Was Cell any good? *spawn

[3dilettante]

And looking only at the picture through the prism of the ps3, say if for example the ps3 would have been a 4 cores Xenos (mostly the same silicon budget as the cell) and the RSX, Sony could have afford more memory. I believe that twice the main ram would have made more of a difference (visible than more than a handful of games) than whatever the SPU brang on the table.

The homogenous multicore would have been closer to what IBM wanted when it was working with Sony and Toshiba, with Toshiba apparently more in love with the SPU concept. This seems borne out by the SPURS product that barely made a blip anywhere.

Where's the math coming from stating that having a 4-core Xenos would make doubling the system RAM affordable?
The choice to skip the SPU may have meant no Toshiba as a contributing design partner, for better or worse.

 

[Billy Idol]

Sorry guys but Aaron (and not only him by the way)has won the point.

Huh, I do not understand what you are saying here...which point exactly does he won? That CELL is worse a CPU than Xenos with respect to raw (real achievable) performance? That is the first time I hear this on this forum, typically all dev state the exact opposite! All experience and observation this gen point us to the conclusion that the PS3 has the better CPU compared to Xbox360.

With respect to the point if CELL was a good investment and a success...well that is something completely different...I only know that in HPC, it was not worth it in my opinion, I could not profit from CELL and at the end, there was no point in going the CELL route (which at the end turned out to be indeed the correct thing to do), as there is at the moment in my opinion no real point in going the HPC-GPU route...

 

[liolio]

The homogenous multi-core would have been closer to what IBM wanted when it was working with Sony and Toshiba, with Toshiba apparently more in love with the SPU concept. This seems borne out by the SPURS product that barely made a blip anywhere.

Where's the math coming from stating that having a 4-core Xenos would make doubling the system RAM affordable?
The choice to skip the SPU may have meant no Toshiba as a contributing design partner, for better or worse.

Well it's a bet of mine that DDR2 is cheaper than both GDDR3 or xdram. But it sounds like a safe bet.
Xenon goes away without much more bandwidth and most likely horrid latencies to access the main ram.


Anyway I don't think either that either the Cell and the RSX was the way to go for Sony, I gave my pov on the matter in the proper thread, I will prevent my self to go further down this road.

 

[liolio]

Huh, I do not understand what you are saying here...which point exactly does he won? That CELL is worse a CPU than Xenos with respect to raw (real achievable) performance? That is the first time I hear this on this forum, typically all dev state the exact opposite! All experience and observation this gen point us to the conclusion that the PS3 has the better CPU compared to Xbox360.

With respect to the point if CELL was a good investment and a success...well that is something completely different...I only know that in HPC, it was not worth it in my opinion, I could not profit from CELL and at the end, there was no point in going the CELL route (which at the end turned out to be indeed the correct thing to do), as there is at the moment in my opinion no real point in going the HPC-GPU route...

Well if you don't understand is point I don't know what to do, it's pretty clear to me.
It was a bad design and was called out as such by plenty of people (not me real hardware or software engineers/ architects) and they got it right it went no where.
He thinks the investment was pointless (and you can track his pov for a long while if you want) history prove him right.
The ps3 was significantly more expansive than the 360 both in BOM and R&D, it ended in the majority of the case below or on par this its competitor no matter the cell.

As for a straight comparison between the Cell and Xenon it is quiet pointless, Xenon is significantly tinier and there was quiet some less money pored into the design.

Imho it's not a matter of redoing close to 8 years after launch a console dick contest, when it's all say and done he is right. Money would have undisputed been better invested elsewhere (vs the 360 it could be extra core, more ram, etc.).

EDIT
I do get that he states his pov in a strong way but I definitely understand the stance. It's like if you are doing some maths and make a mistake not a stupid one but one that as that show your reasoning is wrong. That the case with the Cell people called the very idea behind the design as "wrong", Aaron is among them. From his hardware engineer(architect?) pov it must look indeed like a big mistake, pointless investment or whatever you call it and there is not saving it or finding it some merits. Pretty much like an OT, there is no saving it, I remember my day at school in discipline like philosophy if you don't get the subject properly, no matter you filed your work with good references, the plan for your work is well set out, the stuff is well written you were still receiving a really low grade (like 5 at best out of 20).

