xbox 360 specs (unofficial, but believable)

Titanio said:
This doesn't really tell me much..unless the X360 CPU is a hypothetical, equivalently clocked K8 core and your assumption is correct ;)

Few of us here have access to actual hardware - I don't - so forgive me if for now I work with paper technology. Comparing theoretical to theoretical seems fair to me, for now, in the absence of benchmarks/hardware, and I feel that a significant difference on paper isn't going to be so betrayed in the real world as to reverse the situation.

But with the caveat that we are looking at all the numbers correctly. Bits, MHz, GFLOPs, etc... they only tell us so much.

The reality is thus: The xCPU and CELL (which, if you ask Sony fans here, the CELL shown 2 months ago is no clear indication of what we may or may not see in the PS3) are TOTALLY different paradigms. For example, GFLOPs:

If we compare the GFLOPs of the Xbox 360 to the PS3, we are hearing numbers like this:

xCPU: 80GFLOPs
PS3 CELL: 256GFLOPs

Ok, but this number neglects to mention two things:

#1 Not all game code can/should run on the SPEs/Vector units
#2 What are the xCPUs 3 PPC cores doing at this time? That is a lot of untapped power that we are not counting because we are focusing so narrowly on GFLOPs.

They are just very different. Whereas on the xCPU you can have code running on the PPC cores, on the CELL--since the PPE is going to be tasked with keeping the SPEs fed--will be too busy to do these tasks many times so the code will need to be designed to run on the SPEs unless it absolutely cannot.

I expect the PS3 to be a good measure more powerful than the Xbox 360 in GFLOPs, and therefore exclusive games that deal with math intensive tasks like physics should perform MUCH better.

But to remind us of this generation, more GFLOPs on the PS2 did not generate better looking games (where the Xbox is considered to be the most powerful). And yet in its own little corner the well balanced GCN not only held its own, but produced some of the best looking games this gen--including possibly the best in RE4 (and I am not even a fan of the game).

I am not saying GFLOPs are not important, but there needs to be an * because the xCPU and the CELL are VERY different paradigms of how to deal with processing code. I believe BOTH Xbox and PS developers are going to be pulling their head out with multithreading issues for years to come and in that regards development tools will be vital.

I will have to agree with others though that in general I am not expecting much difference--at least not on the surface. ATi and nVidia have been close for 3 years in performance (give or take on features and performance in specific settings). The two systems are being released 4 months apart. There are so many factors that affect a systems performance... but it ultimately comes down to how much support MS/Sony give developers, how good their HW/tools combo is, and how many people they get on board. Great developers can make any HW shine. Looking at games like GT4 and RE4 really show that the HW differences, while not unimportant, can really be totally overblown on a forum like this. Definately GFLOPs alone are not going to tell us what kind of games we can expect.
 
I don't know if this has been mentioned yet (can't read this right now) but has anyone taken note of the GPU using "unified shaders".

Up to this point SM3.0++ was what was expected...could we now expect SM4.0 being used with Xbox titles?

The R500 architecturally would be like the R600 which would be a SM4.0 part and MS no doubt have the Xbox 360 use a custom version of directx...err WGF...why not a custom version of WGF2.0? I mean they don't have to listen to any IHV but ATI with Xbox 360 because it's their system.

Sorry if someone else brought this up.
 
Up to this point SM3.0++ was what was expected...could we now expect SM4.0 being used with Xbox titles?

Well considering sm4.0 or whatever the future of shaders will be called wont be out till 2007 i doubt this part has anything at all in commen with it . However i would expect it to be more advance than the other sm3.0 parts on the market when it launches . It may not be a fillrate beast but where it matters it will be more powerfull than the r520 .
 
Without reading the whole thread, but paying attention to some arguments about the CPU power on PS3 & XB360..

I'm thinking that are everyone forgetting that PS3 CPU (Cell) is doing a bunch of vertex calculations etc too?
You shouldn't compare them in any way, since they're not comparable.

Full systems could be compared if it would be possible to build fair benchmark which would run on both machines, but pure CPU power not, when they're not doing the same job anyway.
 
Kaotik said:
I'm thinking that are everyone forgetting that PS3 CPU (Cell) is doing a bunch of vertex calculations etc too?

Well, we actually don't know that to be the case as yet. Unless you have information that the rest of us don't of course.
 
xbdestroya said:
Kaotik said:
I'm thinking that are everyone forgetting that PS3 CPU (Cell) is doing a bunch of vertex calculations etc too?

Well, we actually don't know that to be the case as yet. Unless you have information that the rest of us don't of course.

No, of course, it's only rumors until confirmed that nVidia's gfx chip doesn't include any vertex capabilities, which would be left to be handled by Cell, but that's what, as far as I've understood, is pretty much accepted to be near fact info?
 
