Was Cell any good? *spawn

[Shifty Geezer]

Huh? Cell was a piece of ¤&#/. It's 5 years after launch, and still almost no games actually use any of that theoretical horsepower. At this point it is accurate to say that PS3 had a slow single-core processor with some extra silicon put next to it for marketing reasons.

Extremely inaccurate, and ignorant. Have you not seen the powerpoint slides from the likes of Guerilla showing SPE usage? The best games on PS3 are thrashing the SPEs, and that's the only reason it can hold its own this gen. Plenty of criticism and debate can be had about PS3's overall design and the value of Cell given difficulty in using it effectively, but Cell did carry the platform as it was expected to.

 

[tunafish]

Extremely inaccurate, and ignorant. Have you not seen the powerpoint slides from the likes of Guerilla showing SPE usage?

Yes. Do you realize why the slides are relevant? Because, even after 5 years, it's notable that someone uses the SPEs at all. The SPEs were supposed to give the PS3 a major edge -- and yet, in vast majority of all the PS3 games they are not being used. For them to be worth it, it's not enough for someone to finally use them. They were expensive. The 150mm² of die area that was used for them could have been used for 2 more real cores, more eDRAM than 360 has, doubling the GPU power of PS3 (which would have made it not need insane SPE tricks to keep up with 360), or, to drive home their pointlessness, it could all have been turned into about 4MB of cache, which would have been vastly more useful than the SPEs were.

The best games on PS3 are thrashing the SPEs, and that's the only reason it can hold its own this gen.

The best games are mostly using the SPEs to augment the piss-poor GPU that PS3 has, or for tasks that could likely be ran faster on proper cores with real load-store units. And again, my point is not that nobody uses them. If you've ever had a look at some cool demo compos, you know that there are truly insane and gifted people out there who wrest the very last of power from any architecture, whether it's cell or C64. But they don't represent the game industry as whole, and most games are not written like that. The vast majority of games, and especially the multiplatform titles that everyone seems to use for comparing the relative merits of the consoles, are written using standard tools. And generally do not even initialize the SPEs for use.

Plenty of criticism and debate can be had about PS3's overall design and the value of Cell given difficulty in using it effectively, but Cell did carry the platform as it was expected to.

It did that at huge cost, and if they had spent that equal cost on something that wasn't cell, it wouldn't need programming heroics to keep up with the competition, it would easily win over it using simple raw power.

Cell was an unmitigated disaster. Much more so than, for example, Bulldozer is shaping up to be. I do not understand why some people seem to have such a need to defend it.

 

[AlNom]

Cell was an unmitigated disaster.

Yes well, unless this is tying into some grand post about why this is relevant to predicting the next gen, we don't need to shit up the thread with these doom and gloom statements that also ignore the circumstances surrounding the choice and development of Cell.

Can we please move on?

 

[TheWretched]

Yes. Do you realize why the slides are relevant? Because, even after 5 years, it's notable that someone uses the SPEs at all. The SPEs were supposed to give the PS3 a major edge -- and yet, in vast majority of all the PS3 games they are not being used. For them to be worth it, it's not enough for someone to finally use them

So you really think that current multiplatform games run equally well on PS3 and 360, with 360 using 3 cores and PS3 only one? That's not just highly doubtful, but even quite impossible.

 

[Shifty Geezer]

Yes. Do you realize why the slides are relevant? Because, even after 5 years, it's notable that someone uses the SPEs at all. They were expensive. The 150mm² of die area that was used for them could have been used for 2 more real cores, more eDRAM than 360 has, doubling the GPU power of PS3 (which would have made it not need insane SPE tricks to keep up with 360), or, to drive home their pointlessness, it could all have been turned into about 4MB of cache, which would have been vastly more useful than the SPEs were.

That's a complaint about the system design. That's different to evaluating Cell as a processor, and within the context of the system Cell is contributing a great deal.

The vast majority of games, and especially the multiplatform titles that everyone seems to use for comparing the relative merits of the consoles, are written using standard tools. And generally do not even initialize the SPEs for use.

You'll need to present evidence for that. If true, PS3's 'piss-poor GPU' and it's crippled PPC core are achieving about the same results as XB360...

Cell was an unmitigated disaster.

Again, you're talking about it in terms of choice of the whole system design.

I do not understand why some people seem to have such a need to defend it.

For the sake of accuracy. If we're going to have a proper understanding of developments and learn the successes and mistakes of the past, we need an honest assessment. Cell as a processor brought superb, attainable, sustained performance, at a decent enough thermal envelope, with serious issues about developing for it, formed a catalyst for better understanding of the importance of data structures, and helped cultivate a love of real caches among developers. Among other things. Dismissing it out of hand is unfair.

