A look into the history of Half Life 2's development, as part of the Half Life 2 20th Anniversary Update: https://half-life.com/en/halflife2/20thE3 2003: The...
The industry was so different back then, Unreal Tournament 2004 still shipped with a software renderer.
The industry was so different back then, Michael Abrash documented that renderer’s development in Dr. Dobbs, an actual ink-on-paper journal.
Here’s one thing that hasn’t changed: Intel. This renderer, Pixomatic, was all hand-optimized assembly, from the guy John Carmack hired to outclass him. At one point he realized one instruction in a very tight loop was redundant. Removing it made the loop slower. Which is, in technical terms, some bullshit. Doing less should not take more time. It wasn’t from alignment or cache or pipelines or any sensible cause. Abrash called in favors so he could study the actual traces of the Pentium 4, because he just had to know what the fuck was happening under the hood - and Intel made him sign a stack of NDAs, so we the public will never find out.
The industry was so different back then, Unreal Tournament 2004 still shipped with a software renderer.
The industry was so different back then, Michael Abrash documented that renderer’s development in Dr. Dobbs, an actual ink-on-paper journal.
Here’s one thing that hasn’t changed: Intel. This renderer, Pixomatic, was all hand-optimized assembly, from the guy John Carmack hired to outclass him. At one point he realized one instruction in a very tight loop was redundant. Removing it made the loop slower. Which is, in technical terms, some bullshit. Doing less should not take more time. It wasn’t from alignment or cache or pipelines or any sensible cause. Abrash called in favors so he could study the actual traces of the Pentium 4, because he just had to know what the fuck was happening under the hood - and Intel made him sign a stack of NDAs, so we the public will never find out.