• Join Us |
  • |
  • Sign in with:

ATI's Next-Gen

by Steve Gibson, Feb 28, 2000 6:41pm PST
Related Topics – ATI

Just when we got a hold of some more specs for the NV15, some specs for the next-gen ATI 3D integrated board have been leaked out. ReactorCritical has gotten a hold of some info. This is going to be built into motherboards and sounds pretty impressive:

Standart PC-133/100/66 SDRAM up to 1GB First integrated North Bridge with hardware-accelerated geometry T&L 12,5 polygons per second 4 pixels/clock > 330M textured pixls per second Revolutionary new parallel rendering architecture for state of the art pixell fill rates Industry leading 8x Virtual AGP graphics performance and functionality 4MB to 64MB of main memory configurable for display memory




Comments

45 Threads | 45 Comments

  • As people here have said, this particular ATI chipset (note - it\'s not their next big project, that\'s still unrevealed) is just for hardwiring on to the motherboard. It\'s actually fairly beneficial in that it really raises the standard for basic 3D. Look at it this way: right now, your best hope in built-in video is a Voodoo 3, and that\'s probably the 2000 model. If you don\'t wind up with that, then you\'re probably using a Rage Pro! When you can assume that even the sub-$1000 users can have T&L acceleration, that changes how you design your games.


  • Well, I don\'t know about 12,5 polygons a second,
    but then again, looking at the fillrate, that seems a bit iffy.

    No one may have commented on this. But, > 330million textured
    pixels per second at 4 pixels a clock puts it at a very low clockspeed.
    It doesn\'t say dual textured even.

    However, they say they have a parallel rending architecture. This
    could be similar to that of the Maxx, with alternate frames, but
    with built in T&L and clipping, I would say that the T&L chip
    could probably have multiple clip zones, and then send them to two
    screen zones in segmented memory. This would allow
    multiple chips to render multiple triangles at once with no
    performance hit. It would require some pipelining, but it would
    be doable. You would have to put the clipping plane between
    the two buffers in the correct place for performance though.

    For example you wouldn\'t want a mid screen clip plane from side
    to side in a landscape, because more polygons would go
    to the bottom zone (you could adapt the plane position for optimal
    performance, but it wouldn\'t allow the segmented display
    buffer).