 

[hoho]

Answering the original question, yes, Cell was very good considering what was availiable at 2006. Yes, programming it wasn't easy but I'm fairly certain it was walk in the park compared to PS1/PS2. As was said it's biggest problem was it had to do half the GPU work, more in some titles.

Now I wonder what would have been possible had PS3 come with multiple Cells instead of a Cell+RSX combo

 

[Tahir2]

I wonder how Cell would be used if RSX was as strong as the Xbox360 GPU (as an example).

I know DiCE pretty much used Cell to bring effects to the PS3 from the PC version something they did not achieve with the Xbox360 version. Cell has benefits for sure, developers have used in a variety of ways.

 

[kagemaru]

Can't believe these types of posts even exist anymore. There's no denying that the use of the Cell has carried the PS3 a great deal this gen and that developers are making great use of the chip. Really think you should read up on some dev slides before making these claims Tuna.

I know DiCE pretty much used Cell to bring effects to the PS3 from the PC version something they did not achieve with the Xbox360 version.

AFAIK, they achieved similar results, but through different methods. It's not as if either console version has effects the other is missing from what I understand.

 

[patsu]

...

Ah, but I think you have fallen for the marketing. There are many components in PS3. When a software underperforms, it may or may not be due to Cell and only Cell. ^_^

The small and split memory pool is probably more limiting than the Cell. The latter was used to overcome GPU bottlenecks, play high end audio and Blu-ray, calculate intensive physics and security operations, and perform assorted jobs (like CPU MLAA processing).

The latest feat is probably to emulate PS2 in software.


It's actually doing very very well compared to Itanium (also requires significant software changes) and LRB. As for developers taking advantage of Cell. It really depends more on the developers than the h/w. Some may hit a ballpark performance and call it a day, some will try to squeeze every last ounce from it.

 

[tabs]

Wow, MS should've just gone with one core and a weaker GPU. How are these extra 2 cores and graphics grunt going to waste? Must be an awful design.

 

[Acert93]

Obviously the discussion needs to be limited to tech available in 2006. Another thing to take in consideration for those rebutting the OP (which is not strongly defended) is looking at budgets, name arguments like, "Well, Cell makes up a lot for the GPU." Yes, but put Cell into a system side by side another system with similar budgets (chip area, power draw & heat) and the dynamic changes some. Is the current Cell a better compromise than just getting a better GPU at the time? What if Sony had shifter their area budget over to a larger GPU with a more Xenos like structure and had gone with a smaller Cell or a different CPU altogether (e.g. an AMD X2 or whatnot)?

Cell has allowed some nice things to occur, e.g. more advanced graphical workloads that GPUs are just now able to do due to advanced in programability. Also with 8 cores it really did encourage a job/task model which seems to be the right way to go going forward. On the other hand the 60% sized Xenon seems to do just fine in many workloads and has seen more developers get what they need out of it within their budgets more often this gen. But that could be skewed by all sorts of factors--Xbox out first, middleware, later gen software using the market leader is the design platform, etc.

Oddly, now that Cell has a lot of familiarity, if you are looking at a 32nm or 28nm process in 2013 and need a design that has high performance for many cores and have an area budget in the 200mm^2 range and want an IP that is licensable, developer/middleware familiar, etc Cell probably ranks in the handful of real options. It will have some questions in regards to accessibility and it also has a new, unlikely, contender for many workloads (GPU), but I am sure both Sony and Microsoft have seriously looked at Cell as an option.

So by that metric it is not a complete failure. If it ends up in another console AND developers embrace the console (even begrudgingly) then, again, that would need to be counted as some sort of win.

I guess ultimately if Cell was good or not depends on who you are, what your metrics are, and what your options were and how long are you looking down the road.

Not to derail the thread, but in light of the PR and "consensus" I would say the real "not-loser" was the unexciting Xenon. It did its job against Cell--a bigger chip with a larger power draw and the extended cost of special memory--and while uninspiring and some poor performance in some areas and a not-so-sexy peak performance it has been sufficient to allow continued improvement of software over a 6 year window. And it allowed the luxury of accessibility early on. In theory, considering the size of Cell and RSX and the cost of XDR I think, in theory all partnerships and licenses aside, Sony could have gotten more bang for buck out of the area/power draw. Then again these things are VERY fluid. Plans fail, things get delayed, etc. You have a lot of mitigating factors to consider, some we are not familiar with, and others that if you were in the drivers seat in 2003 aren't so obvious for a 2005 or 2006 console.

 

[JB9861]

Answering the original question, yes, Cell was very good considering what was availiable at 2006. Yes, programming it wasn't easy but I'm fairly certain it was walk in the park compared to PS1/PS2. As was said it's biggest problem was it had to do half the GPU work, more in some titles.

