Future CPU performance

[3dilettante]

Fair points, but it is also noteworthy that Cell is about 50% larger in die space and transistors, which explains some of the proportion figures you are using.

When talking about how future games will perform on the two consoles, it doesn't really matter.
Die size doesn't really figure into performance in future titles.

On the other hand Cell does make some significant compromises beyond what even Xenon makes (Local Stores are not coherant like L2 cache and streaming can eat into LS, no branch prediction in SPEs, total chip memory may be large but it is broken into a number of smaller segments, assymetric design, and so forth) which not only have an impact in "getting your head around the architecture" but it isn't always a performance win for every sort of processing task.

To clarify my position, I'm only looking at the competition between the two consoles.

MS and Sony have targeted their processors for the same workload, one which is heavy on multimedia and glitzy effects.
One of the processors in question is simply better at most of that work.
If Cell is terribly hard to program for, the launch titles seem to show that failure to utilize it well leaves it at rough parity with the Xbox 360.

There are areas where Cell absolutely rocks (or can/will), especially when you can throw a lot of SPEs at the problem and it scales fairly linearly and can fit into the Local Store, but from what I have heard this isn't nearly always the case. There is huge potential with Cell but not every problem is SIMD or easily solved (at this point, and this has been an area of work for decades) with gross parallization.

Too bad Xenon is relying on similar assumptions that Cell is working from. The reduced die budget, coupled with similar design targets, means that Xenon is trying to do the same job as Cell with far less.
Cell overall is a powerful, if sometimes awkward chip. Xenon is a good chip, with somewhat less awkwardness.
Both rely on in-order narrow cores that are individually unimpressive.
Cell does more to make up for that than Xenon.

There are other concerns besides raw performance that can blur the picture, but none of them are insurmountable. In the end, Xenon's well will run dry first.

 

[mesyn191]

"If Cell is terribly hard to program for, the launch titles seem to show that failure to utilize it well leaves it at rough parity with the Xbox 360."

The way I look at is those devs had almost a year more to refine their work on hardware that was much closer to final spec than what X360 devs had (and IIRC, was actually identical feature wise as far as the CPU and GPU were concerned...), given how much we've all been told about how Cell stomps all over Xenon in most all tasks by large (ie. 30-50%+) margins, the fact that their end results are only roughly on par vs. X360 counterparts is a MAJOR let down.

These games should be mopping the floor with X360 games, if not in graphics alone, then other ways like physics or AI, which are work loads that are supposedly embarrassingly parallel in nature (or can be made so) and so should be a relative cake walk to get working and working well on Cell.

IMO if we don't see a widespread (IOW, not just the AAA 1st party titles that'll have the huge budgets to do almost whatever they please, but smaller but still competent devs too) difference between PS3 and X360 games equal to or greater than the difference between Xbox and PS2 games than the PS3 is a failure and Cell a dead end for use in gaming systems.

 

[Shifty Geezer]

"If Cell is terribly hard to program for, the launch titles seem to show that failure to utilize it well leaves it at rough parity with the Xbox 360."

The way I look at is those devs had almost a year more to refine their work on hardware that was much closer to final spec than what X360 devs had (and IIRC, was actually identical feature wise as far as the CPU and GPU were concerned...), given how much we've all been told about how Cell stomps all over Xenon in most all tasks by large (ie. 30-50%+) margins, the fact that their end results are only roughly on par vs. X360 counterparts is a MAJOR let down.

By the same rationale, XB360 has such amazingly dev-friendly tools that the ability for XB360 developers to tap the hardware is far easier. Thus though PS3's devs have had more time, the XB360 devs have managed the same work but in less time...

 

[3dilettante]

"If Cell is terribly hard to program for, the launch titles seem to show that failure to utilize it well leaves it at rough parity with the Xbox 360."

The way I look at is those devs had almost a year more to refine their work on hardware that was much closer to final spec than what X360 devs had (and IIRC, was actually identical feature wise as far as the CPU and GPU were concerned...), given how much we've all been told about how Cell stomps all over Xenon in most all tasks by large (ie. 30-50%+) margins, the fact that their end results are only roughly on par vs. X360 counterparts is a MAJOR let down.

How let down people are is irrelevant to the discussion of which CPU has more long-term potential.

The 360 has been out for a year, something that has been a major boon to the platform. The big games coming out for the 360 cannot be considered launch titles a year after the launch.

