nV now owns Tile-based rendering technology (?)

[mrsabidji]

I just spotted that on W2S.

In regards to the tile based rendering, Nvidia now ownes this technology (happened during the Gigapixel -- 3dfx IP purchase). So just think of what might be possible, a GeForce 6800 Ultra with tile-based rendering? That would blow any type of competition out of the way by far. But of course all of this might not be as easy to integrate as you might think/hope.

I was sondering if some of you guys had heard of it and if it was BS and what you thought about it and all.

mrsabidji

 

[Nebuchadnezzar]

I'ts been known since quite a while... but that doesn't mean Nvidia will use it in the near future.

 

[mrsabidji]

I had never heard of it, personally . I thought it sounded strange. I don't know much about GPU architecture but I guess TBR isn't something you can just 'add' to the nv4x GPUs. And if they had it, we would all know it by now.
But what about nv50 ?

mrsabidji

 

[WaltC]

Nothing here, IIRC, if we can believe the several statements Kirk made after nVidia's buy out of 3dfx in order to bury the patent suit ongoing at the time between 3dfx and nVidia. Kirk's answer to that question (the several times it was asked of him) at the time was that nVidia had no plans to ever use any of of the GP tech it had acquired in the 3dfx purchase.

 

[mrsabidji]

Oh, ok. But does the fact that they own the technology mean that no-one else can use TBR technology ? And if it does, does it affect TBDR tech as well ?

mrsabidji

 

[Heathen]

Oh, ok. But does the fact that they own the technology mean that no-one else can use TBR technology ? And if it does, does it affect TBDR tech as well ?

Nope, lots of people have TBR technology (Imagination are the most famous), just like anything in this world there are lots of ways of doing the same thing. Just means that nvidia will never use use the GigaPixel based TBR tech they acquired. TBR is just one element of TBDR.

 

[WaltC]

Oh, ok. But does the fact that they own the technology mean that no-one else can use TBR technology ? And if it does, does it affect TBDR tech as well ?

mrsabidji

No, it only means that nobody else can use any of the TBR implementations covered under the GP patents (whatever they were) without talking to nVidia first. I have no idea as to what GP's patents covered. I suspect most of them have been rendered moot by the progress in other technical directions in the 3d-chip sector since that time.

 

[mrsabidji]

Phew, what a relief !
So I take it nv's never gonna use this 'outdated' technology at all. And since it doesn't prevent anyone from making a TDBR GPU...
Now, the article was mostly dealing about STM to stop working together with PVR on the Kyro3. Does it mean Kyro3 based cards will never see the light of day ?

mrsabidji

 

[loekf2]

Phew, what a relief !
So I take it nv's never gonna use this 'outdated' technology at all. And since it doesn't prevent anyone from making a TDBR GPU...
Now, the article was mostly dealing about STM to stop working together with PVR on the Kyro3. Does it mean Kyro3 based cards will never see the light of day ?

mrsabidji

Forget about Kyro3, I think was a STMicro product name. In the past few years, IMG has slowly conquered the handheld graphics market (I think both TI and Intel licensed the MBX core, which is nothing more than a simplified tile-based 3D graphics core). Rumours tell there's still PowerVR series 5, targetting the PC space, but I would be amazed if IMG can compete with ATI and Nvidia. It's not just technology, but also marketing... After the 3DMark craziness, the FPS hype, we now get to who can render the sexiest polygon-babe (see Nalu vs. Ruby).

 

[mrsabidji]

The sexiest so far is still Mufu's Diginalu, in my opinion.
Concerning Series5, I know nothing about its specs, but if IMG could make a card based on it which could compare to the 6800 (non-ultra, what's its name again?) performance-wise... well, I think it would sell quite well. Or is Series5 that far from a final product ?

mrsabidji

 

[loekf2]

The sexiest so far is still Mufu's Diginalu, in my opinion.
Concerning Series5, I know nothing about its specs, but if IMG could make a card based on it which could compare to the 6800 (non-ultra, what's its name again?) performance-wise... well, I think it would sell quite well. Or is Series5 that far from a final product ?

mrsabidji

BTW... are you the same person posting here ?

http://www.datafuse.net/page.php?news=147

This is ancient stuff ;-)

Already two years ago, STM pulled out of the GFX business.

Further, Nvidia doesn't own TBR just because they acquired GigaPixel and 3dFX. IMG owns a whole lot of patents on the concept of TBR. TBR main purpose is the bandwidth problem to your framebuffer. Some people solve this by massive amounts of embedded DRAM (see BitBoys), others (take Nvidia or ATI) by doing smart compression techniques, wide busses (256 bits) and new DRAM technology (GDDR3).

IMO TBR is ideal for the mobile/embedded market. Here you want to put as many functions in big SOCs (saves costs). Many functions requiring access to the same memory leads to bandwidth issues. Graphics do require a lot of bandwidth (next to video processing of course).[/url]

 

[mrsabidji]

BTW... are you the same person posting here ?

http://www.datafuse.net/page.php?news=147

This is ancient stuff ;-)

I'm not sure exactly what you mean, but I'll try to answer anyway. I'm not the guy who posted the article at datafuse, but I'm the same mrsabidji since the beginning of this thread.
I really don't know much about TBDR and I needed some enlightenment, so maybe you were just a little confused by my newbie questions... Anyway, I thanks for the infos.

mrsabidji

 

[Ailuros]

Accelerators have had tiled (or chunk based) memory layouts since the Voodoos for the record. In that sense NVIDIA and all others are using tiles for a long time now, they just don't defer the rendering or to be more accurate since the advent of early-Z, they do not defer the majority of the rendering process.

Yes NVIDIA has indeed gotten Gigapixel's patents when they acquired 3dfx and no they haven't built so far any deferred renderer and most likely won't ever do so either. That doesn't mean of course that the technology portofolio from GP was entirely useless to them. I recall them having quite some interesting ideas/patents on specific caching techniques back then which NV might have put to good use in former products (if then since NV25).

Some people solve this by massive amounts of embedded DRAM (see BitBoys), others (take Nvidia or ATI) by doing smart compression techniques, wide busses (256 bits) and new DRAM technology (GDDR3).

There's more to it and that's what the bandwidth saving algorithms like early Z, fast Z clear, hierarchical Z, colour compression etc etc are for. Having a high amount of bandwidth will not save your day if don't incorporate sophisticated bandwidth saving algorithms into your pipeline.

Isn't it better to have a smaller bucket that gets emptied constantly so that it'll never fill up, than having one large bucket that'll get you into trouble when it fills up?

IMO TBR is ideal for the mobile/embedded market. Here you want to put as many functions in big SOCs (saves costs). Many functions requiring access to the same memory leads to bandwidth issues. Graphics do require a lot of bandwidth (next to video processing of course).

Granted with UMA architectures the advantages should be more straightforward, yet that doesn't mean that bandwidth isn't a considerable headache in all other systems too.

 

[Megadrive1988]

I think it's been known that Nvidia owns tile-based rendering technology ever since they purchased 3Dfx assests, which included GigaPixel. when was that, late 2000 ?