Arstechnica Write Up

It's not the hardware itself that is the target of his being impressed, but rather procedural synthesis on the whole, and Microsoft's attempts to bring it to the forefront. What impressions Hannibal has with regard to the 360 in general will seemingly be the subject of his next article.
 
that is a fantastic article! nice write up imo. shows us that xCPU will impress even 4-5 years down the line.

"interesting points:

1) IBM drew on the results of the same Broadband Processor Architecture (BPA) initiative that produced the Playstation 3's Cell processor in order to furnish the next-generation Xbox with a truly next-generation core that looks set to last at least the five-year life of the average game console.

2) Codenamed Xenon, the multicore PowerPC CPU that IBM designed for the Xbox 360 boasts a number of unique features that set it apart from any microprocessor seen to date.

3) MOST INTERESTING: This patent (patent number 20050099417) that Microsoft was recently granted covers a method for using procedural synthesis to do real-time skinning of 3D characters. The basic idea behind the patent appears to be as follows. Artists using standard tools (i.e.., motion capture, 3D rendering tools, etc.) generate a character model along with a series of key poses in an animation for that model. This model consists of a set of bones that have been skinned with a deformable skin."
 
dukmahsik said:
3) MOST INTERESTING: This patent (patent number 20050099417) that Microsoft was recently granted covers a method for using procedural synthesis to do real-time skinning of 3D characters. The basic idea behind the patent appears to be as follows. Artists using standard tools (i.e.., motion capture, 3D rendering tools, etc.) generate a character model along with a series of key poses in an animation for that model. This model consists of a set of bones that have been skinned with a deformable skin."
This point I thought the least relevant. Interpolating frames from various 'captured states' seems a small progression. I expect next-gen, certainly for main characters, to have skeletons animated by physics, Inverse Kinematics, etc., and have proper rag-doll properties.

The article did make me realise how much the processors are going to be doing next-gen though. It's been said that if GPU is doing grapihcs and say you had a PPU in, what's left but AI? The reality is next-gen CPU's will be doing game executable, AI, Physics, Sound, interim graphics such as model creation and maybe even texture/material creation. There's a hell of a lot going on. Making the most of it IS going to be mammoth. Things will be might impressive, but also the chance of bugs is going to be much greater :? What if the procedural models go screwy and your arm disconnects from your body? The developers, perhaps mostly for the middleware, are going to have to really know their stuff!
 
X-AleX said:
The writer seems quite impressed by the hardware.
And you?

"In all, the Xbox 360's procedural synthesis capabilities show great promise." ;)

Not just procedural synthesis as a technology in and of itself.

I was intrigued by this statement

"If this idea of small pools of per-thread private, local storage that is accessible to the GPU via switching network sounds familiar, then you must have read my Cell coverage. This is essentially what the SPE local storage provides for the Cell's individual SPE. Xenon achieves this same effect using the L2, and Xenon's scheme has the advantage that it is more dynamically adaptable to the needs of the application, since it's a single store that can be partitioned dynamically. However, what the Xenon gains in adaptability it loses in flexibility, since unlike the SPE local storage, which is just a flat memory space that can be used in any way the programmer sees fit, the Xenon's write buffers can only be configured in one specific way and for one specific purpose (as described above)."

Makes me think that they should have increased the L2 size to no less than 2MB though.
 
That is something that concrns me to. If half the cache is isolated for throughput, you're limiting the rest of say 4/5 threads to a pretty small L2 cache. 1 Mb between 6 threads? Is that likely to cause thrashing? Cell has what by comparison? 2 Mbs level 2 + a total of 256x7 = 3.75 Mb local storage?

It almost looks like Xenon has the less demanding GPU design in terms of bandwidth consumption, and PS3 has the advantage for processing bandwidth consumption.
 
Fafalada said:
These two games actually procedurally generate whole buildings in a modern city-scape.

Ratchet & Daxter ;)

Fascinating article btw. I didn't find the promise of procedural genration (which is just software that devs will have to write of course) as salient as the actual details on the Xenon function.

The cache locking especially was very remeniscent of the Gekko, I was thinking, and then Hannibal went and mentioned it :)

You know, the more I find out about the Xbox360, the more I come to think what an amazing architectural feat it really is. In many ways it seems to be the successor to the system balance that the Gamecube achieved this generation, which makes me think that the X360 may well keep pace with the PS3 for quite some time after the developers get the hang of it.
 
Titanio said:
Shifty Geezer said:
Cell has what by comparison? 2 Mbs level 2 + a total of 256x7 = 3.75 Mb local storage?

2MB of L2? It's 512KB of L2. So 512KB + (7*256KB) = ~2.3MB.

Furthermore, nobody's said yet how much of the cache can be put into streaming write mode anyway. It's not necessarily going to be 50%, possibly less, but definitely not more.
 
Shifty Geezer said:
2 Mbs level 2 + a total of 256x7 = 3.75 Mb local storage?

I am only seeing 512K L2 cache on the PS3 CELL. 256KB x 7 SPEs is right though, which brings it to a total of 2.25MB. The pre-MTV Gamespy leak mentions 64KB of L1 cache for each PPC cores on the Xbox 360 but I cannot find that info on the official specs.

EDIT: Got beat...
 