Microsoft is counting on its lead-time to keep its software a generation ahead of Sony's. I argue that even if this is the case, each successive generation will not yield the same amount of improvement there was over the last. With time, Cell will demonstrate it is technically superior.

Other factors will determine whether that really matters from a business standpoint.

These games should be mopping the floor with X360 games, if not in graphics alone, then other ways like physics or AI, which are work loads that are supposedly embarrassingly parallel in nature (or can be made so) and so should be a relative cake walk to get working and working well on Cell.

I'm not sure it's safe to say any task escapes the learning curve needed to program for Cell. With time, more developers will gain the necessary knowledge.

IMO if we don't see a widespread (IOW, not just the AAA 1st party titles that'll have the huge budgets to do almost whatever they please, but smaller but still competent devs too) difference between PS3 and X360 games equal to or greater than the difference between Xbox and PS2 games than the PS3 is a failure and Cell a dead end for use in gaming systems.

That's something determined by factors outside of the technical aspect. I honestly don't care either way about how much money either console loses or makes, or what the market decides. MS went for the lead time, which may gain it significant market share.
Sony went for an expensive very high-end device with price, lack of titles, lack of BR titles, and some design issues external to Cell that has limited the buying market at present.

Most games will likely be close, because multiplatform projects will keep their versions roughly the same.

These issues are external to what I'm saying, that the AAA exclusives (I'm assuming there will be more than 1) will have more to offer on the PS3 from a technical aspect. Whether the gameplay sucks or if it is a critical success, a huge seller or a market slowpoke is none of my concern.

From a technical aspect, it is those in the PS3's camp will have the most to work with.
Assuming that programmers don't become retarded in the next 5 years, that means they will make use of the CPU's sizeable advantages by then.
Those working on Xenon will be doing the same, but once utilization becomes relatively close on both platforms, one has more raw performance.

 

[Sethamin]

Microsoft is counting on its lead-time to keep its software a generation ahead of Sony's. I argue that even if this is the case, each successive generation will not yield the same amount of improvement there was over the last. With time, Cell will demonstrate it is technically superior.

You say that as if it's a good thing. The fact that it'll take several years to unlock Cell's potential is not a positive and means that only the last generation of games on the PS3 will be able to fully take advantage of the CPU's abilities. Furthermore, it also means that developers will be hard at work for the next several years trying to figure out the best methods to unlock that potential, time that could be better spent implementing new features, effects for their artists, or tweaking gameplay and mechanics.

I would argue that a crucial measure of an architecture's strength is its ability to get out of the developer's way. In an ideal world, a dev could write all the code they want and it runs without any performance issues the first time through. In reality, all sorts of adjustments are necessary, from tweaking various data and code variables to reduce cache misses all the way up to writing whole schedulers from scratch to spread their tasks across multiple threads and cores. But the amount of optimization needed is certainly not fixed. And developers and their time are neither free nor cheap.

All that being said, I want to make clear that I don't disagree with you. The Cell most certainly has far more potential than the XeCPU. To be fair, there are certain types of tasks that XeCPU might actually do better - namely, branchy code, although it still won't do great at that since it's an in-order processor. But realize in the end that the CPU, just like all hardware, is just a tool (or, if you prefer, a means to an end), and not the end all-be all. It's how you use it that matters. For those obsessed with technical specifications and extracting every FLOP of power, this will seem like heresy. But I'd far rather have shorter dev cycles with a lot more emphasis on art direction and style than just trying to pump out raw polygons.

 

[3dilettante]

You say that as if it's a good thing.

I am not. I'm simply reiterating the truth of diminishing returns. The 360 is further along in its curve, and it has less peak performance than Cell.

I'm simply saying that I think that enough of Cell's difficulties will be compensated for, and that future generations of software will use it more effectively.
Should both Xenon and Cell be used to the same proportion, Cell wins out from a performance standpoint.

Odds are, that will be clear in a few years.

The fact that it'll take several years to unlock Cell's potential is not a positive and means that only the last generation of games on the PS3 will be able to fully take advantage of the CPU's abilities.

That has been the case for all platforms. It's hardly an indictment of Cell. Whether it will prove too difficult to effectively harness is a fair question, as for all of its supposed revolutionary features, it is a revisiting of some heterogenous DSP systems that were tried and discarded decades ago.

I still feel that there's a very good chance a sizeable pool of talent and tools will develop to get decent utilization.