For the record, forum devs here have stated plainly before that the PS3 GPU do have vertex shaders. As usual this was either forgotten or ignored as the speculators painted their own speculations about Cell and PS3. So much so that many forum devs can't be bothered to repeat themselves anymore.

If the next question is 'then what do the SPUs do?', this was also discussed and summarised as - 'lots of things we can think of'. The latest very constructive discussion just took place last week actually.
 
the PS3 *GPU* would not be a GPU if it couldnt do vertex shading / geometry processing / T&L :) so of course it has to have vertex shaders, or is capable of using its resources toward vertex shading / geometry / lighting
 
jvd said:
Up to this point SM3.0++ was what was expected...could we now expect SM4.0 being used with Xbox titles?

Well considering sm4.0 or whatever the future of shaders will be called wont be out till 2007 i doubt this part has anything at all in commen with it . However i would expect it to be more advance than the other sm3.0 parts on the market when it launches . It may not be a fillrate beast but where it matters it will be more powerfull than the r520 .

This may not be so relevant. The Xbox has a modified OS and version of directx on it's HD.

The same could apply in some manner with the Xbox 360 presumably with the software would be on the dev kits. That modified OS or more pertinent to what I'm getting at could be something very similar to WGF2.0. We could call it SM4.0 beta or alpha in WGF2.0 beta or alpha.

Seems fair game to me since the NV2A was a Geforce 3 with special abilities that was controlled by a custom directx with a custom OS.

The only thing that would negate this is things just not being ready yet. The R500 which is the HW is done or will be not later than launch. The spec would then be done as well at that point. Since the R500 architecturally is the basis for or is a crippled version of the R600 which will by SM4.0 compliant it's not a stretch to think it could meet the SM4.0 spec or at least for the most part. It's spec then HW built to spec with the assumption that spec is SM4.0 given the R500 core is destined to that end in the R600.

I'm simply stirring up some thoughts is all.

It could be unified shaders that have SM3.0++ limits to instruction lengths for shaders vs. no limits etc etc etc.

It's just some food for thought, but we'll know soon enough I suppose.

Oh and as for whether it will be a filtrate beast or not...we still don't know what type of ram will be bundled with that e-dram nor what it's IPC performance is etc. It may at 500MHz still be a beast if it's more efficient or capable clock for clock than say the R420 like the NV40 is. Then there's other capabilities to consider like load balancing, high sample temporal AA at FPS at or beyond 60FPS, etc etc.

I think we should wait and see what the R500 can do before we look at that 500MHz number and get upset about it. Not to insinuate anyone was upset or anything.
 
Acert93 said:
#1 Not all game code can/should run on the SPEs/Vector units

I'd agree with the latter, but not necessarily the former (?) Computationally speaking, if nothing else..

Acert93 said:
#2 What are the xCPUs 3 PPC cores doing at this time? That is a lot of untapped power that we are not counting because we are focusing so narrowly on GFLOPs.

The SPEs aren't just FPUs either ;) But we do need a metric...people seem to gravitate to Gflops.

Acert93 said:
They are just very different. Whereas on the xCPU you can have code running on the PPC cores, on the CELL--since the PPE is going to be tasked with keeping the SPEs fed--will be too busy to do these tasks many times so the code will need to be designed to run on the SPEs unless it absolutely cannot.

I'm not sure how true this is. We've had no indication thusfar of the overhead involved in SPE management for the PPE - when it is tasked with management. I say the latter because Sony did present models at GDC that would use little-to-no PPE intervention for SPE task management (though, of course, that scheme would be harder to implement). In other words, I think even in worst case scenarios, some code could be put on the PPE. If you play your cards right, virtually the entire PPE may always be at your disposal.


Acert93 said:
But to remind us of this generation, more GFLOPs on the PS2 did not generate better looking games (where the Xbox is considered to be the most powerful). And yet in its own little corner the well balanced GCN not only held its own, but produced some of the best looking games this gen--including possibly the best in RE4 (and I am not even a fan of the game).

I believe the PS2's "Gflops" for the most part went untapped, unfortunately, and when they were tapped it was usually for things the GPU would normally do. There's the argument that PS3 may end up doing the same thing, but I think relatively there should/would be a lot more room there for PS3's CPU to take on that, and still leave plenty more left over.
 
This may not be so relevant. The Xbox has a modified OS and version of directx on it's HD.

What you just said is not relevant . The question is how close or will it be wgf 2 or dx 10 . The answer is not close at all .