Now I wonder what would have been possible had PS3 come with multiple Cells instead of a Cell+RSX combo

You mean without a GPU right? Probably would've been the biggest trainwreck the industry saw.

 

[patsu]

It probably wouldn't make sense for cost reasons. Kutaragi already said it would be too expensive (and too hot ?). Then again, if they use 2 Cells, the memory would be unified like IBM's CellBlade ? We may see a lot of interesting algorithms since the renderer would be software-based (2 PPUs and 14 SPUs @ 3.2GHz working together), unless Kutaragi adds some fixed function units inside.

 

[Arwin]

*snip*

Let me condense this to: if Microsoft hadn't had an outstanding GPU they would have been left behind, and if Sony hadn't had an outstanding CPU they would have been left behind.

It's an oversimplification but I think most would agree that Cell + Xenos would have killed Xenon + RSX?

On a deeper level, I think Cell was a huge step forwards for multi-core design and power efficiency. Of course we've encountered it's limits and there were a lot of software hurdles to overcome, but together with GPU progress it certainly has been important. In terms of development it has also informed a more flexible and unified GPU pipeline, support for post-processing effects etc.

Commercial success is of course a different story. Playstation lost its dominance and went into third place after Nintendo and Microsoft, and the Cell didn't manage to stop that.

 

[hoho]

You mean without a GPU right? Probably would've been the biggest trainwreck the industry saw.

Yes, without GPU. Obviously it would have been horrible if there was no fixed-function HW for texture sampling and also we are getting the more creative use of Cell in rendering 5 years after the release so for first few years it probably would have been pretty bad.

Basically my main point was that technically a Cell could probably have replaced a GPU at huge power and manufacturing cost + tons of extra work needed from devs. It wouldn't have made much business sense though.

 

[Riddlewire]

I laugh everytime someone on here talks about Cell failing. It can't fail, it's infallible.

-Vince, 6/9/03

 

[V3]

I wonder how Cell would be used if RSX was as strong as the Xbox360 GPU (as an example).

Cell would probably be use less if PS3 had better GPU than RSX. I think Cell would have been better utilise if it was pair with something like Graphics Synthesizer I-32 or its successor. I think that was Sony plans when they developed Cell. They wanted to pair Cell with lots of texture units and ROPs as well as eDRAM to support it. Pretty much the evolution of PS2, a better design PS2 with more of everything, like their GScube. Cell was going to do all the shading, obviously with GPU of today, Cell is not necessary. But I think it still has merits at the time of PS3.

The only reason PS4 would have Cell is for Backward compatibility IMO.

 

[liolio]

It's pretty obvious that in the PS3 (and in some other specific cases) the cell was indeed good.
Whereas Sony (and MS) by the way could have done better is another topic.
Say for the same silicon budget (and worse power draw) Sony went with a 4 cores CPU (using 4 PPU) the PS3 would have never keep up (and in some case exceed) with the 360 performances as the extra cores would never make up for the loss throughput of 5/6 SPUs running around. This topic is kind of a "none question" facts proved that Cell was good and hopefully either way if RSX was the best thing money could buy for Sony the system would have done an overall worse show off.

 

[Deleted member 13524]

One could say the Cell was a success because it "saved" the PS3 from using a sub-par GPU (compared to X360) and allowed the console to attain graphical parity with its competitor.

Or.. one could say it was a total failure because it was presented to the world as the future of media-oriented computing, is the result of a brainstorming and R&D aggregate from three of the largest tech companies in the world (Sony, Toshiba, IBM), about $400M were spent to develop the chip and in the end... it apparently leaves no direct offspring.


If 10 years ago Sony, Toshiba and IBM could know the current state of CPU and GPU evolution, would they develop the Cell anyways?
Probably not, IMHO. It would've probably been better for Sony to just get an off-the-shelf PowerPC/x86 CPU and spend a larger piece of their R&D and power budget on a more future-proof GPU.

 

[assurdum]

It's pretty obvious that in the PS3 (and in some other specific cases) the cell was indeed good.
Whereas Sony (and MS) by the way could have done better is another topic.
Say for the same silicon budget (and worse power draw) Sony went with a 4 cores CPU (using 4 PPU) the PS3 would have never keep up (and in some case exceed) with the 360 performances as the extra cores would never make up for the loss throughput of 5/6 SPUs running around. This topic is kind of a "none question" facts proved that Cell was good and hopefully either way if RSX was the best thing money could buy for Sony the system would have done an overall worse show off.

I'm not so sure about this...I have the suspect sony has been too much 'naive' to believe NVIDIA gives them the best relations of cost/performance considering the 'promise' to buy a better/expansive gpu from them to the beginning ....but I could remember wrong.