After pp and multi cores what comes after?  Adding more cores with multiple sub processors and then adding more to that has o stop somewhere.  What is the next process?

There is only so many concurrant threads you can have running, particularly in a video game (hence the "only x% of the cell is being used!" claims)

PP is great for OSs where dozens of programs are vying for processor time at once, but in dedicated game machines, PP potential really tops out at a few cores.

My guess would be new processor technology that moves away from silicon chips. That ceiling is being reached rapidly as well...

Long term, no clue. Short term is actually probably less growth in core number and more growth in memory bandwidth which is a major current limiting factor.

@Crashman Yeah, the end of Moore's Law is quickly coming to fruition.

@Small So more advances in memory gates with DDR3 and DDR4?

Not to argue with anything anyone has said here.

I was at a games conference (GCAP) 3 weeks ago and i sat in on a PP talk and as far as I could tell the direction was a large increase on cores. In the direction of 128 cores.

I believe the future in games is the challange to take advantage of this mass amount of processing power.

Apart from that if you want to think of what the future could be, you should look at the main issues with programming games. aka resource and memory management.

@OGY And what kind of console would have that many cores? The IBM Big Blue @ Home? No I think a new revolution is needed here instead of just playing addition until the machines can't handle it anymore and the memory leaks create Niagara falls.

What should come after parallel processing is software that uses it properly to its best advantage.

But that takes 3-5 years of people stealing each others ideas and another couple to bring it to market and you are just all ready to go with it when - whoops - there's another hardware generation.

So what I'm hoping for is a bit of stability to let the software catch up.

Fat chance.

 Well I am not a game programmer but I use a lot of computing resources, scientific HPC.  What I was commenting on was purely that you can not load enough data from main memory to on chip cache to feed multiple cores unless you get a lot of reuse of data.  That means that for some problems, not sure if this includes games, your going to be fine but if your loading lots of data from main memory your not getting a lot of usage out of all your cores as it is because they are waiting on data being loaded from main memory or a lower cache.  Until you have this solved though adding a whole bunch more cores does not help solve certain problems even if they are a parrallel problems. 

Well I think the presentation I saw was about the challanges faced with console developers when this occurs.

You spoke of memory leaks, well thats exactly right and thats the challange, hence why i mentioned memory management.

I think the next big change will be memory. Some type of new hardware that is built for the idea of a large amount of cores, how im not sure, hell im just a programmer.

Aside from that, i think alot of people miss the oportunities mass amounts of cores provide. If you havnt seen 64k games then google it. Those german guys are making amazing 3d games with large amounts of assets all in the size of 64k, as everything is developed from the code in load time.

I see cores in future consoles being dedicated to realtime asset development, which will reduce any resource issues even deadlock situations. Not to mention the reduced size of games and the eventual future of completely online games.

The next stage will (possibly) be to return to a CISC mindset and increase the instruction set of the cores using more complicated instructions in order to increase performance in specific situations ...

Consider 4x4 Matrix multiplication which is used heavily in videogames, in order to multiply two matricies together it takes 64 floating point multiplications and 48 floating point additions; it is easy to reduce this to take the time of 4 floating point multiplications and 3 floating point additions through parallel execution on a single core (which can be improved further using other methods) but this would (dramatically) increase the size of a core.