Carmack again on PS3-360

What you don't get here is that Carmack is a developer who's words have weight to them, and he uses this chance to try to help the entire industry... His complaints are not neccessarily based on his own problems, you see.

 

According to Amdahl's Law, eventually your speedup will hit a limit based on the amount of code that can actually be parallelized. It won't matter if you have 1000 cores if your code can't use them. And it's not as easy as just making the code so that it just "works". You are constrained by your architecture and workload, not your code. Otherwise, one could just hire a bunch of great coders and be done with it...

 

Just out of curiosity;
- If people have to 'proove' themselves by making a game before they can question Carmack's opinions on the consoles, shouldn't Carmack 'proove' himself by making a console before he started telling them how to do it?

Did Carmack say what was supposed to be in PS3, instead of Cell?
Sure, the PS3 would require less power without the Cell, but it still needs a CPU.
Emotion Engine might be a good chip, but I believe the time were right to move onto something newer and better (like Cell) for their next Playstation.

Since the software producers have come so far that the hardware manufacturers should make their hardware fit the current software, instead as the other way around as it used to be, I think we shouldn't blame Sony.
I think it must have been Carmack wich sent bad beta-code to Sony, while Microsoft got the good alpha-code. Microsoft offcourse managed to design their Xenon to fit Carmacks code, much better than. While Sony didn't manage to make Cell work as easily on Carmacks code.

Atleast Carmack should be pleased that Sony did something right,
Think how hard it would be to code on Cell if it had been clocked to 4 GHz, or if the coders had to worry about double amount of HDMI ports to code for, as initially planned.. :-/

j/k

Megathreading is probably super and all that... But I find it strange while people in Sega says anyone who coded on PS2 is jumping of joy when coding on PS3, while Carmack does nothing but complain.
It seems like some of the developers no longer need some sort of hardware evolution, they don't need Bluray and it's extra storage space since compression gets better all the time, Cell is too hard to code for, etc. etc.. :-/

If multithreading is hard today, Cell should be a perfect training field, especially if it can reach as large a market as it's two predecessors did.

 

All one really needs is a highly parallel solution that is better than an equivalent serial one. They do exist in many places. It may not be optimal, but if it is better than the alternative, what else can you do?

 

Nothing strange about that, fundamentally the PS3 is a very traditional PC like infrastructre with a Nvidia graphics card, and IBM PPE cpu. It must be much easier than what they had with the PS2. Of course they are grateful.

Carmack on the other hand is used to working on traditional systems like this, so of course he's not jumping for joy. He's not focusing on comparisons to PS2 since those are not relevant to him, instead he's making comparisons with singlecore CPU's, and pointing out how much extra time and effort is required with multi-core architectures, and in particular feels asymetrical architectures introduce even more problems.

 

There is no such thing. Any parallel solution can obviously be serialized, and so only ever be as good as a serial solution, never better. Given any perfectly parallel algorithm runing over N processors, you could simply run those same N operations in serial. In other words, a serial processor N times as fast will always be better than having N processors. Parallel algorithms are just inherently more difficult to conceive than serial ones.

 

Sure, but I thought the point I was addressing was a little less hypothetical in nature, where there are N processors possible, but not one processor N times as fast.

 

sorry, a ps3 ****** told me this. serious. i thought JK works for id software and makes sense.

 

JC i mean... what is wrong with me

 

Actually, Raven didn't port Quake 4 so much as they wrote Quake 4. Carmack created the Doom 3 engine, of course, but Raven took the Doom 3 engine and made Quake 4 using it.

 

For those interested, the next part is up now

 

He does have real points about multi threading and concurency nightmares etc. I just feel that if you decide to tackle these challenges, you want to do it on the platform that has the biggest rewards for doing so.

And I don't see avoiding the pot hole issues with mutiple threads/tasks on a symmetric 3 core design significantly easier than a 9 core asymmetric design.

Of course there are other things like forcing the programmer to deal with split memory pools, smallish SPU LS, and forcing the programmer to have to interact with the DMA controler for system memory accesses etc that makes a PS3 programmer's job challenging. They are all complexity for performance trade offs and Carmack probably doesn't like a single one of them. But at least those have straightforward known solutions. And the are all trivial in comparison to the "big issues" with multi threading.

 

better games? You mean visually right?

edit btw I hear this about assymmetric cores on the PS3, and although I am not a tech geek, doesnt the CELL have one PPE core and 7 symmetrical cores? The assymmetry I see is the PPE compared to the other cores. Arent the other 7 (or 8) cores symmetrical?

 

He clearly states that he's coming from the viewpoint of a content creator. I see where he's coming from. Anything that takes energy away from pure creativity is, I guess bad for everyone.

That said, it's diminished by the fact that multi-core tech is the best answer we have at the moment. So his bitching is only valid in with regards to the tools. Multi core is here to stay and he damages his image (in my eyes) by publicly sharing his disdain for the challenge.

 

Yes.

But it's like saying if you took the front tire off your bicycle you would have a unicycle. Maybe technically you do but it's not going anywhere.

 

JC is turning into the cranky old geezer of gaming development world, and talks nothing other than how great the old days were.

 

I fail to make the connection

 

The term symmetrical multi-processor (SMP) machines is usually used for machines with equal access to a shared pool of memory, so the SPU array can't qualify as a SMP machine, even if you take the PPE away.

 

Well thats something I cant quite grasp well since I dont know much about technology