Is the Xenos a shader monster yes or no?

[joebloggs]

in terms of shader performance Xenos is comparable with X1900XT , not X1800XT and in terms of shader performance it looks like RSX is closer to 7800 performance than it is 7900 performance (7900 GTX+ is 700 Mhz compared to 7800 which is 550 Mhz *same as RSX*) , 7900 also has 1800 Mhz memory BandW compared to 1200 from 7800 *which is even higher than RSX* AGEIA is right on the account of architecture but wrong on the account of performance


all you have to do is look at whats available infront of you and you get a pretty consice view of the performance

Well I guess E3 will give us a better understanding.
There's too many opposing theories floating around at the moment.

 

[Robert.L]

in terms of shader performance Xenos is comparable with X1900XT , not X1800XT and in terms of shader performance it looks like RSX is closer to 7800 performance than it is 7900 performance (7900 GTX+ is 700 Mhz compared to 7800 which is 550 Mhz *same as RSX*) , 7900 also has 1800 Mhz memory BandW compared to 1200 from 7800 *which is even higher than RSX* AGEIA is right on the account of architecture but wrong on the account of performance


all you have to do is look at whats available infront of you and you get a pretty consice view of the performance

Bullocks in terms of shader performance M Dogget himself stated that Xenos rivals 1800xt , theoretically though 1800xt has more raw pixel shading power than Xenos , but that's theoretically since 1800xt won't maintain it's peak ops per cycle in the pixel shaders and etc.

and rsx is part of NV4x family and it looks like it's is a modified G71 but to what extent it's still unknown

 

[overclocked]

I was wavering on that, but 48 "full" ALUs vs. 40 (16 full + 16 mini + 8 VS) seems pretty even. Considering R520XT's 25% clock advantage, I suppose you could give it the slight edge. But if we're counting MADDs, then Xenos appears to win, hands down.

Or am I overlooking something? Are you considering threads, batch sizes, tiling with AA, maybe HDR?

No you probably misunderstood me. I waved along you and i would think that it should be in that range ~.
Because its closed enviroment etc a comparison is impossible but i would certainly think that if it has the same "competens" as a X1800XT its very high *already* IMO.

 

[Hardknock]

I've never seen a PC game with textures like this:





Yes, it's a shading beast when in the right hands.

 

[dskneo]

I've never seen a PC game with textures like this:





Yes, it's a shading beast when in the right hands.

my UnrealTtournament 1999 ground/Wall textures in openGL with nvidia card gives those textures a run for their money.. minus the obvious tech such as bump mapping etc that kameo as.


one gpu as to be better than the other, even if its just a hairline.... we just dont have a way to find out who.

 

[!eVo!-X Ant UK]

I've never seen a PC game with textures like this:





Yes, it's a shading beast when in the right hands.

There's texture's in the fisrt halo game that match those, namely the tree bark

 

[TurnDragoZeroV2G]

Bullocks in terms of shader performance M Dogget himself stated that Xenos rivals 1800xt , theoretically though 1800xt has more raw pixel shading power than Xenos , but that's theoretically since 1800xt won't maintain it's peak ops per cycle in the pixel shaders and etc.

and rsx is part of NV4x family and it looks like it's is a modified G71 but to what extent it's still unknown

At lower resolutions say 640x480 or
HDTV then the xbox is most likely faster due to it's slightly higher
shader power and high framebuffer bandwidth. But at higher resolutions
1600x1200 it could possible go to the X1K. They both have similar
performance. In the end it comes down to the type of application.

In any case, it doesn't make much sense for X1800XT (which we're technically assuming is the card Doggett was speaking of there, even though it's pretty dang safe to say it wasn't R580... and all in regards to a question about ToyShop, too--how much of a performance increase did that see from R580 anyhow? I don't think it was close to 200%) to best Xenos in shading. Going back to that comparison earlier in the thread, Xenos also has the addition of another vec unit for Vec4 + scalar instead of vec3 + scalar. Not a very good nor exact science, but (slightly) better than pulling something from an unpleasant region somewhere...

