SO this is what I am getting out of the 360 and PS3...

Good summary Dave And Welcome Back!

That sounds more than reasonable. From the time of the E3 footage you guys have at least a year, if not much more, to not only finish the game content but the streamline the engine. 5-20fps is really not bad considering how early in development you are... considering you are on hardware a good couple notches less powerful/streamlined as the final target doesn't help either.

I cannot wait to see what you guys do for a 2nd gen title from scratch!

 

That's what I'm starting to believe.
PS3 has the edge in raw performance. Xbox has the edge in ease of programming. So in the beginning the two will balance each other out.

Though Sony does have some talented developers on their side who would be able to quickly extract maximum performance from the PS3.

 

I don't know how intrinsically easier XB360 is to program for than PS3. Both need personal approaches. I think development ease will be mostly attributable to tools, which will change over time.

 

It took like 5 or 6 years until ps2 stopped getting garbage like Red Faction, and began gettin amazing games like GOW and GT4.

i wouldn't call that "quickly"

 

but xbox was still better looking

Even though the PS2 had a great and powerful CPU for graphics design the xbox still looked better. Last time it was 4 times as powerful this time its only twice as powerful. Though 100 flops is a huge difference in absolute terms.

The information I am itching to know how will PS3 be able to fully exploit Cell and RSX at the same time. the felxio is the key but nobody really knows "how" it will be done.

 

Can we clone you or something?

You hit the nail on the head, without taking a jab or inserting some side comment at either MS or Sony fans.

You are a machine, huh? Toying with us mere mortals

Keep up the good work Shifty.

 

Re: but xbox was still better looking

The PS3 CELL CPU doesn't have a dot product instruction, while the Xbox 360 CPU does.

The SPE's on CELL will take a big performance hit when it comes to branch misperdicts. The Xbox 360 CPU cores should be a little for forgiving when dealing with branches.

 

Re: but xbox was still better looking

So? In a competition of dot products I know where my money would be

True. But with profiling and hints, where there is a non-equal probability of taking one branch over the other, you should mostly take the right branch. In some instances, might "programmer predicition" (hints) not be a little better than hardware prediction, even?

 

Most likely not..but the spe's aren't going to work with branch heavy code/tasks anyway.

 

"I think development ease will be mostly attributable to tools"

If that's the case, MS has won that war already. They have three huge teams dedicated to making development tools and helping out third parties, in addition to the Xbox Base Software team which is the largest MS Xbox team. It's not even a contest.

"SPEs are evil little bastards" will be the new quote of next generation

 

dunno..there are also hundreds of people at IBM working on cell development tools and Linux integration. Another potential is that IBM made it open source.

 

Really? To me it seemed by the time Xbox and Gamecube were out, PS2 had just about caught up with their graphics. PS2 never caught up in IQ of the image though, though some GameCube and Xbox games got worse in IQ.(I think Halo 2 and Resident Evil 4 both have rather bad IQ)

BTW, anyone think Nintendo might win ease of development?

 

It's true that MS's tools are always really nice and Sony's tools have always been little more than a pile of garbage in the form of 0s and 1s. Even so, we're forgetting that PS3's tools are largely the work of IBM, so it's not quite the 1000:1 difference that you'd expect.

Either way, I expect MS's tools to be no different than they've always been - clean, simple, easy to use, and prone to crash. I'm still of the impression that this whole XNA thing is overblown. If there's anything I'd expect IBM to have the edge on, it'll be the compilers.

IBM and Microsoft aside, the big thing missing from both of them (and probably always will be) is just art pipeline. Art is always going to be the expensive part of things, and at the same time, the most important. So what if XNA will enable me to get my renderer up and going quicker? Is it going to help my artists produce quicker at lower cost? Is it going to make it any easier to get content into my engine? Is it going to enable me to get by with a smaller team of artists?

 

Isn't that what improved use of procedural synthesis is going to allow? (at least as I understood Anand's article to say)

 

Nope not the work of IBM.
Sony even turned down the IBM compiler, arguably one of the best multithreaded compilers around........ They are using the Gnu compiler again.

There is currently no way to debug native SPE code in the Sony toolset, and just single stepping is a hideous operation in the debugger.

Sony has a LONG way to catch up to MS on the tools side. I do know that in the long term they are committed to getting there, but they're certainly not there yet.

 

Procedural stuff can't be used for everything. And you need both programmer and artist input for that, usually - programmers to actually code the stuff, and artists to tweak the parameters and so forth to ensure it looks good. You wouldn't necessarily save time with it, depending on what you're doing - though in some cases, you would, yes, and it could be very worthwhile.

 

And 360 CPU has 2-3x longer instruction latencies. You know how they say nothing comes for free

Obviously since the prediction is automatic - but when the mispredict does happen, the penalty is worse on XeCore then on SPE.

No need to be nice about it - they were probably just being cheap not wanting to pay licensing to IBM for it

 

I found it highly amusing that the IBM Cell FFT presentation was using XLC not GCC...

 

Sort of OT, but: I guess this is due to the longer XeCore pipeline? Also, do you know how exactly branch hinting in the SPEs works? If you have 3 branches, do you simply hint "go with branch b", since b is the most common case, or are there more sophisticated options available (like based on the last branch taken etc.)?

 

Re: but xbox was still better looking

SPEs will not take a performance hit when it comes to mispredicted branches..cause SPEs don't have a branch prediction unit!
IIRC, if a branch has not been previously hinted, a SPE assumes a conditional branch is always taken.
If the branch is not taken (and it's not hinted) then there would be a severe performance hit.