By: David (glowbin.delete@this.me.com), August 1, 2016 4:09 pm
Room: Moderated Discussions
Joel H (joel.hruska.delete@this.gmail.com) on August 1, 2016 1:57 pm wrote:
> David (glowbin.delete@this.me.com) on August 1, 2016 1:22 pm wrote:
> > If the claim is that Nvidia introduced these changes with Maxwell, wouldn't it
> > be wise to show this test running on Kepler and/or Fermi GPUs as well? All that's
> > been shown is that Maxwell and Pascal rasterize differently than AMD.
> >
> > Something is missing...
>
> I believe David states that we know NV used IMR prior to Maxwell. That's not in question.
You're right, that is not in question. Nor is it my point. In the first comment above David says:
> Starting with the Maxwell and Pascal architectures, Nvidia high-performance GPUs use tile-based
> immediate-mode rasterizers, instead of conventional full-screen immediate-mode rasterizers.
Which can be summarized as:
- Maxwell and Pascal: tile-based immediate-mode rasterizing
- pre-Maxwell: conventional immediate mode (non-tile based) rasterizing
Thus one would expect pre-Maxwell cards to exhibit different behaviors within the triangle test. I'm suggesting this be added to the video. I'd run the software myself if I still had my Kepler GPU.
> David (glowbin.delete@this.me.com) on August 1, 2016 1:22 pm wrote:
> > If the claim is that Nvidia introduced these changes with Maxwell, wouldn't it
> > be wise to show this test running on Kepler and/or Fermi GPUs as well? All that's
> > been shown is that Maxwell and Pascal rasterize differently than AMD.
> >
> > Something is missing...
>
> I believe David states that we know NV used IMR prior to Maxwell. That's not in question.
You're right, that is not in question. Nor is it my point. In the first comment above David says:
> Starting with the Maxwell and Pascal architectures, Nvidia high-performance GPUs use tile-based
> immediate-mode rasterizers, instead of conventional full-screen immediate-mode rasterizers.
Which can be summarized as:
- Maxwell and Pascal: tile-based immediate-mode rasterizing
- pre-Maxwell: conventional immediate mode (non-tile based) rasterizing
Thus one would expect pre-Maxwell cards to exhibit different behaviors within the triangle test. I'm suggesting this be added to the video. I'd run the software myself if I still had my Kepler GPU.
