Was I right or what

[alanschu]

I think that's a bit of a copout. Here's the article.

 

http://domino.research.ibm.com/comm/resear...innovation.html

 

Written at a pretty high level for a mere "PR release."

Edited June 13, 2006 by alanschu

[mkreku]

See? Alan knows how to make a point in an argument. Claim something, then back it up with something other than "I heard a rumour and innuendo that [iNSERT HARDWARE HERE] is weaker than [iNSERT OTHER HARDWARE HERE]".

[angshuman]

I think that's a bit of a copout.  Here's the article.

 

http://domino.research.ibm.com/comm/resear...innovation.html

 

Written at a pretty high level for a mere "PR release."

Thanks for the link, there is definitely some technical content there.

 

I still think it's a PR release though.

[alanschu]

Well, if you're still not convinced, Berkeley seems to feel it has a role as a gaming processor (they analyzed it for usefullness in scientific computing)

 

http://www.cs.berkeley.edu/~samw/projects/cell/CF06.pdf

 

While discussing the approach they made with respect to memory:

 

If however, the data access pattern lacks predictability,

then the advantages of software-managed memory are

lost. This more aggressive approach to memory architecture

was adopted to meet the demanding cost/performance

and real-time responsiveness requirements of Sony

[angshuman]

http://www.cs.berkeley.edu/~samw/projects/cell/CF06.pdf

(w00t) Now that's the kind of document I like to read. Thanks a bunch Alan. Instead of ranting I could just as well have gone and searched for these myself . I'll post my comments on the article soon.

[Cantousent]

alan's articles show one of the reasons that evaluating consoles is such a hairy problem.

 

Apparently, Cell was built with more than gaming in mind. However, was it built primarily for gaming or not? Now, Cell gets it from both sides. On one hand, folks say that it's a great beast, but not built to be a gaming beast. On the other hand, other detractors say that it has limited function outside of gaming.

 

I used to think this was all double talk. Then, however, it dawned on me that, since the console isn't even released, folks don't really know how it will perform at all. It's not duplicitous. It's simply unknown at this point.

 

I hate to interject this again, but I don't think most folks will care whether the console provides nightmares or angst for the programmers as long as the end product is fun. Great games make for happy customers.

 

The only thing that I can say in regards to this argument is that 10k makes a good point that developers with a more limited budget could be hurt if the costs of programming for the PS3 is excessive. That does suck, but it's also part of business. I don't like the smaller guy taking the shaft, but business has always been an uncaring ordeal.

 

Hell, I think developing costs have risen to incredible levels already, for all platforms. It's not like the PC hasn't left a pile of bodies in its wake. Many of us here are refugees from the IP boards where we followed the BIS division.

 

In angs' defence, hoewver, virtually every article I read about these consoles sounds like PR talk. Sometimes I doubt the veracity of the article. Other times, I don't really think they're lying, just putting a favorable spin on the issue. A lot of the time, I'm sure they believe what they're saying, but I happen to have heard counter-arguments. The IBM article, from what I could grasp as an outsider didn't sound like something that was written to be a press release, but it did have some of that PR glow, only technical PR glow written for other eggheads like the author.

Edited June 13, 2006 by Eldar

[alanschu]

The Berkeley one is a good read, though it's an article about how they would improve upon the Cell to make it better for scientific research.

[10k fists]

Well, you must have heard a claim like that somewhere.  If you didn't find it by reading up on it, then where?

 

Like I said, I don't read Cell documentation from the internet all that often... So you can believe what I say or not, it doesn't really make much difference to me.

 

That's just it.  The impression I'm getting from IBM, is that the Cell's strength is gaming.

<{POST_SNAPBACK}>

 

Strength for gaming in that it's a floating point beast. It's great for pushing great graphics, but we all know graphics don't make a game.

[alanschu]

Like I said, I don't read Cell documentation from the internet all that often... So you can believe what I say or not, it doesn't really make much difference to me.

 

 

All I asked is if you didn't find it on the internet, then where? NDA?

Edited June 13, 2006 by alanschu

[LoneWolf16]

That's just it.  The impression I'm getting from IBM, is that the Cell's strength is gaming.

<{POST_SNAPBACK}>

 

Strength for gaming in that it's a floating point beast. It's great for pushing great graphics, but we all know graphics don't make a game.

<{POST_SNAPBACK}>

Just to jump in here, since everything before this has been confusing and laced with techno-babble...

 

What does this have to do with anything related to the Cell? Graphics don't make ANY game, not just those powered by this particular piece of hardware.

[Cantousent]

From what I can gather, and I'm a reasonably intelligent fellow, This is what 10k is saying: Cell has sheer calculating power. This Xenon business makes it easier somehow to code for a great game. I guess the xbox has three cores, all the same, and that you can have the cores do different things, or simply work with another core to do a hefty job of something. The upshot is that the cores, being the same, make it easier for a programmer. Cell, on the other hand, has a powerful master core, the PPC(?), and a bunch of slaves, SPEs(?). The PPC and SPEs has a huge Gflop advantage.