 

[forumaccount]

Nah, that's nonsense. For instance ever since Havok 4.5, Havok runs quite a bit better on PS3. You could tell occasionally that the PS3 performed better in there in multi-platform titles, but generally mp devs have kept physics relatively simple in order to cater for the weakest platform (as with everything, obviously also in PS3's favor plenty of times). A game like Motorstorm went pretty wild with the Havok stuff though, and at the time of that games release, Havok ran better on Cell than almost any other CPU other there (that would run games that have Havok).

It may not have been a great investment, but if you would replace it with something like the 360's CPU and leave everything else the same, I think the PS3 would have been much worse off.

Havok isn't a game, it's an individual system that pretty much hits the exact niche that Cell works best in.

Mature engines with 4-5 years of resources put into them can run faster on Cell, it'd be dumb to dispute this. But people don't understand the time it takes to get there, and "it's difficult to program" is a poor assessment of why it takes time to get there.

The question was if a better GPU would make Cell a better investment, not whether a different CPU would make PS3 a worse console. The answer is no, a simpler CPU keeps up well enough when paired with a decent GPU to not make any noticeable difference in game quality as we've seen.

 

[Arwin]

Motorstorm is very much a game and strongly reliant on Havok tech - there's a stupendous amount of cars, moving physics objects and moving track components for which proper collision is handled. And I am willing to bet something like Super Stardust HD's thousands of colliding objects at full 1080p @ 60fps is going to be pretty hard to pull off in any other configuration of current gen hardware components.

Your other arguments are moot, as I've already agreed with them - the investment vs reward for many wasn't enough, even if this was frequentky skewed by a lack of programmers for the PS3 version to begin with (there have been very many stories of just one guy responsible for all the PS3 version of a multi-platform engine). However that does not negate various cool stuff developers have been able to do with Cell that few other CPUs would have been capable of.

Of course this discussion is as old as the console generation by now, and compared to previous threads, any other type of thread then one that just accumulates all arguments, known facts, quotes from developers and so on at this point is fairly pointless.

 

[Sonic]

If one is arguing that CELL is a failure in the marketplace outside of PS3, then that is a given. It was a hyped up technology that couldn't deliver. Much like Larrabee, even though that lives on in other forms.

 

[archangelmorph]

Ain't that costed Sony a lot of work? (all the library?)
Sorry guys but Aaron (and not only him by the way)has won the point. Plenty of not only software but hardware people has vouched the Cell a failure or more a dead end at launch even before release.
They've been proved right, it was a wrong solution for the problem they wanted to tackle. It's dead now along with this approach.

And looking only at the picture through the prism of the ps3, say if for example the ps3 would have been a 4 cores Xenos (mostly the same silicon budget as the cell) and the RSX, Sony could have afford more memory. I believe that twice the main ram would have made more of a difference (visible than more than a handful of games) than whatever the SPU brang on the table.

The CELL certainly was a failure both from a business & strategic perspective...

However as a specific piece of technology built to solve a particular problem, it was far from it.

Sony made the best of it & in the long run it worked out for them, offering the only positive differentiator the PS3 ever had outside of Blu-ray which is by no means an insignificant data point. However the strategic decision on whether or not to continue with the architecture I suspect had far more to do with economics, cost, investment & ROI than it being "hard to develop for" which seems to be the typical definitive reason constantly touted around here & the rest of the internet.

I can pretty much guarantee that if the CELL hadn't cost Sony as much as it did, or had it of worked out better as a more suitable uniform solution across a wider range of Sony's product line, they would have continued to invest heavily in the architecture & it would still be moving forward today, regardless of how many crappy Xbox360-PS3 ports there were or how many times Gabe Newell & John Carmack complained about it.

 

[Shifty Geezer]

Well if you don't understand is point I don't know what to do, it's pretty clear to me.