Even if not, it's still a fair question whether Xenon can manage to win, even in that case. Its design philosophy isn't as aggressive as Cell's, but they share commonalities.
If Cell has difficulties with some tasks, it is not guaranteed that Xenon will win all that decisively either.

On the other hand, anything Cell does well will leave Xenon far behind.

Furthermore, it also means that developers will be hard at work for the next several years trying to figure out the best methods to unlock that potential, time that could be better spent implementing new features, effects for their artists, or tweaking gameplay and mechanics.

And those that learn relatively fast and well will produce (technically) stunning games.

Those with less heavy hardware will learn faster, reach the limit and produce games that are less so from a technical standpoint.

That doesn't mean that the more technically advanced games will be better to play, or that they will sell more.

It also doesn't mean that a developer that spends extra time early on learning Cell can't put out a game that is amazing to play.

The Xenon core is not identical to the PPE. We don't have enough info disclosed on the Xenon to say if it is better overall than the PPE. But we do know Xenon cores have 128 VMX registers each, while the PPE only has 32. This should reduce the need to go to memory/cache as often when running vector heavy code.

(edit: never mind this Your data is old in this regard. The PPE's VMX unit was upgraded with the second revision.)

As to width, I misremembered the SPE's issue restrictions as being mostly scalar with some restricted dual-issue. If it is more general, this only reinforces my overall point, since Xenon's issue width per core would not create as large a differential.

The rest of what was said is pretty much what I was saying, that Xenon--if fully tasked with emulating the behavior of the SPEs, could dual-thread to match six of them in terms of the number of independent instruction streams being run. The cost is that each thread only gets a fraction of the core it is on.

 

[Crossbar]

Your data is old in this regard. The PPE's VMX unit was upgraded with the second revision.

Are you saying the VMX unit of the Cell PPE has more than 32 registers or what upgrade are you refering to?

 

[3dilettante]

Are you saying the VMX unit of the Cell PPE has more than 32 registers or what upgrade are you refering to?

I can't back that up, my memory of is too hazy these days. I reviewed some coverage of the DD2 revision of Cell, which said it added vector resources, but not that it increased the register count.

 

[mesyn191]

By the same rationale, XB360 has such amazingly dev-friendly tools that the ability for XB360 developers to tap the hardware is far easier. Thus though PS3's devs have had more time, the XB360 devs have managed the same work but in less time...

Nah, the tools make it easier to do a few things, but a lot of stuff is still beyond the grasp of a compiler at this point in time and must be done laboriously by hand. When all is said and done Xenon (X360) isn't really developer friendly, just more straight forward to work with compared to Cell, and all the tools the world has to offer can't change that.

 

[inefficient]

The April 06 Cell BE handbook says 32 VMX per thread.link

I'm almost sure you are wrong about the number of instrucion tthe vmx128 unit can issu in one cycle.

I've read the decoupled fp/vmx unit as its own queue list for instructions and can issue two instructions per cycle.

It seems you are right you can dual issue vector/fpu instructions as long as you follow the rules and only issue Type1 with Type2.
Type 1: VXU simple, VXU complex, VXU floating-point, and FPU arithmetic instructions.
Type 2: VXU load, VXU store, VXU permute, FPU load, and FPU store instructions.

 

[mesyn191]

How let down people are is irrelevant to the discussion of which CPU has more long-term potential.

You can have all the potential performance in the world but if its never put to use or used properly for one reason or another then who cares? In the end potential means nothing if it goes unused.

The 360 has been out for a year, something that has been a major boon to the platform. The big games coming out for the 360 cannot be considered launch titles a year after the launch.

I think given amount of time those PS3 developers had to work on them with dev kits that were far better than what X360 devs had for launch games and that they actually had some help from Sony in the form of better (vs. PS2 release) documentation and tools that they sure are comparable.

Microsoft is counting on its lead-time to keep its software a generation ahead of Sony's. I argue that even if this is the case, each successive generation will not yield the same amount of improvement there was over the last. With time, Cell will demonstrate it is technically superior.

Well duh, given all the Cell hype and the cost and how much later it came out it damn well better. But will it be superior enough to show a significant (ie. equal to or greater than Xbox vs. PS2 graphics quality in games) difference in graphics/AI/whatever vs. X360 games without forcing devs to either have massive budgets and/or significantly longer development time and/or cut content out of their games? Thats my question, and personally I'm very skeptical as to wether it is that powerful and if it isn't than given then amount of money, time, and effort they put into Cell I'd call it a failure.

