John Carmack interview talks PS3 & X360

[Lysander]

hey, version, you shrank X2cpu pic on x-axis

 

[Gholbine]

hey, version, you shrank X2cpu pic on x-axis

It looks pretty accurate to me, judging by the size of the Xenon cores and the PPE.

Edit: Can anyone tell me what the vertical black strips are on each SPE? Is that the local store?

 

[Colourless]

Yes, it's the local storage

 

[Nite_Hawk]

It will be very interesting to see which companies in 2 years have produced the most impressive games (be they multithreaded or not).

Nite_Hawk

 

[ihamoitc2005]

I hope most games do not dedicate a CPU core for compression related tasks as recommended by Allard. Its the best way to make sure that the games will look horrible because Xbox360's best chance of matching PS3 graphics is to use full GPU power for pixel-shading and do all geometry on CPU. Enough CPU bandwidth for that. Of course some games wont have as much pixel-shading requirement, but hopefully, many will since those games look better.

 

[Lysander]

looks pretty accurate

no it is not, can someone make comparision with equal size ppe/X2core
(without photo manipulation)

 

[chachi]

no it is not, can someone make comparision with equal size ppe/X2core
(without photo manipulation)

Please, no. There's already been 57 different versions of the comparison, the PPE and XCPU cores should be the same size. I have no idea why some platform advocates like to pretend there's a big difference between the two (i.e. the XCPU core = DD1 Cell!!!@ theory) when the die photo and documentation suggests otherwise, but it'd be nice if this didn't keep creeping into what seems like every thread.

The similarity comment is in regards to the programming model. The PC and 360 share one (SMP) while the PS3 needs a completely different model for the SPEs which have more limitations on what you can do (e.g. LS). It'd be easier to share code between PC and 360 without massive rewrites but the only way to write code that could reasonably be used on all platforms is if you wrote with the PS3 in mind and that would put the others at a disadvantage because they don't have the same limitations.

 

[SubD]

It'd be easier to share code between PC and 360 without massive rewrites but the only way to write code that could reasonably be used on all platforms is if you wrote with the PS3 in mind and that would put the others at a disadvantage because they don't have the same limitations.

What exactly are these limitations supposed to be?

 

[overclocked]

Please, no. There's already been 57 different versions of the comparison, the PPE and XCPU cores should be the same size. I have no idea why some platform advocates like to pretend there's a big difference between the two (i.e. the XCPU core = DD1 Cell!!!@ theory) when the die photo and documentation suggests otherwise, but it'd be nice if this didn't keep creeping into what seems like every thread.

The similarity comment is in regards to the programming model. The PC and 360 share one (SMP) while the PS3 needs a completely different model for the SPEs which have more limitations on what you can do (e.g. LS). It'd be easier to share code between PC and 360 without massive rewrites but the only way to write code that could reasonably be used on all platforms is if you wrote with the PS3 in mind and that would put the others at a disadvantage because they don't have the same limitations.

The last sentence doesnt sound right.
I take it as you mean "A game that uses SPEs heavily will put the others at a disadvantage
because they HAVE limitations". I just get confused because you write that it would be easier to code for PS3 as base and if you mean just using the PPE then it would make sence. But why would you not want to use the SPEs on the Cell IF you use it as your base and port over?

Or maybe you are speaking from a more "general" POV?

 

[ERP]

What exactly are these limitations supposed to be?

He's just stating that the others have more flexible access to main memory and structuring code for SPU's is probably not structuring it optimally for a more conventional Shared memory multithreaded environment.

i.e. you might be able to run code designed for an SPU on one of the XB360's processors, but it probably isn't the best way to do it. Conversely designing code for an XB360 processor means it likely won't run in the SPU's limited local memory.

 

[chachi]

The last sentence doesnt sound right.
I take it as you mean "A game that uses SPEs heavily will put the others at a disadvantage because they HAVE limitations". I just get confused because you write that it would be easier to code for PS3 as base and if you mean just using the PPE then it would make sence. But why would you not want to use the SPEs on the Cell IF you use it as your base and port over?

No, I mean that because the SPEs have limitations that aren't present in the other cores (e.g. PPE, x86) and that you have to design a thread that runs on one in a way that takes those limitations (LS, context switching will suck, no branch prediction, etc.) into account. There are advantages to the SPEs as well, but you're not going to want to use the same strategy for programming an SPE as you would the PPE (and vice versa).

That's not to say that Cell is bad, and on exclusive games the point is moot, but it makes things more complicated for cross-platform. If "what works on an SPE" is the measure of what will be assigned to a thread in a cross-platform multi-threaded game then that is where the others are disadvantaged.

Does that make more sense? If you go back and read some of the cross-platform developer interviews (Carmack, the Valve guy, the Crytek guy) you'll see they're all saying more or less the same thing.

 

[Bobbler]

No, I mean that because the SPEs have limitations that aren't present in the other cores (e.g. PPE, x86) and that you have to design a thread that runs on one in a way that takes those limitations (LS, context switching will suck, no branch prediction, etc.) into account. There are advantages to the SPEs as well, but you're not going to want to use the same strategy for programming an SPE as you would the PPE (and vice versa).

That's not to say that Cell is bad, and on exclusive games the point is moot, but it makes things more complicated for cross-platform. If "what works on an SPE" is the measure of what will be assigned to a thread in a cross-platform multi-threaded game then that is where the others are disadvantaged.