I'm reading something ambiguous in the origin statement about PS3's CPU advantage, so the point is unclear. In his reply that PS3 had no CPU advantage, is Aaronspink saying that the PS3 system had no advantage over XB360 even though it had a more capable CPU? That's a point many would agree with. Or is he saying PS3's CPU didn't offer an advantage over XB360's CPU? That idea is a head scratcher and the one people are reacting to. Surely the PS3's CPU offered considerable performance advantage over Xenon, and is all that's propping up the PS3 platform as a whole. If the Cell is no better than Xenon, and RSX is way worse than Xenos, then how on earth can PS3 manage to be competitive? The only way would be if devs are severely underutilising XB360. Hence the general consensus at this point that:

Xenos > RSX
Cell > Xenon

The debate about the platforms is then over whether XB360 > PS3 or PS3 > XB360. But everyone at large is happy to accept XB360 had a GPU advantage over PS3, and PS3 had a CPU advantage over XB360, hence the wide-eyed response to Aaronspink's remark that PS3 had no CPU advantage.

 

[aaronspink]

Sorry, I'm still a bit confused by this, because I wasn't asking about Xenos.

Putting it another way, you are saying that Cell doesn't outperform Xenon, XB360's CPU, in any way, and Xenon vs. Cell will be an even match in most gaming scenarios?

Sorry, got the names switched.

No I'm not saying that Xenon and Cell would be an even match in most gaming scenarios. I'm saying that Xenon is BETTER than Cell in most gaming scenarios. Given the amount of resources required to get CELL performing in parity with Xenon, you could get significantly more performance out of Xenon or shit significantly earlier. Cell provides a lot of peak performance, but the realized performance is really no better than Xenon, it just requires a lot more work.

If Sony had gone with 3-4 PPU design we would of had overall better games and more games this generation.

 

[aaronspink]

No they didn't... They "had" 1 dual-issue PPU & 6 SPUs...

The vast majority had 1 PPU.

There's no way in hell you'd get similar speed ups with Xenos if you spent "vast" amounts of resources on optimizing CPU code...

Yes you could. People do it every day on friggin Intel x86s. You know, processors that are an order of magnitude easier to program for than the PPU which is an order of magnitude easier to program for than an SPU.

Xenos did what developers "needed" but nothing substantially more. If you wanted to use Xenos to make God of War III or Uncharted II you couldn't because it just doesn't have the raw vector processing power nor the available parallelism to be up to the job...

If it did what developers needed than it did everything that was needed.

And GoW3 and Uncharted 2 could of easily been done on 360.

 

[aaronspink]

Huh, I do not understand what you are saying here...which point exactly does he won? That CELL is worse a CPU than Xenos with respect to raw (real achievable) performance?

Anyone caring about raw performance is caring about the wrong thing. It is like arguing that the WWII German rail guns are the most powerful artillery that ever existed even though they were basically unusable.

 

[aaronspink]

If the Cell is no better than Xenon, and RSX is way worse than Xenos, then how on earth can PS3 manage to be competitive? The only way would be if devs are severely underutilising XB360. Hence the general consensus at this point that:

They are spending significantly more time optimizing for Cell than Xenon because they have no other choice. Given equal levels of optimization for both platforms, Cell will almost always come out behind.

But everyone at large is happy to accept XB360 had a GPU advantage over PS3, and PS3 had a CPU advantage over XB360, hence the wide-eyed response to Aaronspink's remark that PS3 had no CPU advantage.

The reality is that dev have to do a lot more optimization for PS3 both in CPU and GPU in order to reach the same performance. For RSX vs Xenos it is more of a wash as Xenos has it own issues with the tiling and such.

 

[Brad Grenz]

You keep saying "alot" and "significantly more" without quantifying those terms. I'd guess you have a pretty exaggerated sense of how great that disparity was and the overall financial impact that actually has on a modern AAA game. Having 3 guys work 50% more on Cell optimizations than they do on Xenon is negligible in the overall accounting for a game where you have 200 other guys pumping out art assets for 3 years. And since you wrongly assume there is no ceiling to what can be accomplished on Xenon with additional optimizations, it's hard to take you seriously.

 

[Shifty Geezer]

You keep saying "alot" and "significantly more" without quantifying those terms. I'd guess you have a pretty exaggerated sense of how great that disparity was and the overall financial impact that actually has on a modern AAA game. Having 3 guys work 50% more on Cell optimizations than they do on Xenon is negligible in the overall accounting for a game where you have 200 other guys pumping out art assets for 3 years.

