Yamauchi on the PS3: "...beginning of a new world..."

[MBDF]

GPU SHOPS are like GPU FLOPS. You can trust every GPU manufacturer to keep stretching the definition to get bigger numbers than their rivals.

Well at least I went with Controversial 112 shops instead of the Controversial 136 shops

 

[nAo]

What exactly is the standard defenition... if there is one

As I said before, there isn't, even over a single GPU architecture (ATI sometime calls a shader op a vec4 operation and sometime calls a shader op a scalar operation!).

Btw, how is it working with the 7800 GTX? *too many questions*

What?

 

[Powderkeg]

I've heard this said before (most likely by you:smile What exactly is the standard defenition... if there is one

What more is there to it?

I am always eager to learn

Btw, how is it working with the 7800 GTX? *too many questions*

Basically, you will never run into a situation where 100% of the shaders are being used to anywhere close to maximum efficiency. It's comparing theoretical maximums that can never be achieved.

And in this case, you are comparing completely different architectures that work at completely different efficiency levels, and totally ignoring about a million other factors that would effect performance.

With consoles, there really is no way to do a proper comparison on a spec-level, and using PCs as a base is even worse since PCs lack the hardware-specific coding that console software normally achieves.

Really the only thing to do is wait until at least the 3rd year of the console and look at it's 3rd and 4th generation exclusives, and compare those. That's the most effective comparison, because then you can see precisely what the hardware is really capable of. Before that, there really is no way to fairly compare cross-platform abilities.

 

[Powderkeg]

Why don't you try talking about something you know something about?

Antialiasing is NOT a postprocess and can not be performed as one, unless you count simple downsampling.

I see 4X MSAA, Stencil and Z Test, and Alpha Blending all taking place in the EDRAM, not the GPU.

So, I guess you should tell ATI that they are wrong, and they don't know how their chip works, eh?

Or maybe you should take your own advice, and remain silent.

 

[MBDF]

As I said before, there isn't, even over a single GPU architecture (ATI sometime calls a shader op a vec4 operation and sometime calls a shader op a scalar operation!).?

I see I have more to learn... so it's all just marketing eh? That's a shame... something should be done about that.

What?

Well, it was none of my business, but I've recently heard that your with Ninja theory, and I was hoping to get an understanding of what it was like to be optimizing for the RSX's little brother, seeing as far as I know... that much isn't under NDA.

 

[Laa-Yosh]

Or maybe you should take your own advice, and remain silent.

I suggest that you
1. read up on how a GPU generally works
2. what antialiasing means and how it's implemented
3. stop offending people who are already beyond 1. and 2. and are right in what they say, unlike you.

 

[nAo]

IWell, it was none of my business, but I've recently heard that your with Ninja theory, and I was hoping to get an understanding of what it was like to be optimizing for the RSX's little brother, seeing as far as I know... that much isn't under NDA.

[SIZE=-1]"Premature optimization is the root of all evil." -- Donald Knuth[/SIZE]

 

[Powderkeg]

errr
http://www.beyond3d.com/articles/xenos/index.php?p=04

the xenos while has two cores on the package is very diffrent from two sperate chips . Its better to think of it as one chip split into 2

The question is where is the ps3 going to find 26 - 134gb/s a second of bandwidth . Its either going to eat up the texture ram or its going to eat up the xdr ram bandwidth just for 4x fsaa .

The PS3 will use the same lossless compression that PC's use. You know, the first part of the paragraph that you conveniently edited out, that's what the PS3 will do.

Right but the edram is 10 megs so the tiles are in 10 meg chunks . What is the ram for the spu 256kbs ? or is it less . That is alot of tiles to stream in . It iwll be a hell of alot more than 3 titles and swaping them back and forth is going to eat up bandwidth. Then of course you have to send the finished product back to the rsx to be displayed.

The PS3 would require the same amount of RAM as a completed frame on the 360 to do the same total job, and RSX does not stream tiles.

And this bandwidth isn't coming from the GDDR3, it's coming from the XDR, which will likely have plenty to spare.

I'm sure the ps3 can do fsaa and hdr i just don't think it will be worth the trade off in rendering performance. However that is if the rsx can not do fsaa + hdr

Just a note. This "trade off" is already being done on the PC, and the result is improved performance. Just something to consider.

 

[Shifty Geezer]

I see I have more to learn... so it's all just marketing eh? That's a shame... something should be done about that.

