PS3 Cell CPU

[Paddi01]

Hello,

I'm just curious what was so special about the architecture of the PS3 and why was it such a challenge to develop games for it?

Consoles have the same hardware specs but why was Cell so hated?

[FuzzyLogic]

I'm pretty sure it was just because it's a bit of a pain to code everything for a high core-count CPU. Other than that, I don't know much about it.

[Tabs]

3 minutes ago, Paddi01 said:

Hello,

I'm just curious what was so special about the architecture of the PS3 and why was it such a challenge to develop games for it?

Consoles have the same hardware specs but why was Cell so hated?

Cell was an IBM Power architecture chip that was designed for use in mainframes where security was the first concern. It was heavily multi-threaded, which was very very different from other architectures at the time, which prioritised a few, high performance cores which are very flexible. The SPU's in Cell were very inflexible and were just really awful for the type of content (games) that were expected to be run on it.

 

So a developer had a choice - optimise a game for the almost alien Cell architecture, and then have to do way more work to get the same game to run on other platforms (xbox, pc). Or, they could optimise for xbox and pc, which were the lions share of the market, and then end up with a terribly optimised ps3 version (or no ps3 version).

 

Middleware eventually became available to help with this transition but it was a hell of a time for game devs. 

 

Sony basically tried to replicate the success of their PlayStation 2, which had a really weird architecture as well - but the difference was, when the PS2 launched, Sony had massive developer market share and massive player base to make designing for the weird ps2 architecture worthwhile. It just didn't have that kind of clout once the PS3 came out, since it was competing with the xbox 360, which for all intents and purposes was a beefed up xbox and was familiar to code for for anyone who had coded for pc or the original xbox.

[Stefan Payne]

3 hours ago, Paddi01 said:

Hello,

I'm just curious what was so special about the architecture of the PS3 and why was it such a challenge to develop games for it?

Consoles have the same hardware specs but why was Cell so hated?

Essentially it had just one General Purpose Core and 7 other, more like Graphics Card Shader Things that were simple but also hard to impossible to use, depending on the Game.

 

And that's one of the Reasons some Games, even exclusives (Drakengard 3) run like shit on the PS3.

Well, the Graphics core is pretty bad too...

[Stefan Payne]

3 hours ago, Tabs said:

the xbox 360, which for all intents and purposes was a beefed up xbox and was familiar to code for for anyone who had coded for pc or the original xbox.

Nope, that's not true.

The Original XBox used an Intel P3/Celorn Thingy with an nForce Chipset and an nVidia Geforce 3 based thingy.

 

The 360 used a Power PC Architecture (and yes, Spectre and Meltdown do not apply here) with a proprietary AMD Chip wich had the TMU on a different die, with some fast eDRAM as well. 

So completely different architecture with some quirks but was by far the superior console. Though the Slim consumes much more power than the slim PS3.

 

And to make matters worse for Sony: The X360 had something like an APU as well -> single Chip solution that's common in consoles today with a unified memory architecture. While the PS3 had (expensive!) RDRAM main memory for the CPU and also some GDDR3 for the nVidia G70 based Chip (wich also was rather shit at the time because of the Pipeline stalling)...

 

 

What people say is that the Cell Things should have been used as a Graphics processor...

But that's just a rumor...

 

PS: the PS2 wasn't that weird.

The CPU was pretty standard for the Time. The Graphics chip was a bit strange in parts and the worst part was that it didn't have texture compression and had a rather small texture buffer...

[Mira Yurizaki]

The problem with the Cell was its SPUs. It was expected that there would be a lot of data and number crunching going on, as a result, the SPUs aren't really cores like you think of today as cores. They're more like self-contained CPUs. The biggest take away on that is that the SPUs cannot access main memory, they only have access to 256KB of SRAM (and that's all they get, there's no cache) that the PPE is supposed to feed it.

 

It was likely that Sony wanted the Cell to act similar to what a GPGPU does before the concept was even a thing. Today there's a lot of things being calculated on the compute path of the GPU that the SPUs of the Cell could've handled. But the concept wasn't really a thing back then, so developers couldn't really make use of the system to its fullest until towards the end of the system's life.

[Tabs]

3 hours ago, Stefan Payne said:

Nope, that's not true.

The Original XBox used an Intel P3/Celorn Thingy with an nForce Chipset and an nVidia Geforce 3 based thingy.

 

The 360 used a Power PC Architecture (and yes, Spectre and Meltdown do not apply here) with a proprietary AMD Chip wich had the TMU on a different die, with some fast eDRAM as well. 

So completely different architecture with some quirks but was by far the superior console. Though the Slim consumes much more power than the slim PS3.

 

And to make matters worse for Sony: The X360 had something like an APU as well -> single Chip solution that's common in consoles today with a unified memory architecture. While the PS3 had (expensive!) RDRAM main memory for the CPU and also some GDDR3 for the nVidia G70 based Chip (wich also was rather shit at the time because of the Pipeline stalling)...

