, May 4, 2005 - link

    ET

    I meant that it's not related to video memory size. If it were system memory related then it would affect both cards equally, but the test is cached and I noticed no disk related slowdowns during the benchmarking. If you really want I'll run it with 2GB of memory though, just ask and I shall do :)

    Take care,
    Anand
  • ET - Wednesday, May 4, 2005 - link

    Thanks for the comment, Anand. Why do you say "it's not related to memory size". Do you get the same scores with 2GB of RAM?
  • Anand Lal Shimpi - Wednesday, May 4, 2005 - link

    For those wondering, I also ran tests under Doom 3 at 1600x1200 with 4X AA (8X AF) on the High Quality setting. The performance between the two cards is as follows:

    512MB - 31.0 fps
    256MB - 31.0 fps

    This agrees with the High vs. Ultra performance numbers we saw in our first Doom 3 GPU review [http://www.anandtech.com/video/showdoc.aspx?i=2146...].

    There's no magic increase in performance if you back down to the High quality setting, so if you're seeing numbers contrary to what I've shown here something fishy is going on and it's not related to memory size.

    For those of you wondering why Doom 3's Ultra Quality mode doesn't require a 512MB video card, keep in mind that the test we're running here is reporting average frame rates. Doom 3's timedemo function doesn't output minimum frame rates, but through actual game play the 512MB card is a bit smoother with the Ultra quality settings enabled (only in Doom 3 however). The difference isn't huge in the scenarios we tested.

    I have no doubt that 512MB boards will make sense sometime down the road, but not paired up with this GPU.

    Take care,
    Anand

Log in

Don't have an account? Sign up now