Inside The Pipes

The pixel pipe is made up of two vector units and a texture unit that all operate together to facilitate effective shader program execution. There are a couple mini-ALUs in each shader pipeline that allow operations such as a free fp16 normalize and other specialized features that relate to and assist the two main ALUs.



Even though this block diagram looks slightly different from ones shown during the 6800 launch, NVIDIA has informed us that these mini-ALUs were also present in NV4x hardware. There was much talk when the 6800 launched about the distinct functionality each of the main shader ALUs had. In NV4x, only one ALU had the ability to perform a single clock MADD (multiply-add). Similarly, only one ALU assisted in texture address operations for the texture unit. Simply having these two distinct ALUs (regardless of their functionality difference) is what was able to push the NV4x so much faster than the NV3x architecture.

In their ongoing research into commonly used shaders (and likely much of their work with shader replacement), NVIDIA discovered that a very high percentage of shader instructions were MADDs. Multiply-add is extremely common in 3D mathematics as linear algebra, matrix manipulation, and vector calculus are a huge part of graphics. G70 implements MADD on both main Shader ALUs. Taking into account the 50% increase in shader pipelines and each pipe's ability to compute twice as many MADD operations per clock, the G70 has the theoretical ability to triple MADD performance over the NV4x architecture (on a clock for clock basis).

Of course, we pressed the development team to tell us if both Shader ALUs featured identical functionality. The answer is that they do not. Other than knowing that only one ALU is responsible for assisting the texture hardware, we were unable to extract a detailed answer about how similar the ALUs are. Suffice it to say that they still don't share all features, but that NVIDIA certainly feels that the current setup will allow G70 to extract twice the shader performance for a single fragment over NV4x (depending on the shader of course). We have also learned that the penalty for branching in the pixel shaders is much less than in previous hardware. This may or may not mean that the pipelines are less dependent on following the exact same instruction path, but we really don't have the ability to determine what is going on at that level.

No More Memory Bandwidth No More Shader Replacement
Comments Locked

127 Comments

View All Comments

  • mrdeez - Thursday, June 23, 2005 - link

    Derek-
    Please post benches with resolutions that are commonly used or this article becomes a workstatin graphics card article and not one for gamers....I mean really 2046x3056 or whatever the hell...lol...#1 who games at that res??? While this card is powerful it should be mentioned that unless you use a res over 1600x12000 this card is unnecessary.......lol those were some ridculous resolutions though.......and again post some benches with 1280x1024 for us lcd'ers.....
  • Shinei - Thursday, June 23, 2005 - link

    #95: Did you pay to read this article? I know I didn't...

    #94: I guess you missed the part where they said that all resolutions below 1600x1200 were essentially identical in performance? If you only play in 1024x768, why are you reading a review about a $600 video card--go buy a 6600GT instead.
  • jeffrey - Wednesday, June 22, 2005 - link

    #83

    Has the staff at Anandtech not ever heard of "Vacation Coverage"?

    The excuse of your Web Editor being on vacation is, in reality, an admission of improper planning.

    A major hardware site that is dedicated to cutting-edge technology should have planned better. New high-end GPU launches happen by NVIDIA only about 2-3 times a year at most.

    This was one of the HUGE launches of the year and it was messed-up becuase the team didn't feel it was important enough to get some help for the article. There was damage done to Anandtech today due to the article errors and due to the casual admission in post #83 about not caring to properly cover a "Super Bowl" type of product launch today.

    Save your apologies to the message board, give them to Anand.
  • geekfool - Wednesday, June 22, 2005 - link

    How about benchmarking some useful resolutions? This review was essentially useless.
  • JarredWalton - Wednesday, June 22, 2005 - link

    86 - Trust me, most of us other editors saw the article, and quite a few of us offered a helping hand. NDAs a a serious pain in the rear, though. Derek was busy pulling all nighters and functioning on limited sleep for several days, and just getting the article done is only half the battle. Getting the document and results into the document engine for a large article with a lot of graphs can take quite a while and is an error prone process.

    The commentary on the gaming benchmarks, for instance, was written in one order and posted in a different order. So please pardon the use of "this is another instance" or "once again" when we're talking about something for the first time. Anyway, I've got a spreadsheet of the benchmarks from early this morning, and other than non-functional SLI in a few games, the numbers appeared more or less correct. The text also didn't have as many typos. Crunch time and getting the final touches put on a major article isn't much fun.

    Thankfully, I'm just the SFF/Guide man, so I'm rarely under NDA pressure. ;)
  • robp5p - Wednesday, June 22, 2005 - link

    I would love to see someone start benchmarking in widescreen resolutions! 1920x1200 begs for a fast video card like this. As was pointed out, the only real benefits of the 7800 come at high resolutions, and many people buying high resolution monitors these days are getting widescreen LCD's

    and btw, my 2405fpw is sitting in a box right next to me in the office, begging me to open it up before I get home...this thing will be fun to get home on the subway
  • patriot336 - Wednesday, June 22, 2005 - link

    Where is the Monarch and Tiger love?
    TigerDirect http://www.tigerdirect.com/applications/searchtool...

    Monarchcomputer http://www.monarchcomputer.com/Merchant2/merchant....

    Both are 599.00$
  • BikeDude - Wednesday, June 22, 2005 - link

    $600 for a card that only features single-link DVI outputs? Yeah right, pull the other one nVidia!

    --
    Rune
  • ta2 - Wednesday, June 22, 2005 - link

    As a player of eve-online, I can tell you that the game is entirely CPU dependent. On that matter, it will 100% any CPU you have. I mean ANY CPU. Also for the testing, you should use 1600x1200 max AA and AF and go into an area with many player ships on eve-online. I guarantee you will not get 60 FPS. Impractical and unscientific, but would still give better results than this review.
  • TinyTeeth - Wednesday, June 22, 2005 - link

    I am very impressed of the performance of the new chip. Nvidia seems to have fixed the problems SLI had during the 6800 generation.

    I am also pleased they have managed to deliver the cards so quickly. That also puts some pressure on ATI.

Log in

Don't have an account? Sign up now