I'm not sure it's safe to say any task escapes the learning curve needed to program for Cell. With time, more developers will gain the necessary knowledge.

Oh come on, if embarrassingly parallel tasks are that difficult to get working on a arch. like Cell then they've got some real problems, problems which have no easy answers despite decades of work by people who had much more human resources and with FAR larger budgets than a lot of these game developers have to work with.

That's something determined by factors outside of the technical aspect. I honestly don't care either way about how much money either console loses or makes, or what the market decides. MS went for the lead time, which may gain it significant market share.

I think you misunderstood or I'm misunderstanding you, I was talking about differences in graphics between games on the 2 platforms, not market share, etc.

These issues are external to what I'm saying, that the AAA exclusives (I'm assuming there will be more than 1) will have more to offer on the PS3 from a technical aspect.

Oh I'm sure there'll be more than 1 too, perhaps 5 or 6 AAA titles that really try to do everything possible to get as much out of the PS3 throughout its lifetime, but if that is all there is due to the difficulties in wringing performance out of Cell than all that potential is going to be spending most of its time being wasted.

From a technical aspect, it is those in the PS3's camp will have the most to work with. Assuming that programmers don't become retarded in the next 5 years, that means they will make use of the CPU's sizeable advantages by then.

Oh come on now, they're retarded if they can't make practical use of much of Cell's performance in that time frame?

 

[inefficient]

...

The PS2 was worse. People claimed it would never be fully utilized since the design seemed so totally alien. But people did eventually come to grips with it and amazing stuff was produced.

One could argue that the fact that games got steadily better visually from mediocre to great was good for the platform as it gave consumers something to look forward to and they kept coming back because it always felt fresh,

I see "it has room to grow" as exciting. How boring if just a year into a consoles life cycle you already seen it at peak potential and then all subsequnt releases are only on par. And it's not like PS3 games look bad now. We just have a lot more to look forward to. Exciting times ahead.

 

[Crossbar]

I think given amount of time those PS3 developers had to work on them with dev kits that were far better than what X360 devs had for launch games and that they actually had some help from Sony in the form of better (vs. PS2 release) documentation and tools that they sure are comparable.

So you are basically telling us that having final dev kits a year before the competitor means nothing?
Please elaborate, do you think Sonys sdk:s are comparable to Microsofts?

BTW If we look at the Edit: Alfa kits I think MS started handing out PowerPC based dev kits in the beginning of 2004 according to this. and Sony started handing out Cell based kits in the late 2004 and beginning of 2005 according to this.

 

[Titanio]

There are always compromises depending on platform strengths and weaknesses, but the specific things you mention are things already done on year old titles (e.g. large scale armies in games like Kameo and N3, cloth simulation in NBA 2K6, and so forth).

I think he was talking about having a multiplicitly of such tasks running all at the same time in one title (in each frame)...I don't know at all if the specific mix he was talking about w.r.t. HS represents a good example of something that could be done on Cell that would be difficult on Xenon, but I don't think it's beyond the pale to say more generally that there is more potential for doing more simultaneously, or more total per-frame on Cell than on Xenon. To take one of the tasks he mentioned..say, hair or cloth simulation, without having a Xenon benchmark on me but looking at the results achieved by others with similar stuff on SPUs vs other processors, it seems fairly likely to me that a SPU would probably perform about as well (if maybe not better) than a Xenon core with that kind of work. If that were true, to dedicate one third of your processing power to that on Xenon might seem expensive compared to less than one sixth on Cell..you've a lot more left over for doing lots of other things than on Xenon in that kind of scenario. If you have two such "SPU champion" tasks being worked on over the entirity of the frame, then you're left pitting one Xenon core against a PPE and 4 SPUs. Even if the rest of the tasks in the mix are more suited to a Xenon core than a SPU on an individual basis, I think the sheer brute force and numbers left still available with the latter makes it difficult to say that you could not be doing an awful lot more with those remaining processors above and beyond the single remaining Xenon core. If you find still more SPU champions, well..it could be embarassingly more.

 

[mesyn191]

So you are basically telling us that having final dev kits a year before the competitor means nothing?
Please elaborate, do you think Sonys sdk:s are comparable to Microsofts?

BTW If we look at the Beta kits I think MS started handing out PowerPC based dev kits in the beginning of 2004 according to this. and Sony started handing out Cell based kits in the late 2004 and beginning of 2005 according to this.

But MS didn't have final dev kits a year before Sony, they had beta dev kits which were significantly different from the final console hardware up until very close to launch IIRC almost all work was done on those kits and only a few key developers got final kits at all before launch.