 

[Hardknock]

There's texture's in the fisrt halo game that match those, namely the tree bark

Come now? Do you have pics to back this up? I haven't seen parellax (sp?) mapping this detailed in any game on the market.

 

[czekon]

There's texture's in the fisrt halo game that match those, namely the tree bark

they should make oblivion for x360

 

[Robert.L]

In any case, it doesn't make much sense for X1800XT (which we're technically assuming is the card Doggett was speaking of there, even though it's pretty dang safe to say it wasn't R580... and all in regards to a question about ToyShop, too--how much of a performance increase did that see from R580 anyhow? I don't think it was close to 200%) to best Xenos in shading. Going back to that comparison earlier in the thread, Xenos also has the addition of another vec unit for Vec4 + scalar instead of vec3 + scalar. Not a very good nor exact science, but (slightly) better than pulling something from an unpleasant region somewhere...

where is that qoute from ? and i'm pretty sure he did compare it to X1800 xt , i'll take a look again .

 

[Guilty Bystander]

they should make oblivion for x360

WTF?
I sure hope you're being sarcastic.

Originally Posted by Mike Doggett
At lower resolutions say 640x480 or
HDTV then the xbox is most likely faster due to it's slightly higher
shader power and high framebuffer bandwidth. But at higher resolutions
1600x1200 it could possible go to the X1K. They both have similar
performance. In the end it comes down to the type of application.

I have this idea the Xenos has about the RAW fillrate and Shaderops of a X1800XL and the Pixel Shader power of a X1900XTX.
The Xenos does 8GTexel/s (16 TMU's x 500MHz) fillrate like the X1800XL and has a maximum of 48 Pixel Shaders like the X1900XTX.
I've heard somewhere problems with the Xenos and the bad framerates it produces has to do with bad branch prediction and flow controll of the ALU controller.
Is there any truth to this?

 

[Hardknock]

WTF?
I sure hope you're being sarcastic.

He's saying built from the ground up for the 360.

I have this idea the Xenos has about the RAW fillrate and Shaderops of a X1800XL and the Pixel Shader power of a X1900XTX.
The Xenos does 8GTexel/s (16 TMU's x 500MHz) fillrate like the X1800XL and has a maximum of 48 Pixel Shaders like the X1900XTX.
I've heard somewhere problems with the Xenos and the bad framerates it produces has to do with bad branch prediction and flow controll of the ALU controller.
Is there any truth to this?

Where had you heard this?

 

[TurnDragoZeroV2G]

where is that qoute from ? and i'm pretty sure he did compare it to X1800 xt , i'll take a look again .

Directly from the original topic here

You're right, though, in that X1800 is mentioned in the question that was posed (which I didn't read again when I c+p'd it), but like I originally thought, it's also regarding how Xenos would run the ToyShop demo itself. In such a case, "slightly more shader power," if he meant effectively in the demo rather than just theoretically, might not have meant "all" of Xenos' shading units, so much as "all the shading units it could use considering a TMU bottleneck." Considering the extra width of Xenos' ALUs (relative to X1K pixel shader ALUs) plus having a total of 48 with branch execution units compared to 8 full VS +16 full PS ALUs + 16 mini-ALUs, with 25% higher clocks, might suggest that it's at least possible. But, I doubt there'll be any answer regarding that.

Edit:

I've heard somewhere problems with the Xenos and the bad framerates it produces has to do with bad branch prediction and flow controll of the ALU controller.
Is there any truth to this?

Since, AFAIK, no GPU does any kind of branch prediction, and if that term was used in what you "heard," then I'd seriously doubt it myself. According to later presentations from MS, Xenos has dedicated branch execution units, which would suggest low costs for the actual branch calculations (i.e., either 1 cycle or effectively "free" due to being done parallel to the rest of the processing like X1K)

Dynamic branching performance should be lower than X1800 though (4x4 pixel batches) and a little worse than X1900/X1600 (4x12 pixel batches? 48 total, regardless) as Xenos is a total of (8x8) in size. Depending on the type of branching going on, the extra work done can get significantly larger and cause the X1900 to perform worse than the X1800. (don't have the link on hand for where one of those benches was done, though, nor can I remember what it was). But I doubt developers would be using such a horrendously performing thing on the GPU unless it had excellent, excellent benefits. You usually go with what works, avoid what doesn't work, and weight the costs and benefits appropriately.

