By: David Kanter (dkanter.delete@this.realworldtech.com), August 2, 2016 8:45 am
Room: Moderated Discussions
Peter McGuinness (peter.mcguinness.delete@this.gobrach.com) on August 2, 2016 8:43 am wrote:
> David Kanter (dkanter.delete@this.realworldtech.com) on August 2, 2016 8:30 am wrote:
> > Peter McGuinness (peter.mcguinness.delete@this.gobrach.com) on August 2, 2016 8:11 am wrote:
> > > David,
> > > Unfortunately, your assumptions about TBR are in error. This test actually demonstrates conclusively
> > > that these GPUs are NOT tile based; the previous commenter who asserted that the apparent tile organisation
> > > is due to optimised cache-to-memory transfer patterns is much closer to the truth.
> > >
> > > The fundamental characteristic of a tile based renderer, deferred or not, is that each tile is rendered
> > > to completion in on-chip memory and only after all visible objects have been rendered is the final result
> > > written out to display memory. This means that except in some very rare circumstances you will never
> > > see a partially rendered tile so the fact that occluded triangles are visible in the video is conclusive
> > > evidence that nvidia is still using a classic brute force immediate mode approach.
> >
> > Peter, if you read my article you will see I never claimed Nvidia was doing tiled *rendering*,
> > they are doing tiled rasterizing. Of course their approach is totally different than IMG.
> >
> > But I suspect this captures most of the benefits of TBDR.
> >
> > David
>
> There is no distinction. The only part of a tile based renderer which is actually tile-based is
> the rasterisation portion; tiles do not exist except in screen space. Immediate mode renderers have
> been using tile-oriented memory access techniques since GPUs were invented and the batching of square
> regions of pixels for shading (which is what the small regions are) is also very common.
The distinction here is that they are keeping the tile data in on-chip buffers. Normally, that would be streamed out to DRAM.
While this is probably common for mobile, its not for desktop. That's what makes it interesting! We are seeing mobile techniques moving upwards.
You are welcome to call it what you wish, but I chose the term that seemed most appropriate to me. How would you distinguish between a TBDR and a TBR in your mind?
David
> David Kanter (dkanter.delete@this.realworldtech.com) on August 2, 2016 8:30 am wrote:
> > Peter McGuinness (peter.mcguinness.delete@this.gobrach.com) on August 2, 2016 8:11 am wrote:
> > > David,
> > > Unfortunately, your assumptions about TBR are in error. This test actually demonstrates conclusively
> > > that these GPUs are NOT tile based; the previous commenter who asserted that the apparent tile organisation
> > > is due to optimised cache-to-memory transfer patterns is much closer to the truth.
> > >
> > > The fundamental characteristic of a tile based renderer, deferred or not, is that each tile is rendered
> > > to completion in on-chip memory and only after all visible objects have been rendered is the final result
> > > written out to display memory. This means that except in some very rare circumstances you will never
> > > see a partially rendered tile so the fact that occluded triangles are visible in the video is conclusive
> > > evidence that nvidia is still using a classic brute force immediate mode approach.
> >
> > Peter, if you read my article you will see I never claimed Nvidia was doing tiled *rendering*,
> > they are doing tiled rasterizing. Of course their approach is totally different than IMG.
> >
> > But I suspect this captures most of the benefits of TBDR.
> >
> > David
>
> There is no distinction. The only part of a tile based renderer which is actually tile-based is
> the rasterisation portion; tiles do not exist except in screen space. Immediate mode renderers have
> been using tile-oriented memory access techniques since GPUs were invented and the batching of square
> regions of pixels for shading (which is what the small regions are) is also very common.
The distinction here is that they are keeping the tile data in on-chip buffers. Normally, that would be streamed out to DRAM.
While this is probably common for mobile, its not for desktop. That's what makes it interesting! We are seeing mobile techniques moving upwards.
You are welcome to call it what you wish, but I chose the term that seemed most appropriate to me. How would you distinguish between a TBDR and a TBR in your mind?
David