 

 

What people say is that the Cell Things should have been used as a Graphics processor...

But that's just a rumor...

 

PS: the PS2 wasn't that weird.

The CPU was pretty standard for the Time. The Graphics chip was a bit strange in parts and the worst part was that it didn't have texture compression and had a rather small texture buffer...

 

In basic terms, I was alluding to the fact that the Xbox 360 had fully functional and programmable "cores" as we use the term today, unlike SPU's as @M.Yurizaki said. They could be coded for in relatively normal fashion, for anyone who had worked with any kind of "standard" microarchitecture in the past. Cell required a whole new way of thinking - a CPU with heavy number crunching duties and very low levels of flexibility. SPU's couldn't physically process general purpose instructions, whereas Xenon cores could. The PPE on Cell is basically the only real "core" it has, and thanks to the SPE's, it's only duty could be trying to provide data for the SPU's to process. It was a tough sell for developers when the performance gains were marginal even on well optimised titles versus the equivalent pc and xbox versions. The architecture just wasn't suited for running games.

 

The PS2 was the first super successful console that had separate programmable graphics and central processors. The Dreamcast did too, but it had a lot of marketing and design failures that led to it being a loss leading endeavour - and since the PS2 landed only around a year after the Dreamcast launched, it was a super tough consumer sell for a completely new platform. Sony had the benefit of PS1 backwards compatibility, which gave it a massive library out the gate. By "weird", I was pointing out that it was a departure from the standard console design paradigm at the time - and it succeeded because Sony had the marketing clout and developer support to make it a non-issue. That wasn't the case when the PS3 arrived, a year later than the xbox 360, more expensive, and with few reasons to program for it over the existing competition.

[Stefan Payne]

5 hours ago, Tabs said:

 

In basic terms, I was alluding to the fact that the Xbox 360 had fully functional and programmable "cores" as we use the term today, unlike SPU's as @M.Yurizaki said.

I said as much in the Post above.
And the one you were quoting, I didn't even say much about the PS3 CPU at all. Why bring that up??
I only said that the X360 was better and the far superior console at the time (and in hinsight I am happy to get the 360 and not the PS3 at the time).

Though now I have one 360 and two PS3 (a Slim and a Super Slim).nd to 

 

And to make matters worse, Microsoft could integrate the CPU and GPU to one Chip while Sony could not - be it because nVidia wouldn't let them or whatever. The Result is that nVidia won't be seen in a Sony console any time soon because of that...

 

 

And what you said about the sales isn't true either.

Both consoles of that Generation sold about the same, though its safe to assume that M$ made some money off the 360 while Sony didn't.

 

5 hours ago, Tabs said:

The PS2 was the first super successful console that had separate programmable graphics and central processors.

I think you give the PS2 too much Credit because at the time there was SEGA's Saturn and the Nintendo N64 as well.

The PS2 was just the Second Console from Sony...

 

5 hours ago, Tabs said:

The Dreamcast did too, but it had a lot of marketing and design failures that led to it being a loss leading endeavour - and since the PS2 landed only around a year after the Dreamcast launched, it was a super tough consumer sell for a completely new platform.

And that was the Problem of the Dreamcast...
It was something in between and SEGA lost many fans because of the Saturn failure.

It wasn't too shabby, but they just didn't have the Developer Support and also lost some with the previous failure. Something that Nintendo also did - either with the N64 or the Gamecube, I don't remember right now...

5 hours ago, Tabs said:

Sony had the benefit of PS1 backwards compatibility, which gave it a massive library out the gate.

No.

The PS2 got traction because it was the cheapest DVD Player at the time. Though its a rather shitty DVD Player tbh...

 

The PS1 Libary might have been a Bonus but that it could play DVD while the Dreamcast could not was a more important benefit for most users.

Something the Gamecube also lacked for whatever reason.

5 hours ago, Tabs said:

By "weird", I was pointing out that it was a departure from the standard console design paradigm at the time - and it succeeded because Sony had the marketing clout and developer support to make it a non-issue.

No, it was the Zeitgeist.

PC was a bit further ahead at the time and Consoles far behind. Just look at the XBox, when it came out and what it had to offer....

 

5 hours ago, Tabs said:

That wasn't the case when the PS3 arrived, a year later than the xbox 360, more expensive, and with few reasons to program for it over the existing competition.

Yeah, because Sony was still in their world and thought to do something completely different. WIch did work for the PS2 but did not for the 3.

 

They messed up the CPU quite a bit and they choose the wrong graphics chip. The Xenos seems to be the long lost and rumored R400 or something like that and even had unified shader. 

And the Pipeline didn't stall when doing texture operation like it was the case on the nVidia CineFX Architecture that was still used in the PS3 GPU, wich seems to be little more than an off the shelves G70...