Site hosted by Angelfire.com: Build your free website today!
 

rage128-titre.gif (1844 octets)

 

r128chip.jpg (7322 octets) English translation of French article from: http://france.gagames.com/tech/rage128.html
.
.
.
.
. Here it is, I am able finally to offer an exclusive article to you on PHS. Nothing less than astonishing this article treats forthcoming chip of ATI, RAGE 128. Indeed, ATI is the first company in which I had a contact. Thanks to this contact, I have true first hand hardware now, then level you.

ATI has always had a presence in the world of the display, and this as well on the level of the graphics boards as to that of the cards TV, MPEG... Indeed, the cards of the Canadian manufacturer were especially famous for their multiple capabilities (Rage 2 was an old chip 2D/3D like Virge and having an output TV, the Rage Pro is a chip 2D/3D with output TV, decompression MPEG...), and not inevitably for their performances: not the Rage Pro was definitively not Voodoo killer.

And then more nothing: since this last chip 2D/3D which made its glory at OEMs, ATI did not come out anything any more. Better, whereas all its concurents major (3Dfx, NVIDIA, Matrox) announced (very in advance) here and there their next chips, the coyotes of ATI did not do anything of it. After the outputs of Voodoo then TNT, G200, Savage 3D, Banshee, everyone were said: " Good, ATI loupé notches it like S3 in its time ". Eh well not: ATI has delay, but ATI did not loupé at all notches it: they had in preparation RAGE 128, processor 2D/3D of third generation (like the TNT, Savage 3D...).

RAGE 128 will be declined in two versions: the RAGE 128 GL, intended for the workstations OpenGL and general public (plays 3D), and the RAGE 128 VR for OEMs.

Let us see a little which are the design features of this chip:

  • Graphics processor 128 bits
  • Support of the SDRAM and GDR SGRAM (Double Dated Spleen SGRAM)
  • Support optionel of an output TV and a unit of video capture
  • The techniques used by RAGE 128 are: the technology of returned super scalar (speed 2D and 3D), Twin architecture Hides (meeting of the masks textures and pixel from where increase in the bandwidth), Multitexturing on the way (following the example of Voodoo and TNT), and finally a very optimized management of the AGP.
  • Decoding MPEG-2 (thus DVD) in hardware, therefore with little use of the CPU
  • Optimization DirectX 6 and OpenGL
  • RAMDAC to 250 MHz. Perfect stability until in 1920x1280 in true color

The functionalities 3D of RAGE 128:

  • Complete alpha blending
  • Fog based on a table or a vector
  • Dynamic lights
  • Reflection
  • Shades
  • Bump mapping
  • LOD biaising
  • Morphing of texture
  • Z-buffer in 16, 24 or 32 bits
  • Returned into 16, 24 or 32 bits
  • Stencil plane

RAGE 128 will be equipped with 16 or 32 MB of memory (SDRAM with up to 143 MHz, GDR SGRAM up to 125 MHz for a bandwidth going up to 2 GB/s, equivalent to AGP 8x...).

Here the diagram of operation of RAGE 128:

rage128.gif (42850 octets)

Good on the level of the functionalities 3D, all goes well, it is " standard ". RAGE 128 does not have anything to envy its concurents, except perhaps with RIVETTED TNT, which can make him anisotropic filtering, better than the bi/tri-linear. But good, to take advantage of this filtering, it is necessary that the object is more or less detailed according to its dimension.

The technology of manufacture is obviously the 0.25 microns (little dissipation of heat, high frequencies). On the level of fill-misses theoretical (with memory to 125 MHz and a frequency of 100 MHz), one is still in the same category as the TNT: 200 MPix/s in single texturing, 100 MPix/s into dual texturing (RIVETTED it TNT, for the moment, has memory to 112 MHz and is given rhythm to 90 MHz, therefore 180 MPix/s in single texturing, 90 MPix/s into dual). In any case, RAGE 128 carries out as the TNT the multitexturing on the only one way, without loss of performance.

Now, I propose to you to return in the details, and it is there that the article becomes an exclusiveness.

Here photographs of screens illustrating the principal functionalities of chip, thus q' a photograph of Quake 2. They have warp taken with the following configuration: Pentium 2 450 MHz, 128 MB of SDRAM, XPert 128 (RAGE 128) 16 MB SDRAM, DD 10 GB, Windows 98 and Windows NT 4.0. The drivers used are the following: Beta bluid 15 pennies 95/98, Beta build 06 pennies NT. The photograph of Quake 2 was made under NT, so that it cannot have cheating on the level of the ICD OpenGL (under NT, not of wrapper Direct3D/OpenGL such that of G200 under 95/98). ATI thus has well already a OpenGL driver. For Quake 2, images in high resolution will follow (the card will go up in 1920x1200x32).

Quake 2:

driver.jpg (34736 octets)

quake2_img1s.jpg (31305 octets)

Good first image, not of comments, one sees well that Quake 2 charges all the possible OpenGL modules, that returned is in 32 bits, Z buffer of 24 bits.

