Xenon System Block Diagram

[Panajev2001a]

The Shader units are 24 and not 48...

48 Shader ops = 24 units * ( 1 scalar op + 1 vector op )

24 * ( 8 ops + 2 ops ) * 0.5 GHz = 120 GFLOPS.

 

[nAo]

Well, like I said in another thread, it has to be more sophisticated than tiling, it has to be a deferred renderer like DreamCast. Otherwise, you wouldn't be able to flush and re-read tiles fast enough, and eventually you'd be stalled flushing the tile to system RAM or reading it back.

All current GPUs are tile based.

Or it can subdivide the screen in N viewports corresponding to N tiles and render one tile at time.
No need to flush all the tiles..at most it might need to flush all the tiles one time after rendering to do some global (on the image) filter, like tone mapping, but even in this scenario it would be pretty efficient.

ciao,
Marco

 

[Crazyace]

Maybe each alu op is a scalar op - so effectively 6 fmac/fmul per pixel per clock, enough for blending op + texture op..

Interesting things is 64bpp at edram - taking 720p ( 720x1280 ) 64bpp+Z/stencil give 720x1280x(8+4) = 10.5Mbytes..

 

[Panajev2001a]

Tough, to say, CPU race on the desktop could speed up, but I think you're basically on the money here. I just feel like take some potency out of this statement. Price performance, this will likely make many things its bitch.

So, you basically just want to be mean, huh ?

 

[Panajev2001a]

Maybe each alu op is a scalar op - so effectively 6 fmac/fmul per pixel per clock, enough for blending op + texture op

Crazyabe, that would give the GPU a rating of only 48 GFLOPS which is a bit more than half of what the CPU can do: it seems odd to me.

 

[darkblu]

Except that no chip with 32bit native color from recent history could store 24bpp in different alignment then 32bit, which means it eats up the same space as 32bpp. Even GS - which had a lot of consideration done in that regard due to same eDram limitations.

2D videocards used to do that commonly, i.e. when there were still 2D videocards

 

[V3]

The Shader units are 24 and not 48...

48 Shader ops = 24 units * ( 1 scalar op + 1 vector op )

24 * ( 8 ops + 2 ops ) * 0.5 GHz = 120 GFLOPS.

How can you came to that conclusion ?

 

[Panajev2001a]

The Shader units are 24 and not 48...

48 Shader ops = 24 units * ( 1 scalar op + 1 vector op )

24 * ( 8 ops + 2 ops ) * 0.5 GHz = 120 GFLOPS.

How can you came to that conclusion ?

Each Shader Unit can co-issue a Scalar and a Vector operation ( remember, this Shader unit should be able to do both Vertex Shading and Pixel Shading ).
I am thinking about MADD as the operation for the Vector and Scalar ALUs which means respectively 8 FP ops/cycle and 2 FP ops/cycle.

 

[Devourer]

http://www.gamesindustry.biz/content_page.php?section_name=dev&aid=3364

 

[V3]

and also noted that the CPU diagram doesn't mention that the CPUs used will be dual-core (effectively making Xbox 2 into a six processor unit), another important factor.

Wow, the spec is upgraded already.

Its a fake afterall

 

[Teasy]

Not necessarily, its more likely to be an older internal concept diagram of XBox 2.

 

[pc999]

Not necessarily, its more likely to be an older internal concept diagram of XBox 2.

or like already someone had say the first diagrame for dev kits

 

[Guden Oden]

2D videocards used to do that commonly, i.e. when there were still 2D videocards

They also lost performance, because one pixel got spread across two words and required two operations to update it.

Besides, 24-bit buffers isn't going to help when we need at least 64-bit buffers next gen.

 

[DemoCoder]

Well, like I said in another thread, it has to be more sophisticated than tiling, it has to be a deferred renderer like DreamCast. Otherwise, you wouldn't be able to flush and re-read tiles fast enough, and eventually you'd be stalled flushing the tile to system RAM or reading it back.

All current GPUs are tile based.

Or it can subdivide the screen in N viewports corresponding to N tiles and render one tile at time.
No need to flush all the tiles..at most it might need to flush all the tiles one time after rendering to do some global (on the image) filter, like tone mapping, but even in this scenario it would be pretty efficient.

In other words, like I said, tile based deferred rendering. How else is it going to do the subdivision without doing scene capture and sorting geometry into tiles?

 

[bbot]

Has anyone noticed that Teamxbox has pulled ther article? No one else seems to have done so.

Does anyone think the figure of 500Mtri/sec for the peak polygon thruput rate is a bit too low? There's gpus already capable of 800Mtri/sec.

 

[marconelly!]

>>and also noted that the CPU diagram doesn't mention that the CPUs
>>used will be dual-core (effectively making Xbox 2 into a six processor
>>unit), another important factor.

>Wow, the spec is upgraded already.

Actually, the 3CPU / 2 Core setup has been rumored since the first rumors appeared earlier this year. That 3CPU setup on the diagram gave the whole thing some credibility, IMO. Text below doesn't talk much about any of the hardware's aspects, why should it be expected to give precise details of the CPU...[/list][/quote]

 

[ninelven]

edited

 

[Saem]

So, you basically just want to be mean, huh ?

Shhh! You're not supposed to notice.

Would a deffered render be that hard in terms of patents or would the actual difficulty of implementing it be more difficult?

 

[Megadrive1988]

but these CPUs for Xenon are not actually duel core. they're meant to be based on the consumer PowerPC chip, not the Power4/Power5.

my understanding is, they'll do 2 seperate threads. whereas a duel core Power4 will do 4 threads, 2 per core.

 

[V3]

Actually, the 3CPU / 2 Core setup has been rumored since the first rumors appeared earlier this year.

From that diagram, you can came up with 6 virtual cores, but not 6 real cores. The rumors were number 3 and 6. For what those numbers are for, I don't know, never got an answer. If it was 6 real processors, I expected the diagram to at least show the cache for each processor core. So from this diagram you would call 3CPU/1 core setup, or just simply 3 processors.

That gameindustry article however seems to confirm its actually 6 real processors, instead of 3 real processors.