TBR Patent Infringement

Eldar

Newcomer
My understanding was that:

The PowerVR uses tile based rendering instead of immediate mode rendering, used traditionally by graphics renderers. This means the 3d image is constructed by first dividing the screen into a number of tiles and then processing each tile at a time. This means the renderer can pull in all the polygon information for a single tile and process it through to the final image before writing the tile out to the frame buffer and moving onto the next tile (Source Arm/Imagination Technology regarding MBX)

and that:

PowerVR MBX’s unique patent-protected technologies enable it to deliver class-leading performance from a very small core, and its sophisticated power management techniques exploit both module and register-level clock gating to ensure the lowest possible active and standby power dissipation (source Imagination Technology)

So have:

1. Falanx (Mali), Ati/Nvidia (rumoured to be using TBR on future handheld chips) etc taken out a yet to be announced license from Imagination Technology? Or

2. Said companies do not give a stuff about Imagination Technology patents and/or do not expect Imagination Technology to protect their patents (through the courts if necessary?) Or

3. Imagination Technology's patents do not cover the TBR used by other companies?

I always had the impression, based on everything I have read from Imagination Technology, that they had sufficient patents to effectively "own" TBR

I am confused and yes I know about Microsoft and Talisman or whatever it was called.

Eldar
 
IMG would be hard pressed to protect TBDR in general. I'm pretty certain deferred renderers existed before PowerVR, though I could easily be wrong.

The patents more likely deal with the intricacies of making a TBDR work with an IMR API like OpenGL or Direct3D. But that's not to say there aren't plenty of other ways to do it.

Perhaps Simon or Kristof would be able to do a better job explaining all this than I can....
 
And, by the way, TBDR is the correct term. Tile-based rendering is used by almost everybody these days. Deferred rendering is what's different about PowerVR's technique.
 
There are 2 different things related to deferred rendering


1. hidden surface removal before rendering
Rendering is done in 2 phases:
1. hidden surface removal phase
2. actual rendering / color calculations

In order to be able to do rendering on 2 phases, all triangles have to be stored.

Tile-based rendering(not really deferred rendering)
Screen is rendered smaller part("tile") at time.
all triangles that are inside a tile are rendered into tile, then the tile is copied into backbuffer and next tile is being operated.

This also requires that all triangles have to be stored
(so that they can be rendered into every tile at different time)


some intel extreme graphics cores do do tiling, but does not do HSR before rendering.

via/s3 deltachrome does do HSR before rendering, but does not do tiling.

powerVR does both, and combines them well.

HSR before rendering and tiling are good match for each others due

a) using either of those alone would has the negative side of storing the triangles.. but when you do both you still store them only once

b) HSR before rendering can be done with very simple/cheap Z-units, and one chip may contain huge number of these units. These units however require tons of bandwith, and memory bandwith is easily bottleneck. When doing tiling, the active framebuffer(tile buffer) can be so small that it can be SRAM with huge bandwith inside the chip.
 
Without any TBDR based cores (other than from PowerVR/IMG) released in silicon I'd suppose it would be hard to talk about patent infrigiments.

Falanx's Mali hasn't been released yet and neither Gigapixel's IP made it into any product in the past nor did up to now NVIDIA release any product based on the Gigapixel patents or a TBDR.
 
This also requires that all triangles have to be stored
(so that they can be rendered into every tile at different time)

Not on a PowerVR though; MBX utilizes so called small object culling, whereby if a pixel doesn't cross a triangle it gets rejected.

In order to be able to do rendering on 2 phases, all triangles have to be stored.

Not true either. See example above. Things are way more complicated than you describe them.
 
Chalnoth said:
Tile-based rendering is used by almost everybody these days.
We were calling it tile based rendering long before anyone ever spoke of alternate rasterization patterns ... and Ill keep calling it tile based rendering thank you very much.

As for what is patentable and what's not, bucket rendering predates PowerVR. That was essentially tiled rendering too.
 
Most graphics processors try to defer rendering into just visible surfaces to avoid the ineffectiveness of outputting overdrawn pixels, and most processors also try to divide the workload into manageably sized tiles to avoid the inefficiency of working across the bus into external memory. The difference in PowerVR solutions seems to have been that they implement these approaches closer down to the device level while conventional methods seem to handle it nearer to the application level.

A balance nearer to the device would appear to be more beneficial until it becomes too inflexible to keep pace with advancement.
 
Lazy8s said:
Most graphics processors try to defer rendering into just visible surfaces to avoid the ineffectiveness of outputting overdrawn pixels, and most processors also try to divide the workload into manageably sized tiles to avoid the inefficiency of working across the bus into external memory. The difference in PowerVR solutions seems to have been that they implement these approaches closer down to the device level while conventional methods seem to handle it nearer to the application level.

A balance nearer to the device would appear to be more beneficial until it becomes too inflexible to keep pace with advancement.

I think I'll keep that paragraph for reference; beautifully worded :)
 
PowerVR is protected by patents, but they don't own every patent associated with TBDR.

Talisman was the first ever tiler made by microsoft, but abandoned for reasons I am unaware of but I suspect it was a little ahead of its time.


What I do think might get dangerously close to IMGEC's patents is actually ATi's multi-chip load balancing method of splitting the screen into tiles and distributing them between rendering units.
 
Sort first parallel rendering is pretty ancient, Bitboys were also going to use something like it (sort middle).
 
What I do think might get dangerously close to IMGEC's patents is actually ATi's multi-chip load balancing method of splitting the screen into tiles and distributing them between rendering units.

Not only is MfA right, but I have the impression that the specific macro-/micro-tiling patent from IMG you're referring to, had as a primary target to handle the display list space more efficiently.

Techreport has an interesting article about XBox360's GPU. If it would come to thread allocation/handling PowerVR would most likely incorporate more Metagence IP for such a case. ATI isn't from what it sounds to me incorporating something revolutionary either, but it seems quite unique to me as a method.

MfA,

Was it your signature that stated that every interesting idea has been patented....twice? If yes it's a shame you aren't using it anymore ;)
 
MfA said:
Chalnoth said:
Tile-based rendering is used by almost everybody these days.
We were calling it tile based rendering long before anyone ever spoke of alternate rasterization patterns ... and Ill keep calling it tile based rendering thank you very much.

Yep, Pyramid3D chip back in '95 also used tiled rasterization order for better access patterns to dram pages, and is purely IMR.
 
They never talked about what they were doing, powervr did ... as did we on D3D.

Ailuros, I just said that ... would be a bit over the top to sig myself now wouldnt it :)
 
Back
Top