Tell me about polygoncount

[London Geezer]

yeah...?

and i want a ferrari, a villa in the Bahamas, be so rich i never have to worry about money. oh and i want doughnuts. now.

all i want is eliza dushku and michelle branch for my 22nd birthday. Think you can make that happen london boy ? You have till saturday

oh happy birthday!

i'll try and work my connections and see if i can make something happen. would u settle for, say, dolly parton instead of eliza dushku? heard she's a bit busy..... hehehehehehhehehhe

 

[hey69]

so a patched celeron 733 with a pushed Gforce4 [XBOX]is
soo much faster then a 1.4g pc with a radeon 8500 (low geforce4 comparable) ?

and ps2 grosso modo a pc 1.4g with radeon8500? hm, sounds reasonable.

 

[phed]

Taken from EA shader doc :
[img-poly.png]
Take it with a grain of salt though, but at least it gives us numbers from a real world engine even if EA is not famous for taking the most out of each machine.

Oy, nice find. ...

Edit: And notice, since it is using the stanford bunny it is easy to create your own comparable benchmark since it is a public domain.

 

[Zeross]

so a patched celeron 733 with a pushed Gforce4 [XBOX]is
soo much faster then a 1.4g pc with a radeon 8500 (low geforce4 comparable) ?

Don't forget we're talking of a single model displayed here, it's not a real world test with everything the CPU have to do : AI, physics etc... So I don't think CPU plays a great role in this test. Combined that with the fact that Xbox is a closed platform and developpers can code closer to the metal I don't find it really surprising. Oh and in the T&L area GF4 is better than R200 (two full functionnal Vertex Shader)

 

[Qroach]

Actually, the EA tests included AI running in the background if I remember correctly. Anyway, this is the closest we have to a real world comparrison of each platform under the same conditions.

BTW, EA actually do optimize for each platform individually, they just don't take advantage of extra graphical features to make their games look different from platform to platform.

 

[jvd]

yeah...?

and i want a ferrari, a villa in the Bahamas, be so rich i never have to worry about money. oh and i want doughnuts. now.

all i want is eliza dushku and michelle branch for my 22nd birthday. Think you can make that happen london boy ? You have till saturday

oh happy birthday!

i'll try and work my connections and see if i can make something happen. would u settle for, say, dolly parton instead of eliza dushku? heard she's a bit busy..... hehehehehehhehehhe

thanks , but i'm into brunettes. Not old blondes with fake um... fake bags of fat .

 

[London Geezer]

Actually, the EA tests included AI running in the background if I remember correctly. ANyway, this is the closets we have to a real world comparrison of each platform under the same conditions.

BTW, EA actually do optimize for each platform individually, they just don't take advantage of extra graphical features to make their games look different from platform to platform.

Isn't "taking advantage of extra graphical features to make their games look different from platform to platform" the same as "optimising"....


i do believe that EA does optimise their multiplatform titles MUCH better than average.

 

[Zeross]

Actually, the EA tests included AI running in the background if I remember correctly. Anyway, this is the closest we have to a real world comparrison of each platform under the same conditions.

BTW, EA actually do optimize for each platform individually, they just don't take advantage of extra graphical features to make their games look different from platform to platform.

Table 1 summarizes some typical performance numbers achieved
by our system. These figures are sustainable throughput rates for
production art assets. For the sake of comparison, we also include
figures for the Stanford Bunny model. The CPU is largely idle in
these examples, as would be required to allow for an interactive applications use of the CPU. The bottle necks on these systems are
generally in the bus to the GPU, and the transformation and rasterization
engines.

For those interrested in the document it's available here : http://www.cs.brown.edu/~tor/sig2002/ea-shader.pdf

 

[ChryZ]

Sore wa himitsu desu !

So, what's the secret?

 

[cthellis42]

Ps2 pushes more polygons than a pc . But a pc has more textures , more effects and is able to download porno.

<cracks up> Best answer EVER!