From comments on these forums Re: their respective SDK's MS's are the gold standard by which all others are judged while Sony's could only be called minimalist, but time counts for alot.

 

[mesyn191]

The PS2 was worse. People claimed it would never be fully utilized since the design seemed so totally alien. But people did eventually come to grips with it and amazing stuff was produced.

IIRC that main issue with the PS2 was that at launch Sony provided it to the developers with almost no documentation and/or tools and it was very very different from what they were used to working with. Also people were right, it never was fully utilized...

One could argue that the fact that games got steadily better visually from mediocre to great was good for the platform as it gave consumers something to look forward to and they kept coming back because it always felt fresh,

This was true with the Xbox too though wasn't it, and wasn't the Xbox one of the most if not the most developer friendly and easy to work with console of that generation? Sure, it came out a year later but IIRC it cost about the same as the PS2 did while having something to show for it (significantly more powerful and versatile GPU with clear results that you could point to).

I see "it has room to grow" as exciting. How boring if just a year into a consoles life cycle you already seen it at peak potential and then all subsequnt releases are only on par. And it's not like PS3 games look bad now. We just have a lot more to look forward to. Exciting times ahead.

Exciting and suprises are both very bad things when you have a budget and a dead line to work in, they have to make money you know. They are after all businesses, not elitist clubs for coders and academics, and innovation is not something they do very much of or well at all.

 

[Crossbar]

But MS didn't have final dev kits a year before Sony, they had beta dev kits which were significantly different from the final console hardware up until very close to launch IIRC almost all work was done on those kits and only a few key developers got final kits at all before launch.

From comments on these forums Re: their respective SDK's MS's are the gold standard by which all others are judged while Sony's could only be called minimalist, but time counts for alot.

Yeah, the final PS3 hardware shipped less than a year after the 360, the sdk:s shipped a few months later than the hw and the final spec of the hardware (frequencies etc.) was locked late last summer IIRC about a year after the final 360 hw started shipping.

Some developers get shitty shipment of dev kits, we've heard those storys about both MS and Sony, nothing new there.

Anyhow 8, 10 or 12 months, it is still a lot more time with final hardware and I am pretty sure it makes a lot of difference together with Microsofts "gold standard" sdk:s.

Don´t you think it is pretty likely that the PS3 hardware is less exploited by the launch titles than the 360 hardware is with its current titles?

 

[mesyn191]

Yeah, the final PS3 hardware shipped less than a year after the 360, the sdk:s shipped a few months later than the hw and the final spec of the hardware (frequencies etc.) was locked late last summer IIRC about a year after the final 360 hw started shipping.

Some developers get shitty shipment of dev kits, we've heard those storys about both MS and Sony, nothing new there.

Anyhow 8, 10 or 12 months, it is still a lot more time with final hardware and I am pretty sure it makes a lot of difference together with Microsofts "gold standard" sdk:s.

Looks like it was closer to 3 months before the X360 launched that final X360 SDKs were available to a limited number of developers, beta kits became available around June 2005 it seems but work was still being done on the alphas around that time period too.

http://xbox360.ign.com/articles/655/655273p2.html

Don´t you think it is pretty likely that the PS3 hardware is less exploited by the launch titles than the 360 hardware is with its current titles?

Not significantly no.

 

[Crossbar]

Not significantly no.

Why? You say that 360 developers had final hardware 8-10 months longer than PS3 developers and they have significantly better sdk:s.

Do you think PS3 developers have more skills or what do you base your assumptions on?

 

[inefficient]

Exciting and suprises are both very bad things when you have a budget and a dead line to work in, they have to make money you know. They are after all businesses, not elitist clubs for coders and academics, and innovation is not something they do very much of or well at all.

You have a pretty bleak view of it all. Surprises might not good, but I didn't say surprises. I said exciting!

A team that is not excited about the game they are making is a team that is just going through the motions and will put out crap they don't care about. Or will put in just in enough check box features until they can say their product matches or betters the last game in the genre that came out.

I remember Cliffyb saying in an interview recently something to the effect that his main job on GoW was just keeping the excitement level up. And pitching ideas that the programmers and content creators would get excited about.

Of course before sinking a lot of money into something new you will want to prototype it to make sure your core ideas are feasible. This takes time but does not necessarily have to be really expensive. And even if the prototype turns out to be a throw away, the lessons learned are not throw away.