Xenon System Block Diagram

[Guden Oden]

Granted i don't know a lot about power5 bus architecture, but as they seemingly use a shared l2 cache, this is certainly thighter coupling then "copy & pasting" three "G5s" on a core and connect them to a shared bus.

The "G5" cores (PPC976 probably) interfaces to the L2 and write buffers, the L2 and write buffers interfaces to the bus controller interface. The CPUs are actually abstracted from the CPU interface by intermediary hardware so that there are three CPUs on the same die needn't mean that much really.

 

[Squeak]

Isn't it a bad idea to use quad pipelines on a nextgen console GPU? I'd expect most triangles to be very small, especially at 640*480 resolutions, I'd expect highly tesselated and displaced geometry for at least the characters... Or?

I doubt microsoft is going for a ~ 1:1 pixel to polygon ratio. Especially since they are still putting a lot of emphasis on "pixelshading".

 

[cybamerc]

Three CPUs aside does anyone else feel that the specs are a bit underwhelming (assuming that they are true and represent a system launching late 2005)? 10 mb edram... ugh! No 1080p support and you have to run at 640*480 to enable AA.

 

[zurich]

Anyone else notice the hardware video scaler? 8)

 

[Farid]

I would think that R500 and R600 will have more than 16 pipes, R420 is already there.

Sure it will be great, but you have to remember facts such as: Geforce1,2,3,4,FX = 4 Pipelines.
So if todays cards are 16 pipes that doesn't necessarily mean that next-gen cards will have the twice the number of pipes.

 

[ERP]

Three CPUs aside does anyone else feel that the specs are a bit underwhelming (assuming that they are true and represent a system launching late 2005)? 10 mb edram... ugh! No 1080p support and you have to run at 640*480 to enable AA.

Your assuming the entire frame buffer has to be in that 10mb's.

The GPU is very different from any current ATI part, the trade offs that are being made will probably mean that there is never a very similar PC part, although I'm sure there will be architectural similarities.

The CPU configuration (assuming the diagram is accurate ) is very different to what Sony is proposing, When we know what the final configuration of PS3 is, it'll be an interesting comparison. But I'll bet that flops will be a very small part of the overall performance, it'll be more about the ability to move data around or share it between units.

This is of course all speculation.

 

[Megadrive1988]

Sure it will be great, but you have to remember facts such as: Geforce1,2,3,4,FX = 4 Pipelines.
So if todays cards are 16 pipes that doesn't necessarily mean that next-gen cards will have the twice the number of pipes.

well you are right about that. Flipper is also 4 pipes. I suppose we can expect both Xbox 2 and GCNext VPUs to have 16 pipes and lots of shader units per pipe.

 

[cybamerc]

ERP:

> Your assuming the entire frame buffer has to be in that 10mb's.

Point taken. Still... not having access to detailed information it seems somewhat underwhelming. The various busses are not super fast either.

> the trade offs that are being made will probably mean that there is
> never a very similar PC part, although I'm sure there will be
> architectural similarities.

I wonder if that is clever. No doubt Xbox being essentially a PC has been a huge advantage for M$ this gen. Is a shared development environment enough to make up for the hardware difference?

> The CPU configuration (assuming the diagram is accurate ) is very
> different to what Sony is proposing

Yeah, it's going to be interesting to see what Sony is working on and in particular how development is handled. Whatever Sony comes up with three 3.5 GHz CPUs is pretty nice. There won't be many PCs with that kind of CPU power when Xbox 2 launches.


EDIT: what's the "VPU" inside the CPU blocks?

 

[Panajev2001a]

Three CPUs aside does anyone else feel that the specs are a bit underwhelming (assuming that they are true and represent a system launching late 2005)? 10 mb edram... ugh! No 1080p support and you have to run at 640*480 to enable AA.

No 1080p support ? Do you constantly ignore my posts ? ( plus what ERP said ).

 

[Fafalada]

Now, the comments about "a reasonable fraction of 1 TFLOOPS in Xbox 2 in 2005, so why getting closer to 1 TFLOPS for a 2006 machine like PlayStation 3would be impossible to think about ?" like some people here, that know something that not everyone knows, make sense

PSX3 cpu will have 128GFlops, at Most 256GFlops if Sony delays PSX3 until 2007. Max GFlops you can expect for PSX3 is ~320 for a 2007 launch. This info only reaffirms what I said before, MS is still ready to loose tons of money on XBox, while Sony simply can't afford that, so the only way for PSX3 to come close to XB2 power is if it launches 1-2years later.

Ok ok I'll stop now, before I get banned.

Three CPUs aside does anyone else feel that the specs are a bit underwhelming (assuming that they are true and represent a system launching late 2005)? 10 mb edram... ugh! No 1080p support and you have to run at 640*480 to enable AA.

Front buffer is in main memory, so it doesn't necesserily mean the entire FB has to fit, but it remains to be seen how efficient the option will be to do that.
I find the CPUs pretty cool though - for once a console not having a lowend desktop level CPU is IMO exicting news.

well you are right about that. Flipper is also 4 pipes. I suppose we can expect both Xbox 2 and GCNext VPUs to have 16 pipes and lots of shader units per pipe.

The diagram explicitly states 8 pipes But with 48 shader units, that makes for a heck of a lot. Note it also has the same double rate for pure stencil ops NVidia's chips currently do.

