Johnathan Blow had an interesting view on it [1], he talks about the idea of the Cell being you somehow refactor your game to be parallelized across the SPEs. But the whole point of a game is that you're simulating a cohesive little world, and anything that is easily separable from the global view of the world is probably not core to the gameplay.

One example is sound effects, if you're generating sound effects, you can't do a good job of that without the context of the environment and what is making the noise, so you necessarily have very deep data dependencies.

[1]: (at about 22 minutes onward) https://oxide.computer/podcast/on-the-metal-9-jonathan-blow/

>> Had the PS4 been a PS3 with a new GPU and a clocked up Cell with an extra PPE would that have been a bad thing?

Probably, if only because non-exclusive AAA games already cost tens of millions of $$$ to develop, and no other consoles or PC's show any similarity with the Cell architecture. So you would either end up with really crappy PS versions of multi-platform games, or they would not be made for PS at all.

I “loved” the idea of the cell when I heard about it, but the development community’s struggle with it should be a salutary lesson to us all!

It's a challenge game companies have had for a long time.

Sometimes, being bold means trying new architectures... But the important thing is making sure they are easy to adopt.

That's the irony of the 'lesson' of the PS3 being hard to develop for; Sega learned that one the hard way with the Saturn.

I guess that’s my question, was the Cell fundamentally bad or was it just too new, different and hard to adopt?

If it’s the latter then building the PS4 on the cell as well should have been fine as by that point it’s a well-known and well-understood architecture.

Even if it's well-understood, being fundamentally different than the other platforms is still a cost for anyone not exclusively making Playstation titles.

The Cell was poorly suited for video games — video games don’t have much in the way of highly parallelizable, discrete tasks. Games, even modern ones, generally want fewer, higher-clocked cores — something like Threadripper would be of little benefit.