Force is to notament note that quality is not far from the perfection, qu level of the mip-mapping of the ground, one does not observe " breaks ", all is done carefully. The colors they also are irreproachable (it is all the same of the 32 bits). The lights are well contrasted, neither too dark nor too clear as on certain chips. No the dithering either. In short, of the good job.

Rage Rawning Demo:

It is a demonstration developed by 3D Forges in order to show the possibilities of RAGE 128 (apparently there was a small problem with these images, they do not appear very beautiful). *** TRANSLATION ENDS HERE ***

Good it is traditional, one will not comment on on top. It is an animal (finally...) plating of textures.

That it is an example of multitexturing. This method already was implemented and functions with Direct X 6. Three operations are applied to the polygon:

  • Firstly, the affected plating of texture to the first engine of returned (rendering engine # 1)
  • Then the rendering engine #2 carries out Gloss map (plating of the one second texture offering this effect of varnish, of brightness)
  • Finally the reflection is calculated by the lightning engine (driving effects of lights) at the end of the chain.

All that being realized on the only one way.

Third example, Bump Mapping (here, DX6 emboss map), which is very à.la.mode these times.

bumpmaps.jpg (38328 octets)

This technique makes it possible to simulate an effect of relief on textures. Here is how RAGE 128 carries out it:

  • Plating of texture (rendering engine # 1)
  • Height Map (the second texture containing information of altitude, applied by the rendering engine # 2)
  • Application of the source of light on the first texture, according to the altitude provided by Height Map. This application is dealt with by the lightning engine.

Obviously, RAGE 128 carries out Bump Mapping on the only one way (to you had included/understood it, it is because each engine is used only once).

I now will show you why Single Pass Multitexturing (multitexturing on the way) such as one finds it in RAGE 128 or RIVETTED it TNT is better than Multi Pass Multitexturing (multitexturing on several ways) such as one finds it in G200, Savage 3D and Voodoo Banshee:

Here stages of Single Pass Multitexturing with a processor such RAGE 128 (provided with two rendering engines):

  • Reading textures 1
  • Reading textures 2
  • Reading Z-Buffer
  • Writing Z-Buffer
  • Writing pixel

Five stages and a maximum quality, without effects of dithering.

Let us see Multi Pass Multitexturing now (with only one rendering engine)

  • Reading textures 1
  • Reading Z-Buffer
  • Writing Z-Buffer
  • Writing pixel (if rendering 16 bits, dithering 16 bits)
  • Reading textures 2
  • Reading pixel (in 16 bits if display 16 bits)
  • Reading Z-Buffer
  • Writing pixel

Eight stages and a less quality because of the effect of dithering. It is clear that the future is in Single Pass Multitexturing (implemented in the plays OpenGL and Direct X 6). It is not thus in my opinion not viable to provide itself with a card of not implementing this function.

RAGE 128 thus has three engines: two of returned and for the effects of light. It is that which allows Single Pass Multitexturing, as well as Single Pass Bump Mapping.

I did not speak about the part 2D (one tends to forget it lately). Know that it will be doubtless on the level of the remainder, therefore very good (with performances appears étonantes under Windows NT).

I will finish with some precise details on the level of the performances. RAGE 128 will thus be placed concurent direct RIVETTED TNT on this level. But the Canadian company has of other goals: ATI hopes well to infiltrate the market of the professional cards 3D (it is besides for that that you can really count on a ICD OpenGL as of the output of the cards). I have figures, but I will not communicate them (in all manners, I will publish figures whom I will have carried out myself). I can say all the same to you that it is on the level hopes that you can place in RAGE 128.

Then, on the level of returned, you can discount something of as beautiful as the TNT, if not more: indeed, RAGE 128 carries out all its calculations in 32 bits, therefore the quality of a play or a scene 16 bits will be certainly better on this chip than on the TNT.

Here offers that ATI will propose:

XPERT 128: 16 MB of SDRAM (nonextensible), AGP2x, in the 1000 FTTC (without bundle).
RAGE MAGNUM: 32 MB of SDRAM (nonextensible), AGP2x, in the 1500 FTTC (without bundle), intended for the workstations of input of range.
RAGE FURY: 32 MB SDRAM (nonextensible), left TV, AGP2x, in the 1700 FTTC (without bundle).

All will décompresseront the MPEG-2 in hardware (the FURY will have at more one output TV). They will be compatible with Windows 95/98 and Windows NT4, then Linux later (when XFree implements chip). The cards will be initially with format AGP, then ATI will launch a version NCV at the beginning of 1999. RAGE 128 is thus announced like an excellent chip, complete (like the Rage Pro), but which has the possibility of eclipsing the concurence thanks to its quality (as well on the level performances as on the level quality of returned). All this, of course, under the condition which the Canadian company holds its promises (notament those of times, first cards being scheduled for December).

Guillaume Gates October 20, 1998

Updated on October 21, 1998