 

[cybamerc]

Panajev2001a:

> No 1080p support ? Do you constantly ignore my posts ?

Huh? You lost me

> ( plus what ERP said ).

Yeah, I guess.

 

[Panajev2001a]

ERP:

> Your assuming the entire frame buffer has to be in that 10mb's.

Point taken. Still... not having access to detailed information it seems somewhat underwhelming. The various busses are not super fast either.

> the trade offs that are being made will probably mean that there is
> never a very similar PC part, although I'm sure there will be
> architectural similarities.

I wonder if that is clever. No doubt Xbox being essentially a PC has been a huge advantage for M$ this gen. Is a shared development environment enough to make up for the hardware difference?

> The CPU configuration (assuming the diagram is accurate ) is very
> different to what Sony is proposing

Yeah, it's going to be interesting to see what Sony is working on and in particular how development is handled. Whatever Sony comes up with three 3.5 GHz CPUs is pretty nice. There won't be many PCs with that kind of CPU power when Xbox 2 launches.

Three G5 class CPUs running at 3.5 GHz, with 1 MB of shared L2, ultra fast FSB and very fast main RAM will smoke Desktop PCs that are out at the same time Xbox 2 launches ( mid 2005 ).

EDIT: what's the "VPU" inside the CPU blocks?

G5's VMX/Altivec unit.

 

[cybamerc]

Fafalada:

> Front buffer is in main memory, so it doesn't necesserily mean the
> entire FB has to fit

It's clearly a GC inspired design but I think what they are Dave, ERP and Pana are suggestion is that the back buffer is stored in main RAM as well but that it is rendered into the eFB. That way you can have frames that exceed 10 mb in size. But it comes at a cost... and well, it'll be interesting to see how it works.

> but it remains to be seen how efficient the option will be to do that.

Do you think it would be best to just have 10 mb for the developer to manage like PS2 or a more fixed design like the GameCube where you have an eFB portion and a texture cache with the hardware taking care of copying and writing data to and from the edram?

> I find the CPUs pretty cool though - for once a console not having a
> lowend desktop level CPU is IMO exicting news.

Yeah and you have three of them. One 3.5 GHz CPU would be pretty nice for a 2005 system going by previous standards.

 

[Panajev2001a]

Sorry it was posted on GA:

Xbox 1 had no e-DRAM and was able to have 720p and 1080i games .

That e-DRAM would likely be used to store the back-buffer and the Z-buffer.

1920x1080p back-buffer at 24 bpp ( no destination alpha ) and full-screen Z-buffer at 16 bpps would mean 10,368,000 bytes or 9.88 MB so I think 10 MB would suffice for them.

 

[cybamerc]

Panajev2001a:

But then there's practically no room left for render to texture operations or other neat tricks.

 

[Megadrive1988]

The diagram explicitly states 8 pipes But with 48 shader units, that makes for a heck of a lot. Note it also has the same double rate for pure stencil ops NVidia's chips currently do.

then I don't really believe this block diagram represents what Xbox 2 is.

I very much doubt the real final Xbox 2 VPU will have less pipes than R420 (12 and 16), even with 48 shader units.

 

[Fafalada]

It's clearly a GC inspired design but I think what they are Dave, ERP and Pana are suggestion is that the back buffer is stored in main RAM as well but that it is rendered into the eFB. That way you can have frames that exceed 10 mb in size. But it comes at a cost... and well, it'll be interesting to see how it works.

That sounds like going around your ass into front pocket :? It sounds like regular cache with a larger buffer just for the size's sake... that can't be right, can it?

Anyway, according to the diagram, it's not going to store any texture data, so it's not gonna be like PS2 either way.
At first thought I was assuming it would be more like GC then (where you manually control copying data from eFB to main mem), but apparently some people think otherwise?

Yeah and you have three of them. One 3.5 GHz CPU would be pretty nice for a 2005 system going by previous standards.

Indeed - right now even my dream rig (the fastest dual opterons you can buy) would get smoked by this *L*

 

[Dave Baumann]

It's clearly a GC inspired design but I think what they are Dave, ERP and Pana are suggestion is that the back buffer is stored in main RAM as well but that it is rendered into the eFB. That way you can have frames that exceed 10 mb in size. But it comes at a cost... and well, it'll be interesting to see how it works.

Well, I tend to look at it as being whatever fits in the eDRAM comes for a very low cost, whatever is left over is normal speed!

Note, IIRC this is actually very similar to what Bitboys had in mind for their PC solutions in terms of the frame buffer arrangements. Also note, that I would assume that some ATI's other schemes, like HYPERZ (Hier-Z / Z Compression / fast Z clear) will also be present. Not sure though.

AFAIK the CPU's will also have dp4 functionality which should make them handy for vertex processing - I think the idea will be that the ALU's of the graphics part will predomanently be used for fragment ops, but if the dev wants to do VS on the graphics then that option is there, or of there are VS ops that can;t be done on the CPU (such as vertex texturing). (Now, lookm back to some of Tim Sweeney's comment here).

 

[cybamerc]

I think Panajev has a point actually. A 4 gigapixel GPU won't be rendering HDTV resolutions with AA anyway.

 

[Dave Baumann]

Not all the AA specs are available from this information...