 

However, I don't think 10k was saying, "better graphics doesn't necessarily mean better gameplay." What I believe he is saying is that, "graphics are only part of the programming involved in developing a game." Whatever he meant, I'm sure he can explain it.

 

However, I'd be curious if the braniacs here could simplify this for some of us mentally challenged folks.

[angshuman]

Eldar, you are correct for the most part.

 

The problem we are talking about here is Programmer productivity. This is with respect to Graphics, AI, Physics, everything. Sure, the Cell has a huge peak potential, but how much effort does it take to extract all that? A hell of a lot, it seems to me. Why does it matter? After all, all we care about is the games right?

 

Yes, when the platform and the games finally arrive, all these discussions would be moot. However, at this point in time, the technical information about the console is all we can use to *speculate* what kind of games we would be able to see on the console. Why do we care to speculate? Just for fun, being geeks and all. After all, we won't have to buy the damn thing until it is launched.

 

And as far as Cell vs. Xenon goes, it is not just the heterogenous nature of the Cell that is going to make programmers pull their hair out. The individual SPEs are also nothing like traditional processors. Together, they have very high peak processing power, but they lack a good deal of the "intelligent" stuff present on modern processors.

 

This "intelligent" stuff (branch prediction, out-of-order execution, memory disambiguation etc.) discovers properties of the program being executed on the fly and exploits these properties to improve performance enormously. Trouble is, all this extra baggage consumes a lot of power and area. Cell's SPEs have a lot of compute power, but they rely on the Programmer and Compiler to do all of the "intelligent" stuff and spoon-feed it with pre-processed optimized instructions. So, in the ideal case, a guy can sit over a desk and hand-tweak his code for months and end up with a super-fast application. But real life is hardly ideal.

[10k fists]

All I asked is if you didn't find it on the internet, then where?  NDA?

<{POST_SNAPBACK}>

 

Not going to claim NDA status on an internet forum, it's far too common to see... I just know, and that's all I'm really willing to say.

 

Just to jump in here, since everything before this has been confusing and laced with techno-babble...

 

What does this have to do with anything related to the Cell? Graphics don't make ANY game, not just those powered by this particular piece of hardware.

 

I'm not totally sure as to what you were asking... But I'm going to take a shot in the dark and hope this is the type of answer you were wanting, gonna try and remove the "techno-babble" as well.

 

Cell is great for pushing geometry (graphics), and I mean really great. It also has good potential to really push a lot of physics as well. But there are other factors that make a video game what it is. Some of these other factors (such as AI) don't have a lot of room for growth on Cell. I'm not saying it'll be bad, I'm just saying that Cell was designed to excell in one direction, but, in a way, remain still in the other.

 

However, I'd be curious if the braniacs here could simplify this for some of us mentally challenged folks.

 

What did you want simplified?

[alanschu]

 

So you have no problems stating the findings of some people that you talked to (or whatever) regarding the Cell, but you refuse to say that you either know someone working on it, or have an NDA and cannot talk about it because it's too common to see on an Internet forum?

 

[Cantousent]

I basically believe 10k. That doesn't mean I think he's entirely right all the time, but he sounds an aweful lot like someone else I've seen on other boards whom I know is a developer.

 

That's irrelevant. Even if I take him as an authority on the subject, that doesn't mean he can't comment on things like the IBM or Berkley report. The big thing is, in a general sense, the arguments you guys use against each other are pretty much the same arguments I've heard. Actually, while I tend not to be a technical wonk, I've been following all these threads very closely.

 

I finally weighed in because of Epiphany's moderated status, whereupon one of the mods let him create this thread, whereupon he picks a fight. *shrug* shoulda kept my mouth shut.

 

This discussion, however, is quite interesting. Apparently, I don't need any clarification. It seems (thanks angs) that I got the gist of it already.

 

I do have a relatively technical question, although I'm sure it will highlight my ignorance. If the SPEs can't predict very well, couldn't the designers use scripted events to make up the difference? I mean, that would be irritating in one way, but it could save the AI effort for more meaningful parts of the game. Also, how would these systems, Wii, 360, PS3, handle a game like FEAR, which claims excellent AI?

 

I'd appreciate any answers you guys have, even if you're a little irritated answering my noobish question. alan, in particular, has been excited by his studies in AI for quite some time.

[alanschu]

Branch prediction can be done in software, though that will take up memory. There are also ways to avoid branching, such as unrolling loops so that they are inline. The problem with this is you might need a lot of registers, which the SPEs do have.

 

I'm doing some reading up on it right now, so I'll be able to contribute a bit more in a bit.

Edited June 14, 2006 by alanschu

[alanschu]

The introduction from a paper from MIT (which was a really good read).

 

The increasing importance of multimedia, game applications, and other numerically intensive workloads has generated an upsurge in novel computer architectures tailored for such applications. Such applications include highly parallel codes, such as image processing or game physics, which have high computation and memory requirements. They also include scalar codes, such as networking or game artificial intelligence, for which fast response time and a full-featured programming environment are paramount.