Oooo, I was way off the mark! Do the SPE's need to go through cache to access XDR? I'm thinking that L2 is primarily for the PPE. Or is Cell gonna have even bigger cache issues than XeCPU?
 
Shifty Geezer said:
Oooo, I was way off the mark! Do the SPE's need to go through cache to access XDR? I'm thinking that L2 is primarily for the PPE. Or is Cell gonna have even bigger cache issues than XeCPU?

Afaik each spe is connected via flexIO, so I don't think they should need to communicate with the PPE's cache.

Nite_Hawk
 
This was a pretty interesting article. The xbox360 really seems like it was designed with very specific purposes in mind, and this is definately one of them. It's really interesting how similar the xbox360 feels to the gamecube of the last generation (which I personally think is a good thing).

I think this is in a way the biggest difference between the xbox360 and the PS3. The PS3 feels a lot like the PS2. It seems like thier primary goals were to build a very very fast, yet flexible architecture, and built the entire machine with the that thought in mind. (And are leaving developers with the task of figuring out how to exploit it all).

Microsoft seems more to have taken a bunch of ideas/goals (procedural synthesis, memory bus efficiency, cheap antialiasing, unified shaders) and built the entire design around those goals. It doesn't seem to be nearly as flexible as the PS3, but it seems that their goals are pretty compatible with most developers and will probably be a win for them anyway.

I have a feeling that the xbox360 is going to be very good so long as you play by Microsoft's rules (don't use really high precision HDR with antialiasing, exploit procedural synthesis, use 3DC, etc). The battle for developers to use their system as the primary development system is probably going to be **very** important this generation.

Nite_Hawk
 
Microsoft seems more to have taken a bunch of ideas/goals (procedural synthesis, memory bus efficiency, cheap antialiasing, unified shaders) and built the entire design around those goals. It doesn't seem to be nearly as flexible as the PS3, but it seems that their goals are pretty compatible with most developers and will probably be a win for them anyway.
In terms of hardware abilities/inherent limitations, I don't see how XeCPU and Xenos are any less flexible than Cell and RSX. Now, if you're talking about the memory/bus configuration, that is something different.
 
Luminescent said:
Microsoft seems more to have taken a bunch of ideas/goals (procedural synthesis, memory bus efficiency, cheap antialiasing, unified shaders) and built the entire design around those goals. It doesn't seem to be nearly as flexible as the PS3, but it seems that their goals are pretty compatible with most developers and will probably be a win for them anyway.
Piecemeal, I don't see how XeCPU and Xenos are any less flexible than Cell and RSX. If anything they are more flexible. Now, if you're talking about the memory/bus configuration, that is something different.

I was kind of thinking everything in general. A lot of it is the memory/bus configuration (IE, spending silicon to make AA cheap and a requirement to use, slower GPU memory vs very fast framebuffer memory), but it's more than just that. Everything in the PS3 is designed to talk to everything else extremely quickly. You can get SPEs talking to the RSX and pulling data from all over the entire system. Everything is interconnected and designed to be very very fast at doing it. The xbox360 does this in places, but it feels like it was designed to be elegant where the PS3 was more designed to be flexible.

Nite_Hawk
 
Singh,
nope, I was talking about US based developer. :p
But like I said, I'm sure there are many others aside for those 2.

Shifty Geezer said:
Do the SPE's need to go through cache to access XDR?
No they don't need to, but they can if you want them to.

Incidentially, the option to access memory directly, by bypassing the cache, has been a standard feature on every Sony console to date, and at least a few others (GC I know of, maybe ERP and guys can fill in for N64, DC etc.).
 
Nite_Hawk said:
Luminescent said:
Microsoft seems more to have taken a bunch of ideas/goals (procedural synthesis, memory bus efficiency, cheap antialiasing, unified shaders) and built the entire design around those goals. It doesn't seem to be nearly as flexible as the PS3, but it seems that their goals are pretty compatible with most developers and will probably be a win for them anyway.
Piecemeal, I don't see how XeCPU and Xenos are any less flexible than Cell and RSX. If anything they are more flexible. Now, if you're talking about the memory/bus configuration, that is something different.

I was kind of thinking everything in general. A lot of it is the memory/bus configuration (IE, spending silicon to make AA cheap and a requirement to use, slower GPU memory vs very fast framebuffer memory), but it's more than just that. Everything in the PS3 is designed to talk to everything else extremely quickly. You can get SPEs talking to the RSX and pulling data from all over the entire system. Everything is interconnected and designed to be very very fast at doing it. The xbox360 does this in places, but it feels like it was designed to be elegant where the PS3 was more designed to be flexible.

Nite_Hawk

I understand your take on this and agree. It does seem that. Obviously both approaches have their pros/cons.
 
Nite_Hawk said:
I was kind of thinking everything in general. A lot of it is the memory/bus configuration (IE, spending silicon to make AA cheap and a requirement to use, slower GPU memory vs very fast framebuffer memory), but it's more than just that. Everything in the PS3 is designed to talk to everything else extremely quickly. You can get SPEs talking to the RSX and pulling data from all over the entire system. Everything is interconnected and designed to be very very fast at doing it. The xbox360 does this in places, but it feels like it was designed to be elegant where the PS3 was more designed to be flexible.

Nite_Hawk

The Xbox looks better from a cost point of view. I would imagine it is much cheaper to make.
 
Back
Top