He's talking relatively to the other platform; if another 3 guys also spent the same 50% more on Xenon optimisations, XB360 would have performed much, much better. What aaronspink is saying is that Cell has only added a necessary cost and provides no better peak performance. That's a significant claim and one it'd be good to (dis)prove somehow, although I don't know how that could be achieved. At this point I suppose we can only count on the opinions of devs who worked on both to equal measure. In theory one should see the benefits of optimised Xenon+Xenos in XB360 exclusives, but those sorts of comparisons are hard to make and practically impossible to make without supposed platform allegiances colouring the discussion.

 

[Arwin]

Can anyone remember ever hearing about any developer that actually spent more time on Cell than PC/360? From what I've heard over the years it is usually one single developer responsible for Cell optimisations, and with the engine either developed on PC or 360 in the first place, it's hard to compare the two.

Definitely at the beginning of the generation though the 360 came with vastly better and easier development tools than the PS3. That eventually improved a lot on PS3, as far as I've read.

 

[Billy Idol]

Anyone caring about raw performance is caring about the wrong thing. It is like arguing that the WWII German rail guns are the most powerful artillery that ever existed even though they were basically unusable.

Yes, I understand this Aaron, I know that this is important. That is why I am talking about real achievable performance. It is just surprising what you are saying, because the last 5 years we heard devs all over the place stating it differently. If I remember correctly, e.g. John Carmack and the Crytek devs stated that CELL is a little bit better CPU.

Are you aware of a specific situation (probably you can even state a game or so) where CELL would be no match to Xenon? I once read that in particular AI is rather difficult on CELL? But unfortunately I do not remember where I read it and the reasons why, if it is even true...

Problem is, that it is really difficult to compare different architectures, different engines and different games. We also had same discussion in our HPC group back than. With the amount of effort you have to put into CELL to make it run, I decided that it was simple not worth it to switch regarding the potential performance...

@liolio: I fully agree with you and everyone else about CELL not being a success.

 

[archangelmorph]

The vast majority had 1 PPU.

No they used 1 PPU... Early in the generation... I can categorically state with a pretty firm degree of certainty, that nobody has shipped a game on PS3 that only uses the PPU for at least the past 2-2.5 years... Sony's APIs & toolchain have matured leaps and bounds since the consoles' inception & therefore doing that would (be somewhat retarded &) mean not using the tools provided that handle most of the baseline heavy lifting & SPU work so you don't have to..

Yes you could. People do it every day on friggin Intel x86s. You know, processors that are an order of magnitude easier to program for than the PPU which is an order of magnitude easier to program for than an SPU.

Either you didn't read/understand my response properly or you don't know what you're talking about. I'll favorably assume the former...

If it did what developers needed than it did everything that was needed.

Sure but did it do everything developers wanted? No... Not by a long shot.. Ask the Splinter Cell: Conviction engine boys about that...

Could CELL? certainly in a much greater capacity than Xenon, albeit it will the requirement that it would take longer than if you implemented the same code on an 8-core modern Intel chip. but then no console developer had one of those at the time so it's a moot point...

And GoW3 and Uncharted 2 could of easily been done on 360.

Take it from someone who has worked directly on PS3, attended many technical talks from the Sony/Naughty Dog guys on the work they did in these games and what it took to achieve what they were able to accomplish. You're simply quite wrong.

Finally aaronspink, I think you need to learn/understand/comprehend that writing code so close to the hardware in ways that the architectural differences between Xenon & CELL really matter is nothing more than a sunk cost...
For the vast majority of developers today, that cost has been spent... The work paid for & development moved on...
So any further argument revolving around the CELL's "hard to program for" rhetoric is moot. It simply isn't anymore because there's a far greater knowledge base on how to make the most of the architecture now than there ever has been.

Everyone else has moved on from this idea.. Maybe you should too...?

 

[Billy Idol]

One of the good things due to CELL: game devs had to come up with new cool strategies and algorithms, most prominent being the postprocessing antialiasing (if I am right introduced in the Saboteur PS3 version and of course in God of War 3). That is a good thing in my book, and especially post processing antialiasing had a lot of impact...

 

[Shifty Geezer]

The post processing AA derived from MLAA was pioneered by Intel AFAIK. And if the GPUs were more potent, we wouldn't need post-FX AA. But it is a useful addition to the developer toolset, I'll grant. Not sure that Cell really helped with that.