RSX: 1.1 billion vertices/sec?

[Brimstone]

"And you can't tdo this on XB360 because?
Why do you think they let the GPU read and write from the CPU cache?"

The big problem with that is, there's simply not enough cache on the CPU to hand off that much of a workload from the GPU to the CPU. The cache is already being strained by the 3 cores, you can't put that much more of a workload on it or else it's going to bottleneck what the 3 cores are doing, and thus slow them down.

So GPU's require large cache's for their operations or are you suggesting that for a CPU core to perform a GPU style operation it will need a lot of cache memory because of architechtural design differences?

 

[overclocked]

It's quite clear to me with XeCPU and Xenos. However with SPUs and RSX, I was expecting them to remove the VS units and only leave PS units in RSX. But if SPUs and VS units can work asymmetrically, then I suppose it's irrelevant and even provides more options...

Its good they follow a "safe" easy path for vertex-performance and they have quite a bit of fragmentpower over dont you think?

 

[richardpfeil]

I really would have liked to see Sony and NVidia scrap the vertex shaders from the RSX and write a CG and OpenGL shader compiler for the SPU of the Cell. They could spend the vertex shader transistor budget on some other part of the graphics pipeline, and the SPUs would get some use without writing a line of SPU code. Add in a physics API written for the SPUs and the power of the Cell gets harnessed fairly painlessly.

I can't help thinking that the 360 and PS3 will end up looking something like this to cross-platform developers...
- One Power PC, dual thread core @ 3.2Gh.
- OpenGL renderer.
- 256MB graphics memory.

Running within those specs works well for both consoles, and also the PC.

 

[nAo]

CG and OpenGL shader compiler for the SPU of the Cell

be faithful

 

[version]

32 pixelshader 475 Gflops + 2 SPE for geometry 51 Gflops= 526 Gflops
and 5 SPE for phisics,sound etc..

 

[therealskywolf]

32 pixelshader 475 Gflops + 2 SPE for geometry 51 Gflops= 526 Gflops
and 5 SPE for phisics,sound etc..

32 pixel shaders

 

[Titanio]

32 pixelshader 475 Gflops + 2 SPE for geometry 51 Gflops= 526 Gflops
and 5 SPE for phisics,sound etc..

32 pixel shaders

Version likes to speculate and theorise. Mostly about what he'd like to see. In this instance, he believes Sony should put vertex shading on Cell entirely and have a 32-pixel shader GPU.

 

[ralexand]

32 pixelshader 475 Gflops + 2 SPE for geometry 51 Gflops= 526 Gflops
and 5 SPE for phisics,sound etc..

32 pixel shaders

Version likes to speculate and theorise. Mostly about what he'd like to see. In this instance, he believes Sony should put vertex shading on Cell entirely and have a 32-pixel shader GPU.

That would be cool.

 

[Iron Tiger]

32 pixelshader 475 Gflops + 2 SPE for geometry 51 Gflops= 526 Gflops
and 5 SPE for phisics,sound etc..

32 pixel shaders

Version likes to speculate and theorise. Mostly about what he'd like to see. In this instance, he believes Sony should put vertex shading on Cell entirely and have a 32-pixel shader GPU.

That would be cool.

A 1.1Ghz 16 pipe chip would be cooler, but we're not gonna get that from the RSX either, so there's no point in veering so far off topic with fantasy.

 

[richardpfeil]

The vertex shaders are quite a bit simpler than the pixel shaders (no bi-linear texture fetch, Vec4 + Scalar instead of 2 Vec4). I don't think you could trade the transistor count of one vertex shader in for another pixel shader. I think you could still get to 32 pixel shaders if you also cut the texture units in half, each quad having two instead of four (maybe decoupling them from the first shader unit in the process). This would be bad for legacy PC games, but on modern engines it should keep up with texture demands just fine.

While we're at it, half the ROPs and make each 4 fragment per clock capable, with FP16 enabled (HDR).

 

[nAo]

[version mode on] Remove 8 VS, a video processor, FOG ALUs, and 8 ROPs...Iìm betting there's room for other 8 pixel pipelines [version mode off]

 

[Fafalada]

*tugs nAo's shirt*
add 32bit normalize, don't forget 32bit normalize

 

[nAo]

*tugs nAo's shirt*
add 32bit normalize, don't forget 32bit normalize

bad baby, baaaad baby

 

[Thegameman]

regarding the part I highlighted in bold, I agree. just to add to that, the 500 million polygons per second figure for Xenos/Xbox360 does indeed look more reasonable than the peak polygon figures of Xbox: 300M micro / 150M T&L'd polygons per second originally - later got reduced to 233M micro / 116M T&L'd polygons per sec. still MS claimed 100M sustained even after the two core clockspeed downgrades - and also PS2: 66M transformed / 75M rasterized polygons per second.