Like all these things, people need to measurements to evaluate various properties, quantities and metrics, but without standards they can't be used for fair comparison. Like prior to Elizabethan regulation a pound of butter in one county of England would have been different to a pound of butter in another county, but in those counties half a pound was half the weight of one pound. That's why the measurement was needed though without being standard it couldn't be used to fairly compare different regions.

The only real applicable comparison for these consoles is well constructed benchmarks, which don't exist and never will. Fair benchmarks are rare for any industry AFAIK. Motoring has a few such as average fuel consumption is taken at given speeds etc. The next best console comparison is seeing what the average differences are between titles some years down the line, which is scientific but is at least based entirely on realworld performance. Trying to predict performance from paper specs is nigh on impossible.

 

[Powderkeg]

I suggest that you
1. read up on how a GPU generally works
2. what antialiasing means and how it's implemented
3. stop offending people who are already beyond 1. and 2. and are right in what they say, unlike you.

OK smart boy.

Tell me, precisely where is the hardware logic that performs the AA algorithms located on Xenos?

eDRAM or the GPU?

 

[Laa-Yosh]

And just to make it a bit easier for you: in Xenos, several components of a GPU's pixel pipeline have been moved onto the EDRAM die, so that they can access the framebuffer without a bandwith limit. These parts are on the GPU die in RSX.

 

[Shifty Geezer]

I suggest that you
1. read up on how a GPU generally works
2. what antialiasing means and how it's implemented

Isn't this diagram rather misleading...

This seems to be where Powderkeg's confusion is coming from. The AA requires the subsamples to be rendered on the GPU proper, and the eDRAM only handles the blending, but as it's written there it suggests the eDRAM adds the AA without the GPU proper being involved. Of course one single diagram should eb used to understand the entire intricacies of a render process!

 

[Laa-Yosh]

OK smart boy.

Stop that, now.

Tell me, precisely where is the hardware logic that performs the AA algorithms located on Xenos?
eDRAM or the GPU?

The Xenos is a unique architecture. The graphics pipeline has been split into two parts, because there's no current manufacturing technology to build a single, large GPU die that also has EDRAM.
Phisically the EDRAM is a separate part. But for all other purposes, it is a part of the GPU.

 

[Laa-Yosh]

Isn't this diagram rather misleading...

It is, but as long as there's a long article on Xenos right here on B3D, it's no excuse. And especially no excuse for his behaviour.

 

[Powderkeg]

And just to make it a bit easier for you: in Xenos, several components of a GPU's pixel pipeline have been moved onto the EDRAM die, so that they can access the framebuffer without a bandwith limit. These parts are on the GPU die in RSX.

But those parts can be ignored in RSX and applied by Cell instead.

If you wanted to, you could skip RSX entirely, and render the entire frame straight from Cell.

 

[Powderkeg]

The Xenos is a unique architecture. The graphics pipeline has been split into two parts, because there's no current manufacturing technology to build a single, large GPU die that also has EDRAM.
Phisically the EDRAM is a separate part. But for all other purposes, it is a part of the GPU.

And RSX/Cell is a unique architecture. The graphics pipelines are made in such a way that either, or both processors are capable of rendering all, or part of the frame.

 

[Shifty Geezer]

It is, but as long as there's a long article on Xenos right here on B3D, it's no excuse. And especially no excuse for his behaviour.

I agree.

 

[Laa-Yosh]

But those parts can be ignored in RSX and applied by Cell instead.

If you wanted to, you could skip RSX entirely, and render the entire frame straight from Cell.

The problem is that RSX's blending units are more suited for this task; on Cell, you'd have to use the SPEs which only work with 32 bit floats. It's overkill for the job, and it taks a massive amount of bandwith to work with a 128 bit framebuffer. As I've said, it's been discussed on B3D before, with PS3 developers, and the conclusion was that it's probably not possible in practice.

 

[Powderkeg]

This seems to be where Powderkeg's confusion is coming from. The AA requires the subsamples to be rendered on the GPU proper, and the eDRAM only handles the blending, but as it's written there it suggests the eDRAM adds the AA without the GPU proper being involved. Of course one single diagram should eb used to understand the entire intricacies of a render process!

And again, RSX can read/write directly to XDR, and Cell can perform the same processes as the logic in the edram.

What's the difference?

 

[Laa-Yosh]

And RSX/Cell is a unique architecture. The graphics pipelines are made in such a way that either, or both processors are capable of rendering all, or part of the frame.

Please, you clearly have no idea about the technological background, you just repeat Sony PR talk here...