Architecture RIVA 128

riva 128 gpu



riva 128zx gpu

nvidia s nv1 chip had been designed fundamentally different type of rendering technology, called quadratic texture mapping, technique not supported direct3d. riva 128 instead designed accelerate direct3d utmost extent possible. built render within direct3d 5 , opengl api specifications. graphics accelerator consists of 3.5 million transistors built on 350 nm fabrication process , clocked @ 100 mhz. riva 128 has single pixel pipeline capable of 1 pixel per clock when sampling 1 texture. specified output pixels @ rate of 100 million per second , 25-pixel triangles @ 1.5 million per second. there 12 kib of on-chip memory used pixel , vertex caches. chip limited 16-bit (highcolor) pixel format when performing 3d acceleration , 16-bit z-buffer.

the 2d accelerator engine within riva 128 128 bits wide , operates @ 100 mhz. in fast , wide configuration, nvidia referred it, riva 128 performed admirably gui acceleration compared competitors. 32-bit hardware vesa-compliant svga/vga core implemented well. video acceleration aboard chip optimized mpeg-2 lacks full acceleration of standard. final picture output routed through integrated 206 mhz ramdac. riva 128 had advantage of being combination 2d/3d graphics chip, unlike voodoo graphics. meant computer did not require separate 2d card output outside of 3d applications. allowed 3d rendering within window. ability build system 1 graphics card, , still have feature-complete time, made riva 128 lower-cost high-performance solution.

nvidia equipped riva 128 4 mib of sgram, new memory technology time, clocked @ 100 mhz , connected graphics processor via 128-bit memory bus. provides memory bandwidth of 1.60 gigabytes per second. memory used in unified memory architecture shared whole ram pool both framebuffer , texture storage. main benefit of this, on split design such on voodoo graphics , voodoo², support 3d resolutions of 800×600 , 960×720, higher voodoo s 640×480.

riva 128 1 of agp 2x parts, giving more marketing headroom being on forefront of interface technology. graphics processor built around intel s agp specification targeting intel 440lx chipset pentium ii. nvidia designed riva 128 maximum memory capacity of 4 mib because, @ time, cost-optimal approach consumer 3d accelerator. case partly because of chip s capability store textures in off-screen system ram in both pci or agp configurations.

in 1998, nvidia released refreshed version called riva 128 zx. refreshed design of nv3 increased memory support 8 mib , increased ramdac frequency 250 mhz. these additions allowed riva 128 zx support higher resolutions , refresh rates. next major chip nvidia riva tnt.

image quality

quake ii on riva 128 (final drivers)

at time of riva 128 s release, 3dfx voodoo graphics had firmly established 3d hardware benchmark against newcomers compared. voodoo first 3d game accelerator offer exceptional performance , quality in emerging pc 3d gaming. riva 128 scorned lower quality rendering (compared voodoo) , rendering errors.

with initial drivers, riva 128 used per-polygon mipmapping instead of higher-quality, more demanding, per-pixel variety. caused different texture detail levels pop place player moved through game , approached each polygon, instead of allowing seamless, gradual per-pixel transition. nvidia released drivers allowed per-pixel mode. issue card s texturing use of automated mipmap generation. while improves visual quality , performance in games without mipmaps, caused unforeseen problems because forced games render in way not programmed for.

nv3 s bilinear filtering sharper of 3dfx voodoo graphics. but, while didn t blur textures as voodoo, did instead add light noise textures, because of lower-fidelity filtering algorithm. there problems noticeable seams between polygons.

while initial drivers did present these image-quality problems, later drivers offered image quality arguably matching of voodoo graphics. in addition, because riva 128 can render @ resolutions higher 640×480, card can offer quality superior of voodoo graphics, shown in above quake ii screenshot. final drivers released riva 128 support per-pixel mipmapping, full-scene |anti-aliasing (supersampling), , number of options fine-tune features in order optimize quality , performance.