• Join Us |
  • |
  • Sign in with:

ATI Rage6 Details

by Steve Gibson, Apr 20, 2000 6:07pm PDT
Related Topics – ATI

Well, we've seen info on the Voodoo5 from 3dfx and NV15 (kinda, here , here, here, and here) from NVidia, why not see what ATI has going? IXBT has a preview of the new Rage6 chipset which may finally be a contender with the top of the performance market.

- 25 million transistors - 2 rendering pipelines with 3 texturing units each - 200-400MHz core, 200mhz memory - Pixel fillrate: 400-800 million pixels per second - Texel fillrate:     1 texel per 1 pixel: 400-800 million texels per second     2 texels per 1 pixel: 800-1600 million texels per second     3 texels per 1 pixel: 1200-2400 million texels per second - 8-128MB local graphics memory - 350-400MHz RAMDAC integrated into the graphics core
10:10update: You might wanna check out this SharkyExtreme article ont he chipset as well while you're at it. Thanks ilya.




Comments

77 Threads* | 137 Comments







  • Dunno if anyone's seen this. Someone was asking about how cool DirectX TC would be.

    "Microsoft Licenses Breakthrough 3D Graphics Technology from NVIDIA
    NVIDIA's Volume Texture Compression Formats for Microsoft's® DirectX® Enables Rich Graphics for Internet and Games
    SANTA CLARA, CA — March 23, 2000 — In a move that will help bring stunning 3D graphics to internet users, NVIDIA™ Corporation (Nasdaq: NVDA) and Microsoft® Corp. today announced the adoption of NVIDIA's technology for Volume Texture Compression Format (VTC) for Microsoft DirectX APIs. Today's 3D internet sites are plagued with inadequate image quality due to bandwidth constraints of transmitting high-resolution textures over standard communications systems. Even for high-performance PCs, the limited amount of texture storage forces game developers to use lower-resolution textures, resulting in imagery that lacks detail. Scheduled for release in 2000, NVIDIA's VTC format enables a superior level of image quality that allows web and content developers to produce 3D objects that depict their natural characteristics."

    From: http://www.nvidia.com/News.nsf

    Boogieman






  • Fucking hell. I started disliking Sharky's site when they changed from the black look to that current Cnet-clone shit. I loved the site until that, and I hated the new design. Ever since that change, that site has just gone into the shitter. I kept getting more and more annoyed by the writing, and eventually just stopped visiting the site at all (used to visit it every day). Now, every time I hear about a decent article being posted and go there to read it (or try to, rather), I just shovel thru pages and pages of pointless rambling bullshit, only to reach the end and think "What the fuck was that??". That ATI article was like 15 fucking pages long, I wasn't even reading it and it still took me 5 minutes just to click thru the fucker. Blah blah about bump mapping, blah blah about why more polygons are better than fewer polygons. Fucking DUH!! Jesus fucking Christ, quit raping me for banner ad impressions and just give me some fucking info. Fucking sell-out bitch.











  • Ok Pimpstick in the case of the GeForce you have 4 pipelines. Each pipeline renders a "pixel". At 120mhz you multiply 120 x 4 = 480pixels/sec. Then you include multitexturing (another texture on the pixel). In order to do multitextrung the GeForce has to use one of its 4 pipelines to render the other "texel" on top of the "pixel". This in effect cuts the fillrate in half to 240mpixels/sec. Since you need to use one of the 4 pipelines for the extra "texel pass".

    This in effect is the way most modern 3d acelerators run, V5 runs similiar in that it can't do "free" multitexturing & keep its 2 pixels/clock. It only renders 2pixels/clock when not in multitexturing mode. The NV15 (according to the preliminary pdf file) has 4 pixel pipelines, BUT each pipeline is capable of running another "texel". This in effect does NOT cut fillrate by 1/2 when running in multitexture games since each pipeline is capable of its own free multitexture without actually using one of its 4 pipelines in order to do multitexturing.

    If you do the math on the 1600m/texels you figure 200mhz * 4 = 800mpixels but since each pipeline can render 2 texels you go 800 * 2 = 1600m/texels sec. The V5 as stated earlier (just like geforce/nv10) loses 1/2 fillrate when doing multitexturing.

    In other words if there is not some other bottleneck (possible bandwidth?) the NV15 will be insanely fast in multitexture games.

    Hopefully now you will understand & then you can figure out how the 3 texel per pipeline (2 of them) on the ATI Rage6 works.

  • All these hype numbers mean shit until they hit the bench, IMHO.

    ATI has crazy numbers, Nvidia is throwing out some crazy numbers, and so is 3dfx.

    Nvidia claimed 800Mpixels for the GeForce in early hype and even when it was released at 480Mpix it only worked with the DDR and 480Mpix still wasn't acheivable due to memory bandwidth and texture management design flaws(the DDR was supposed to fix the memory problem and it sorta did but not completely). Nvidia's next chip better not have more than one chip nor need an outside power connector or they will be guilty of doing things they swore they would never do. I think their $349 card will resemble the V5-5500 very closely in performance. The 1.6 is their equivalent of 3dfx's V5-6000.

    3dfx claims 1.33 or 1.5Gpixels for it's V5-6000, although the smart money says that the V5-6000 will resemble the V3-3500 and come out so late and have so many problems that it will probably be D.O.A. 3dfx's real hopes lie in it's next chipset, one they have poored tons of cash and resources on developing.

    ATI also claimed some impressive stats for it's top of the line Rage Fury Maxx and although an excellent card on paper, the driver problems keep it from being a real contender for the hardcore crowd. And the fact that it's a multiple chip solution lowered it in the purists eyes.

    Anyway, benchmarks are what matters.