 

[Dave Baumann]

X1900 will never perform worse than X1800 in branching, at worst it will perform the same.

 

[ERP]

X1900 will never perform worse than X1800 in branching, at worst it will perform the same.

Plus there are actually potential performance benefits to dynamic branching even in the case where both paths are taken...

 

[pjbliverpool]

I know this site has higher standards than simply comparing GFLOPS to estimate the power of a GPU but there seems to be a lot of incorrect information in this thread so I thought it better to start from the basic figures and work from there. Prorgrammable Vertex + Pixel GFLOPS for each is:

Radeon X1800XT : 170
GeForce 7900GT : 208.8
Xenos : 216
RSX : 255.2
GeForce 7900GTX : 301.6
Radeon X1900XT : 410
Radeon X1900XTX : 426.4

There, now that you have the basics, you can continue from a correct starting point.

 

[TurnDragoZeroV2G]

X1900 will never perform worse than X1800 in branching, at worst it will perform the same.

Isn't it possible (and likely, in certain shaders), that a shader with a number of branches could cause elements in the batch to take more than just 2 branches by the completion of the shader? In which case, if the larger batch size caused a greater number of branches to be calculated in full instead of being ignored, despite triple the shading units, X1900 could end up slower than X1800.

Given, this wouldn't be a normal pixel shader, but rather something along the lines of GPGPU stuff. And that's probably where I saw the benchmark (if I ever saw it at all).

All of which really makes my statement moot anyway, since it's not really applicable to what most devs are likely to use Xenos for I suppose.

There, now that you have the basics, you can continue from a correct starting point.

I think it's already bad enough doing a simple inequality between shading units from the same company and relatively close generations of chips that seem to share quite a bit of technology (with, of course, a great number of differences as well). Nevermind from two different companies, and then with these GFlops, Shader ops, and such...

Though that reminds me of one thing. It would be interesting to know if the X1K's vertex shaders are the same as what Xenos uses. From ATI we know that the scalar can't do MADD and is primarily there for special functions like 1/x, cos(x), sin(x), 1/sqrt(x), etc. Does the vertex shader of X1k do madd, and more/less special functions, or are they the same.

 

[Dave Baumann]

Isn't it possible (and likely, in certain shaders), that a shader with a number of branches could cause elements in the batch to take more than just 2 branches by the completion of the shader? In which case, if the larger batch size caused a greater number of branches to be calculated in full instead of being ignored, despite triple the shading units, X1900 could end up slower than X1800.

Its immaterial. Think of it like this: the configuration of X1900 (in term of its quads and how it branches) is exactly the same as X1800, except for the fact that per "hardware quad" it has 3x the shaders. Its impossible (at the same clockspeed) for X1900 to perform worse, although it is certainly possible that it will be less efficient (in relation to each of their shader capabilities). In a pathological case the performance could be the same, but in reality, due to the higher general shader performance X1900 will almost certianly outperform X1800 in all cases.

 

[Squeak]

I've never seen a PC game with textures like this:





Yes, it's a shading beast when in the right hands.

Look at the sides, straight polygons where they should have been lots of curves. It looks even less convincing on motion. The transition from straight to curved looks swimming and fake.

 

[scooby_dooby]

Look at the sides, straight polygons where they should have been lots of curves. It looks even less convincing on motion. The transition from straight to curved looks swimming and fake.

whatever, kameo looks awesome, and there is no (read: no) pc game that has textures like that. The brick work is very much like the Toy Shop demo that people wet their pants over. Now that it's a 360 game: "the transition from straight to curved looks swimming and fake"