 

 

Our first contribution is a set of compiler techniques that provide high levels of performance for the SPE processors. To achieve high rates of computation at moderate costs in power and area, functionality that is traditionally handled in hardware has been partially offloaded to the compiler, such as memory realignment and branch prediction. We provide techniques to address these new demands in the compiler.

 

 

Link

 

The whole article is an analysis of using a compiler to optimize the code generation to reduce the limitations of the SPEs.

Edited June 14, 2006 by alanschu

[mkreku]

The PPC and SPEs has a huge Gflop advantage.

A pet peeve of mine: it's flops. Floating Point Operations Per Second, FLOPS.

 

Or, in this case, Gflops.

 

Now I'll butt out and leave the discussion to alan and Mr. I-Know-Sooo-Much-I-Just-Conveniently-Enough-Can't-Say-Anything-To-You.

[Gorth]

On a not really related, yet humorous note:

 

http://www.winterwind-productions.com/foru...opic.php?t=4297

 

No idea about the authenticity of it.

 

Them Japanese and their English translations

[10k fists]

I do have a relatively technical question, although I'm sure it will highlight my ignorance.  If the SPEs can't predict very well, couldn't the  designers use scripted events to make up the difference?  I mean, that would be irritating in one way, but it could save the AI effort for more meaningful parts of the game.  Also, how would these systems, Wii, 360, PS3, handle a game like FEAR, which claims excellent AI?

<{POST_SNAPBACK}>

 

The SPEs could handle a scripted event fairly easily. I don't think the PS3 or the Xbox 360 will have a problem with FEAR (it's coming to the Xbox 360 already). The PPE can run AI we've already seen with no problem in my opinion. Cell just doesn't have the capabilities to really advance forward in AI. For most people, that's ok, as the AI in games we have today is sufficient enough for them. I and I'm sure other people feel the same, would love to see advances in AI. Cell was designed in a way that would not allow for large improvements.

 

With that said, I really doubt these short comings will come into play, and the majority of games across the two platforms will be identical.

 

 

So you have no problems stating the findings of some people that you talked to (or whatever) regarding the Cell, but you refuse to say that you either know someone working on it, or have an NDA and cannot talk about it because it's too common to see on an Internet forum?

 

<{POST_SNAPBACK}>

 

I have no problem discussing what I know, but I'm not going to reveal why I know it.

Edited June 14, 2006 by 10k fists

[LoneWolf16]

*snip*

<{POST_SNAPBACK}>

That's all I was wondering really. Seemed like you were just taking a shot at the Cell/PS3's capability to make good games....or something.

 

No offense or anything, but it'd be nice of you to post some sort of backing material for your claims, especially about something like this, which is rooted in "cold hard facts", or what passes for facts when dealing with an unreleased product.

[alanschu]

I just thought it was wierd that his reason for not posting even that he has an NDA is because it's too common to see claims of that nature.

[Epiphany]

I just thought it was wierd that his reason for not posting even that he has an NDA is because it's too common to see claims of that nature.

<{POST_SNAPBACK}>

 

I'm under NDA!

[10k fists]

No offense or anything, but it'd be nice of you to post some sort of backing material for your claims, especially about something like this, which is rooted in "cold hard facts", or what passes for facts when dealing with an unreleased product.

<{POST_SNAPBACK}>

 

Sorry, but I'm not going to do it. I'm more than happy to discuss the tech to certain limits, but I'm not going to qualify myself on an internet forum. I'm willing to debate with people that try to dispute what I actually say though. But if someone just doesn't believe me, then that's fine.

 

The problem is, if I qualify myself then my argument loses its ground, and people judge what I say based on my experience in the field. It shouldn't matter whether I'm a high school janitor, one of the engineers behind some of the PowerPC processors, or a game designer.

 

I just thought it was wierd that his reason for not posting even that he has an NDA is because it's too common to see claims of that nature.

 

I said that in jest... I was poking fun at the fact that almost everywhere you go, someone is claiming to have "insider" knowledge, but they're under NDA, so they beat around the bush with vague comments.

[LoneWolf16]

Sorry, but I'm not going to do it.  I'm more than happy to discuss the tech to certain limits, but I'm not going to qualify myself on an internet forum.  I'm willing to debate with people that try to dispute what I actually say though.  But if someone just doesn't believe me, then that's fine.

 

The problem is, if I qualify myself then my argument loses its ground, and people judge what I say based on my experience in the field.  It shouldn't matter whether I'm a high school janitor, one of the engineers behind some of the PowerPC processors, or a game designer.

<{POST_SNAPBACK}>

All well and good, I can respect the attitude, but why give what appears to be factual data without properly backing it with some sort of resource? Been taught that since day one in english classes. Providing proper support for my arguments, that sort of thing. Then again, I obviously never learned that lesson all that well myself.

 

Yes it does matter whether or not you're a highschool janitor or the head of some PC engineering design...thing. I'd think the reason why would be obvious too.