> For some reason, I am very fascinated by the architecture of PS3.

You’re not alone. I also am enchanted by non-standard architectures: the PS3’s Cell, its predecessor the PS2’s Emotion Engine, the Transmeta Crusoe, etc.

There’s a sibling comment to mine that talks about needless complication and dead-ends. That’s fine, but marching along with essentially optimizations to a basic architecture seems boring to me from a creative point of view (not that the achievements made haven’t been, of course, profoundly technically impressive).

And that’s on top of the fact the designing computer architectures can probably be thought of as a huge multidimensional optimization problem (where optimal can change over time or between customer demographics). I think of the approach of iterating as helping us march up that manifold to a local maxima. I think of these “exotic” architectures as sampling far away from those points to see if maybe we can find a more global maxima.

And that’s not to say that the main platforms aren’t innovating: with big-little, NUMA, etc.

But there’s a soft spot in my heart for those wild, long shot bets.

The architecture is unnecessarily complicated and it turned out to be dead end. Which is why SONY dropped it and now uses x86 like everyone else.

Everything from this generation (PS3's Cell included) was some sort of PowerPC, many previous gens also had consoles using PPC or MIPS. Sony and MS went x86 during the same generation (PS4 and Xbox One). Nintendo went from PPC to ARM (though arguably they'd already used ARM for a while because of their handheld stuff).

I believe the GP is referring to the entire computer architecture, not the instruction set architecture. I seem to recall that it had three distinct heterogenous processors that had to be coordinated to get the most out of the system, so porting from, say, PC to PS5 wasn't necessarily straightforward, without leaving performance on the table.

The xbox 360 was the one with the triple CPU, PS3 had the odd combo of one "main" CPU and 7 auxiliary ones.

Iirc there were 8 chiplets on the cell die, one was disabled for yield rate and one was dedicated to the os itself, leaving 6 of them for gaming.

Then they had the super riced out rsx (Nvidia GeForce 7000 series) for the GPU, but there wasn't really a "main" CPU was there?

The Cell had a PowerPC core (the Power Processing Element) as its main CPU, alongside the Synergistic Processing Element co-processor units.

I think the 360 had a regular homogeneous tricore CPU, so programming it was just normal multithread programming. As you say the PS3 had several auxiliary processors that all needed to be told what to do in specific, non-portable ways, which is what made it inconvenient to work with.

Fun fact about the 360, MS bought a bunch of PowerMac G5 towers and installed a PPC build of Windows XP on them to turn them into Xbox 360 dev kits. Makes sense because those were the most readily available and most cost effective PPC boxes at the time, but kind of funny.

OG Xbox was x86 and was released similar time as PS2.

Then they went PPC for 360.

And now back to x86.

IMO they went AMD because that was the only real option left. PPC was dead by then outside of big IBM servers and ARM was just getting into low-end x86 performance levels. If PPC was still getting improved in the embedded space I think they might have went that way for backwards compatibility reasons.

IBM put in a bid for PS4 (based on Power7 IIRC) but it wasn't selected. I think the convenience of getting everything from AMD helped them take over consoles.

AMD was also able to offer up an APU (GPU integrated with CPU) vs separate chips that would have increased cost and complexity.

Developers had to be more creative when the hardware shaped the kind of game you were making. that's what's fascinating to me

In practice it made multiplatform games to look and run worse. For example - GTA IV ran @ 720p on x360, but only @ 640p on the PS3.

Another reason why both the PS and XBox use basically the same hardware under the hood.

I think the other aspect as well is that the 360 ended up with the somewhat better GPU in Xenos.

I'm not sure if it was obvious at the time in-between the Cell BE and "reality synthesiser" marketing etc. but I understand that gave the 360 an edge in a lot of cross-platform titles.

I think this really influenced the design of the PS4 which actually had a really weak Jaguar CPU but a solid GPU in its APU.

On the other hand, weird hardware architectures are the worst from a preservation standpoint. If no one had made a PS3 emulator, MGS4 was at serious risk of becoming lost media.

Is MGS4 really that difficult to port to a different architecture?

I don't see Konami interested in doing that anymore, and I don't see how else it could happen.

They’re releasing a “volume 1” collection of MGS games soon, which contains up to MGS3. Presumably MGS4 will be in volume 2.

I had no idea. Apparently it's coming to PC. Here's hoping I can finally play MGS4 on PC, as well.

Here's hoping they leave in the install screen somehow. Big fan of watching the main character chainsmoke for a couple of minutes.

They announced a MGS3 remake recently. So they might remake MGS4 as well.

One of the hooks of the PS3 architecture was that it sported a core count greater than 2 in consumer hardware (albeit as a very simplified core).

Rumours about Intel Larrabee were also flying around at the time, so it seemed like the future was here.