Intel Conroe: From a Console's Perspective

[Guden Oden]

ATI and IBM were hired to develop specific SOCs for MS. MS owns the designs for these SOCs, they can do with them what they want, including integrating them on a single piece of Si in the future.

Yeah, owns the designs of the finished chips - NOT the IP they're based on, because ATi and IBM relies on that IP in other products. Be kind of stupid to sell that off to MS just because they got a contract with Billy Goat to build him a console wouldn't you say?

 

[Gubbi]

Yeah, owns the designs of the finished chips - NOT the IP they're based on, because ATi and IBM relies on that IP in other products. Be kind of stupid to sell that off to MS just because they got a contract with Billy Goat to build him a console wouldn't you say?

You explicitly said that MS did not own the IP of Xenos and Xenon.

They do.

Xenos is MS' property. ATI cannot start producing PC add-in GPUs based on the Xenos design. They can however make something quite similar based off their know how about how to make such a system. Which is of course where the real value for ATI lies, and which is what I think you meant (so we agree )

Cheers

 

[pjbliverpool]

Just remember it took the PC market till 2005 to finaly being able to match the Emotion Engine (PS2 CPU) in power.

A 1.6Ghz P4 exceeded the EE in single prescision FLOPs. Im pretty sure that came out before 2005. And thats comparing to the EE with a pair of dedicated vector processors, not a general purpose CPU.

Continue ignoring all the devs who say the numbers are meaningless if you like. Continue believing that IBM managed to slap together a super CPU which would outperform Intel future state of the art technology by x times while still being small, cool, power efficient, and cheap enough to mass produce for a console.

I prefer the real world.

 

[superguy]

This is basically the whole "high Gflops vs Out of order/ease of use" debate.

Xenon the former and Conroe the latter.

I happen to believe the latter is better.

But it's also back on Cell. Because Cell only has 2X the Gflops of Xenon. Not 100X or something like people act sometimes. So if you think Conroe would "crush" Xenon, then by extension it would probably crush Cell.

 

[j^aws]

Theoretical numbers are so damn useless it drives me insane to see so many on console forums constantly quote them.

I'll tell you what drives me up the wall even more is when people can't even quote the *correct* theoretical numbers. Nor being able to differentiate between bullshit PR numbers and *actual* theoretical numbers. No wonder most people think they're meaningless because they don't even know what they are looking at...

 

[Asher]

Plus compiler technology is far more advanced for x86 than the PPC architecture.

I know a room full of engineers who would consider those fighting words.

The POWER and PowerPC architecture were performance-oriented from their inception, and so were the compilers. x86 compilers haven't started pushing performance until later.

 

[Asher]

There are also optimizations that have existed for years in IBM's Power/PowerPC compilers that are just now making it into gcc and Intel compilers.

Off the top of my head is the Tree SSA optimization infrastructure: http://people.redhat.com/dnovillo/pub/tree-ssa/papers/nordu2003.pdf

 

[Fox5]

It would be better.

The just released benches show it easily besting a A64 by 20-50%, as I imagine you know.

1. On benchmarks made and optimized by Intel. They're probably real, but don't bet the farm on it.

2. All we know is that it destroys an A64 on branchy code, which PC cpus already do to the consoles. That obviously wasn't the focus of the consoles. (in the case of xbox 360, I'd say the focus was marketting numbers and owning the IP)

Just remember it took the PC market till 2005 to finaly being able to match the Emotion Engine (PS2 CPU) in power.

And yet within a year of release, PC cpus could outperform the emotion engine is just about any real task. And the ones they couldn't, GPUs could.

Using a 20 year old architecture with constantly hold up to date compilers.
So it's simple and there are no suprises but my god it's slow and dated.

Not true, Intel has 100s of pages of Pentium 4 specific optimizations that would have little, or perhaps even a negative effect on other x86 processors.

Originally Posted by Guilty Bystander
An Amd 64 4000+ can do about 6GFlops


I'd like to see a source or your derivation for this.

Sounds like he's comparing sustained performance of the 4000+ to the theoretical max of the Emotion Engine.