1st International Symposium on CELL Computing
- Thread starter nAo
- Start date
Jawed
Legend
Deferred pixel shading on Cell prolly deserves its own thread, but I suppose we'll just have to wait.
I dare say you could have a lot of fun constructing a tile based deferred renderer in software on Cell, with a deferred lighting engine used to shade each pixel in a tile.
It seems to me that you could hide texturing latency quite nicely. When you know in advance per-pixel materials, effects plus anisotropy you should be able to construct a really efficient texture fetch stream.
Jawed
I dare say you could have a lot of fun constructing a tile based deferred renderer in software on Cell, with a deferred lighting engine used to shade each pixel in a tile.
It seems to me that you could hide texturing latency quite nicely. When you know in advance per-pixel materials, effects plus anisotropy you should be able to construct a really efficient texture fetch stream.
Jawed
Vysez said:
A game developer?
Where would RSX fit in this? I presume you're not going to idle it..
edit - just had a look, and it's a SCEA presentation, so I guess that's at least one game developer..
And no, they don't idle RSX at all. I'm very unclear on the specifics of this though What kind of pixel shading is Cell doing? It talks about returning 'shadow textures' - so shadowing/lighting ala this article by DeanoC..?
Last edited by a moderator:
Jawed
Legend
It's worth noting this is by Heirich:
http://alumnus.caltech.edu/~heirich/eghw05_1003.pdf
Which, to me, suits the idea of the kind of "re-ordering" that I was describing earlier. In fact I thought of this paper when I made that comment, jsut hadn't bothered to rummage.
Jawed
http://alumnus.caltech.edu/~heirich/eghw05_1003.pdf
Which, to me, suits the idea of the kind of "re-ordering" that I was describing earlier. In fact I thought of this paper when I made that comment, jsut hadn't bothered to rummage.
Jawed
I'm a little confused. Is this deferred shading in the sense of working out visible fragments for subsequent shading on the GPU, or is it pixel shading on Cell with certain stages deferred (e.g. rasterisation, most obviously), which will occur on the GPU? I'm guessing it's the latter given Jawed's first reply, but just like to be very clear..
Last edited by a moderator:
mckmas8808
Legend
Vysez said:
Don't tell us the company, but can you or nAo explain how this will make games look better? Like what does Mapping Deferred Pixel Shaders onto the Cell Architecture actually mean?
Thanks.
I had like a two-week lead on this topic, but whatever.
http://www.beyond3d.com/forum/showthread.php?t=31564
http://www.beyond3d.com/forum/showthread.php?t=31564
mckmas8808
Legend
xbdestroya said:I had like a two-week lead on this topic, but whatever.
http://www.beyond3d.com/forum/showthread.php?t=31564
Can you answer what Mapping Deferred Pixel Shaders onto the Cell Architecture actually means?
From the site's Goals pagepc999 said:
Specific topics of interest, that will be discussed at CCS-06 include:
- Programming models unifying local, in-process, and remote communication.
- New languages, language constructs and programming models for leveraging the CELL.
- New libraries, and tools for the CELL.
- Design techniques for factoring multicore algorithms.
- Numerical methods for the CELL.
- SIMD, software managed cache and thread programming optimizations.
- Intrinsic based AltiVec / SIMD programming techniques.
- Experience reports in targeting game / high quality rendering engines to the CELL.
- Applications perspective for leveraging next generation processor architectures.
- Experience reports on building CELL based clusters and other peer-to-peer distributed systems.
JF_Aidan_Pryde
Regular
mckmas8808 said:
Deferred rendering uses a different order of rendering to 'immediate' rendering. With immediate rendering, many pixels are computed, then each is tested for visibility. Only the visible ones (those that pass Z and stencil tests) are displayed. With deferred rendering, the visibility of pixels is determined first, then only the visible pixels are calculated using pixel shaders. So the key performance benefit is that no pixel shading work is wasted on occluded pixels.
Because DX9 GPU's architecture is organised for immediate rendering, you have to jump through loops to make it do deferred rendering. With Cell, it's totally programmable, so it can perform deferred rendering more naturally.
I'm not sure about performance though. Anyone care to speculate?
First guess, it can't be too hot. Otherwise why is RSX there and not a Cell renderer? From the off it was speculated that Cell would be great at vertex work, not so great at pixel work. If this is good at pixel work, the idea of a purely Cell GPU would have been feasible. Though maybe it was working but he tools weren't there and that's the sole point to nVidia?JF_Aidan_Pryde said:
I'm guessing this is something that isn't capable of producing whole games on it's own, but can work well with RSX. Also, it may have application in alternative devices like a mobile platform, with Cell doing simpler shading work. That is an area for Cell research if it's to broaden it's appeal.
There's my tu'penny-ha'penny worth!
My knowledge of this isn't so great at all, but here goes..
I'm presuming Cell and RSX together do the initial "generate your per-pixel attributes" and then RSX handles subsequent passes using those? That's where a second Cell chip might have fallen down?
My question is, is the shading being done here on Cell symmetric with the shading on RSX? Or are you doing some specific steps on Cell and then something else entirely on RSX?
Bah, I'm probably not making any sense It'd be nice if someone could outline the steps involved and the possible delegation of work on Cell.
I'm presuming Cell and RSX together do the initial "generate your per-pixel attributes" and then RSX handles subsequent passes using those? That's where a second Cell chip might have fallen down?
My question is, is the shading being done here on Cell symmetric with the shading on RSX? Or are you doing some specific steps on Cell and then something else entirely on RSX?
Bah, I'm probably not making any sense
It'd be nice if someone could outline the steps involved and the possible delegation of work on Cell.mckmas8808
Legend
nAo said:
nAo we are hurting man. Please come in here and answer some questions. Can you at least try to fill in some of the gaps?
Jawed
Legend
That's where the presentation/paper comes in. I expect it'll be publicTitanio said:Bah, I'm probably not making any sense
Ghost Recon: Advanced Warfighter uses a deferred lighting graphics engine for the PC version of the game. Dunno what the XB360 version uses.
I don't know why you'd use Cell to do any of this, to be honest. I don't mind believing PS3 games will use deferred lighting (i.e. like GRAW, with the graphics done by RSX) but it seems to me the only way deferred pixel shading on Cell is going to be useful is when you have a Cell-only computing environment, i.e. not PS3.
Jawed
Jawed said:
If the paper's claim is true (i.e. Cell+GPU>GPU for this), presumably it'd be useful if using deferred shading and bound where Cell could offer relief, or perhaps if you wish to use more complicated lighting evaluation (though that might depend on the extent of Cell's involvement).
Jawed said:
Jawed
Well, the abstract too:
I guess it's a question of whether they follow up on that 'indication' and provide numbers on GPU, Cell (if you could bench Cell on its own..again, a question of how far Cell is involved) and GPU+Cell. But I presume/hope they will..(!) I'm guessing that sentence isn't coming out of nowhere, and they've tested that.
Last edited by a moderator:
Similar threads