The same could apply in some manner with the Xbox 360 presumably with the software would be on the dev kits. That modified OS or more pertinent to what I'm getting at could be something very similar to WGF2.0. We could call it SM4.0 beta or alpha in WGF2.0 beta or alpha.
wgf 2.0 seems to be 2 years away . I highly doubt anything released in the next few months will be anywhere close to the wgf 2.0 feature set esp in regards to shaders .

Seems fair game to me since the NV2A was a Geforce 3 with special abilities that was controlled by a custom directx with a custom OS.
the nv2a didn't support p.s 1.4 which was already out when the xbox launched. Your telling me that a chip released now is going to support the shader requirements of a api that wont be released for about 2 years ?

The only thing that would negate this is things just not being ready yet. The R500 which is the HW is done or will be not later than launch. The spec would then be done as well at that point. Since the R500 architecturally is the basis for or is a crippled version of the R600 which will by SM4.0 compliant it's not a stretch to think it could meet the SM4.0 spec or at least for the most part. It's spec then HW built to spec with the assumption that spec is SM4.0 given the R500 core is destined to that end in the R600.
we don't know if the r600 will be sm4.0 . sm4.0 looks to be a year away from the r600s launch .

It's just some food for thought, but we'll know soon enough I suppose.
I think its a pipe dream . Its akin to saying wow a 4x 32 cell set up would be really nice in a ps3 .

Oh and as for whether it will be a filtrate beast or not...we still don't know what type of ram will be bundled with that e-dram nor what it's IPC performance is etc. It may at 500MHz still be a beast if it's more efficient or capable clock for clock than say the R420 like the NV40 is. Then there's other capabilities to consider like load balancing, high sample temporal AA at FPS at or beyond 60FPS, etc etc.


Well i would expect that the r500 will be more capable and efficient than the r520 and r420 . Its a new tech not based off the r300 . However i think the budget of the high end graphics card will be higher than that of the x360 or ps3 and thus the high end cards of the time will have more fillrate and more ram in total .

Now the edram is very interesting and you will most likely see it perform better at fsaa .

I think we should wait and see what the R500 can do before we look at that 500MHz number and get upset about it. Not to insinuate anyone was upset or anything.
I'm not upset with it . I'm liking how the x360 is shapping up and if these specs are true its around what i thought it would be. I'm not a huge dreamer though .
 
ERP said:
I have seen benchmarks of the same code run on pre-production Cell (PPU and SPU) and xenon CPU's, a lot of people here would be surprised by the results especially when they are compared to current P4 and G5 results for the same code.
Are you surprised? I'm not and I bet you are not too.
Gigaflop/s numbers mean nothing, those numbers give us no info about real world performance.
The sad thing is that a lot of devs I know don't have a clue how to exploit NG CPUs power and they are not that much informed about CPUs architecture.
As example: first time I ported our old CPU skinning code from PC/XBOX to PS2 I profiled it and observed main character skinning (7000 polys) on PS2 took 20 ms!! A full PAL frame!
The same code on XBOX run 10 time faster, It was just running on EE and it didn't make any use of some fancy vector unit, no DMA, no data customization/reordering.
Now, after making some use of VU0 and DMA engine it takes about 2 ms, it's an order of magnitude faster than before and probably it could be, if further optmized, 2x faster than that.
'Exotic' new architectures need a lot of custom work but I believe this time things will not be THAT different between X360 and PS3 like it was on XBOX/PS2.
 
nAo said:
ERP said:
I have seen benchmarks of the same code run on pre-production Cell (PPU and SPU) and xenon CPU's, a lot of people here would be surprised by the results especially when they are compared to current P4 and G5 results for the same code.
Are you surprised? I'm not and I bet you are not too.
Gigaflop/s numbers mean nothing, those numbers give us no info about real world performance.
The sad thing is that a lot of devs I know don't have a clue how to exploit NG CPUs power and they are not that much informed about CPUs architecture.
As example: first time I ported our old CPU skinning code from PC/XBOX to PS2 I profiled it and observed main character skinning (7000 polys) on PS2 took 20 ms!! A full PAL frame!
The same code on XBOX run 10 time faster, It was just running on EE and it didn't make any use of some fancy vector unit, no DMA, no data customization/reordering.
Now, after making some use of VU0 and DMA engine it takes about 2 ms, it's an order of magnitude faster than before and probably it could be, if further optmized, 2x faster than that.
'Exotic' new architectures need a lot of custom work but I believe this time things will not be THAT different between X360 and PS3 like it was on XBOX/PS2.


This particular test was a pretty reasonable task, in that it didn't really favor one architecture.

As for surprises, yes I was surprised by some of the results. Some of them are extremly puzzling.

And FWIW what surprised me isn't the data I would expect to surprise most on this board.
 
Back
Top