IGN Xbox: Last year Seamus Blackley was quoted as saying in an interview with us that "The only changes that you'd ever possibly see -- and I think that the probability of there being any changes is extraordinarily low -- would be upgrades to system performance." What happened?

J. Allard: The honest truth is that the goal that we always had for the system was 3x the graphical and computational performance of PlayStation 2. Initially, we thought that a 600MHz CPU and a 300MHz GPU was about right -- and that was in March. Now that Nvidia's got the NV20 in production, and we've got NV20 cards working with the operating system out in dev kits, and we've got games up and running on NV20s, we learned a bit more about the production and the manufacturing, and we decided that the 250MHz combined with the 733MHz is really the right balance. We'll still hit the 3x, but we guessed bad with the 300Mhz.



IGN Xbox: The poly performance has changed as well though, from 300 million polys a second to 125.


J. Allard: The funny thing about poly performance is that that's a theoretical number, and it's a good number to talk about because competitors talk about theoretical numbers, but we've long held that the most important number is what game performance looks like. So the fact that NV20s are in developers hands is great, and the final hardware will be in their hands in a couple of months. I think we'll have no problem making the 3x differentian in real game performance, and maybe even more.




IGN Xbox: You've said once before that there wouldn't be spec changes -- what's to stop more announcements in the coming months?

J. Allard: It's more of a spec refinement, not a change. Technically it's a change -- that's the clock rate we're going to be able to achieve and still maintain the 3x performance. From a physics change it's what's possible and what's right. It's not like there's a 300MHz part that's appropriate and we're shaving some money. It's really the right thing overall in the system architecture. We need to get final hardware to the developers in the next couple months to get them to write games for 2001 that shine.

This ^^^^ s the end of the story,that last question put J Allard against the wall,you can clearly see that he din't even know what to answer,first he say is not a change then he say technically it's a change.


The problems was that MS lie a little much on those xbox specs,in fact they go on saying that they still hit 3 times the perfomance of the PS2,which is a realy big and open lie,i hope they don't do the same again.

Are we talking about XBox vs PS2? I think 3X is reasonable for ingame polygon performance for some scenarios. For instance I think Sony stated something like 16 million PPS max for games texured, lit, etc.. And I heard that some XBox games go as high as 50M PPS textured and lit.

Also, for car games, Forza, RSC2 and other top XBox games are budgeted like 15,000~20,000 polys per car while I think GT4 is budgeted around 5000~6000 polys. That's why top XBox car games usually have car damage for instance.

The most graphicaly impresive game on xbox doesn't look even 1 time better than the most graphicaly impresive PS2 game,it have cleaner picture sharper graphics and more polys,but not to the point were you can say even 2 times better.

You can talk about specs but the final games like GT4 and Forza is what realy show how far is one from the other.

 

[jvd]

The most graphicaly impresive game on xbox doesn't look even 1 time better than the most graphicaly impresive PS2 game,it have cleaner picture sharper graphics and more polys,but not to the point were you can say even 2 times better.

You can talk about specs but the final games like GT4 and Forza is what realy show how far is one from the other.

Untill u run the xbox games in thier native res. They will look much better at that point .

 

[overclocked]

How can someone even say that game looks 1.23423% better than that?
Has gaming lost all the feelings from when i was young, really sad.

 

[LunchBox]

on the subject of poly-counts on xbox/ps2...

it's been a while, so correct me if I'm wrong, but i remember reading an interview from a developer for LOTR:two towers stating that they used more polygons on the ps2 version than on the xbox version... although they used better textures on the xbox.

 

[Shifty Geezer]

LOTR Games on PS have a pretty dire frame rate though. Can't remember TTT too well, but I definitely remember major frame rate retardation in the halls of the dead, ROTK. Walking throurgh the mist was almost painful

 

[3dcgi]

I didn't notice major frame rate drops in The Two Towers or Return of the King (PS2). And the mist was only painful because it cut your movement rate in half. I.e. couldn't run, only walk.

 

[randycat99]

...and what would-be King is going to be dumb enough to run through a waist deep fog layer? Got to step careful, lest you trip on a rock and break your neck. No neck, no RotK ending, right?