 

[SMarth]

...and is able to download porn.

Yeah, baby!

 

[Squeak]

Yea but polygon counts by themselves don't matter. I'm sure you can make a game with 60 million polygons with just bilinar filtering that would look much worse than a 20 million polygon game that has aniso , fsaa , bump maping , dot 3 shadows pixel shader 2.0 effects and vertex shader 2.0 effects .

Wouldn't 60 million or even 20 million polygons mean that you wouldn’t need bilinear, aniso, FSAA and all that other stuff, because you would inevitably do REYS style rendering?

 

[randycat99]

That would be a seriously interesting approach for the current console generation. I hope somebody tries it...

Could this possibly work on a PS2, or is the Reyes approach utterly out of the question?

 

[Squeak]

If the claim that PS2 can render +50 million polygons per second is true, then that would be over 3 polygons per pixel at 640 x 480 60 fps.
The question is if the fillrate is up to it, as a lot of it is waisted when drawing such small primitives.

 

[randycat99]

Oooh, that's right. I seem to recall someone posting about that elsewhere. Something to the effect that existing rasterizers are designed to achieve maximum performance when projecting a polygon over multiple pixels, not having a pixel cover many polygons. In such a case, the polygon fillrate would fall to some baseline level corresponding to whatever it takes to put a single poly on a single pixel. Did that make any sense? I dunno...

 

[Squeak]

Taken from EA shader doc :

Take it with a grain of salt though, but at least it gives us numbers from a real world engine even if EA is not famous for taking the most out of each machine.

Shouldn’t that be a truckload of salt? The table contradicts everything we’ve ever heard about the three machines, most notably is the fact that the PS2 has such inferior performance in a “rawâ€￾ test, the kind were it would normally be expected to shine.
And the high poly counts on the xbox? I thought someone once said that the maximum realistic performance would be somewhere around 31 million polys? I am aware that it has programmable vertex shaders, which with tweaking(/cheating) could help performance, but still..?

 

[phed]

Taken from EA shader doc :
Shouldn’t that be a truckload of salt? The table contradicts everything we’ve ever heard about the three machines, most notably is the fact that the PS2 has such inferior performance in a “rawâ€￾ test, the kind were it would normally be expected to shine.
And the high poly counts on the xbox? I thought someone once said that the maximum realistic performance would be somewhere around 31 million polys? I am aware that it has programmable vertex shaders, which with tweaking(/cheating) could help performance, but still..?

Vertex shaders does not help on raw polygon data. Usually using these slows down things. And on the PS2, a simple streaming of an object is not what it is good at. If the vus generated additional polygons the number would go up, but these are after all just a static objects.

And, the XBox is very happy with ideal situations because of the lack of parallelisms (if the developers aren't using the tricks like they used in State of Emergency, streaming polys directly to gpu with dma thus leaving the cpu free).

 

[randycat99]

So these tests were just static polygon data being sent from main RAM? That would seem to explain a lot. Wouldn't it [the winner] just come down to whoever has the most memory to store the most polygon data (assuming the GPU bus isn't bottlenecked)?

 

[Tagrineth]

Just for reference...

The highest maximums achieved that I know of:

Dreamcast - between 3 and 4 million per second
GCN - ~15 million per second (Rogue Leader)
PS2 - ~12-14 million per second (Either J&D or R&C, can't remember which)
XB - ~30 million per second (quoted by ERP) - highest known in-game is around 15 million per second though.

 

[zurich]

Just for reference...

The highest maximums achieved that I know of:

Dreamcast - between 3 and 4 million per second
GCN - ~15 million per second (Rogue Leader)
PS2 - ~12-14 million per second (Either J&D or R&C, can't remember which)
XB - ~30 million per second (quoted by ERP) - highest known in-game is around 15 million per second though.

Polycounts can be misleading when there is no animation (moving characters require skinning, IK, etc) though, ie: Rogue Leader, Racing games, etc.