VR-Zone had the chance to meet up with ATi’s President & CEO, David E. Orton and he shared with the press his views on PCI Express, chipsets and the roadmap of ATi. More details inside.

VR-Zone has an opportunity to meet up with the ATi’s President & CEO David E.
Orton over at a Computex and here are some information that he has shared with
the members of the press :

PCI Express Transition

  • OEM to move to PCI Express faster, offering from top to bottom solutions
  • About 70% of the OEM channel will move to PCI Express by end of this year (2004)
  • Channel will be slower and is expected to reach 50% by end of the year (2004)

Chipsets For AMD

  • There will be chipset offerings for AMD ranging from desktops to notebooks by end
    of this year (2004)
  • ATi will continue to work with ULi for its Southbridges

Die Size Increment

  • The die size of the GPU has been increasing at a faster pace in these recent
  • There are several factors to affect power consumption : Die Size, ASIC
    layout, Architecture, Process Technology, Package Design, Thermal & Noise
  • ATi to adopt technologies from the notebook GPUs to reduce power for the
    desktop GPUs


  • No plans to follow MXM standard which is set by NVIDIA
  • To come out with Advanced Express I/O Module which is a proven solution
  • M28 meant for DTR which has broader power range
  • Supports multimedia and has a range of products coming up

90nm Process Technology

  • To surface in 2005
  • Continue to work with both UMC and TSMC on the next generation process

Native vs Bridge PCI Express

  • ATi uses native PCI Express interface while NVIDIA uses bridge (HSI). The benefits of each solution will be clearer within the next 4 weeks.

Enhancements For Next Generation GPU

  • More parallel architecture
  • More pipelines
  • Finer process technology
  • Improved memory interface
  • Higher core and memory clock
  • GDDR3 to dominate the memory technology over the next two years and has
    much headroom ahead
  • GDDR3 memory clock to reach 1Ghz in the near future and offer 1.5x to 2x
    more memory bandwidth
  • Work on higher order surfaces