Does that make more sense? If you go back and read some of the cross-platform developer interviews (Carmack, the Valve guy, the Crytek guy) you'll see they're all saying more or less the same thing.

I think you may have hinted at this, but...

X86/PPC/etc all have rather large limitations... they just happen to have limitations programmers are accustomed to -- that's the only difference (and a very important one for early games and younger programmers).

 

[Lazy8s]

All of the new technologies and flexibility of the X360 would be interesting to coders. It's also the only console that's scheduled to have a state-of-the-art graphics processor at release.

The advantage of a more PC-like development environment is that the tools are more refined.

 

[Acert93]

I hope most games do not dedicate a CPU core for compression related tasks as recommended by Allard... Enough CPU bandwidth for that.

I am pretty sure the "compression" he is talking about it procedural synthesis. It saves bandwidth, memory space, and could use up an entire core. And since you are taking a small list and dynamically generating it one could call it a form of compression because you are taking a very small file and making it into a larger game asset.

 

[one]

I am pretty sure the "compression" he is talking about it procedural synthesis

I think it only literally means CPU instructions that support native D3D-compressed data.

 

[ihamoitc2005]

I think it only literally means CPU instructions that support native D3D-compressed data.

It seems to be a suggestion about how to work around the storage capacity of the DVD and lack of a hard drive more than anything else since that's what gave rise to the streaming technology in the first place no?

 

[wireframe]

Interesting that Carmack feels so strongly pro360... To me he's always seemed liek a coder's coder so I woul dthink from a technical challenge/potential standpoint he'd be PS3 all the way... *shrug* Also intersting is the he never really states why from a technical standpoint. What about the x360 CPU does he prefer? Symmetry? VMX units? Cache?

I'm gonna guess that this is because Carmack prefers to spend his time innovating algorithms and ways to accomplish tasks for his games instead of learning new ways to operate on a new architecture. That is to say, I think Carmack prefers to work on getting good "generic" code instead of learning particular tricks of a certain CPU.

He has embraced smaller changes like MMX, SSE, and SSE2 on the x86 architecture, but I think the reason that ID games tend to run well is because it has good foundations, not little tweaks here and there, but fundamentally sound code. This is very different from how many Japanese developers are coding as close to the metal as possible. From what I understand, many of the better ones (Japanese) make impossible things happen through tricks that cannot be taken advantage of by simply thinking in C/C++, but requires intimate CPU architecture knowledge and assembly code.

 

[overclocked]

No, I mean that because the SPEs have limitations that aren't present in the other cores (e.g. PPE, x86) and that you have to design a thread that runs on one in a way that takes those limitations (LS, context switching will suck, no branch prediction, etc.) into account. There are advantages to the SPEs as well, but you're not going to want to use the same strategy for programming an SPE as you would the PPE (and vice versa).

That's not to say that Cell is bad, and on exclusive games the point is moot, but it makes things more complicated for cross-platform. If "what works on an SPE" is the measure of what will be assigned to a thread in a cross-platform multi-threaded game then that is where the others are disadvantaged.

Does that make more sense? If you go back and read some of the cross-platform developer interviews (Carmack, the Valve guy, the Crytek guy) you'll see they're all saying more or less the same thing.

Okay now im with you. I didnt thought at all you were refering the limits of the SPEs compared to the PPE/xCpu which make perfect sense in the coding department and structuring of data. I guess its a bad idea to post before sleep..

Just want to say my view on esp Gabe´s comments with many threads. I think its natural for a PC guy that just have been used to throw whatever code at all at the latest PC Cpus and it just works anyway. But i have a feeling that if thats Valves stake at the whole we aint going to see any impressive stuff later on compared to the hard working console devs effort on either console.

 

[Acert93]

I think it only literally means CPU instructions that support native D3D-compressed data.

My understanding that the D3D compression is for transferring information between the GPU<>CPU (mainly vertex data, ala procedural synthesis) for the point of saving bandwidth between the CPU and GPU.

I am certain they wont be designating an entire CPU for decompressing information off of the DVD; but we have heard them talk about dedicating one or two cores for procedural synthesis which is a forum of storage compression (both DVD and memory) and saves bandwidth across the board (and even more with D3D compression which is said to double the effective bandwidth for vertex data).

Anyhow, in the context of saving space on DVDs procedural synthesis makes sense sa you can store large amounts of data in small lists (that are later dynamically created) and also save a bit of RAM space as well in storage.

 

[Wunderchu]

summary of the recent G4 interview:

Summary of the interview (I watched this a few hours ago so I may be forgetting some stuff):

- Talked about the Doom movie, showed some footage of it etc

- Talked about Raven and how they are close etc

- Reason for no flashlight/gun at the same time in Doom 3 was that earlier on it brang the frame rate down a lot cause of the dynamic lighting (or something like that). They also wanted to make it more scarier juggling a flashlight and a gun, but carmack said looking back that may not have been the best choice and Quake 4 now has the flashlight/gun combo.

- Thinks in next-next-gen (one coming up after this one, ie PS4/Xbox3) that it will be a lot harder to tell the difference between games and CG unless you closely examine it

- Seems really impressed with the 360, said that this is the first time a console has as good a development environment as PC (tools, documentation etc). PS3 has slightly more raw power, but 360 development is much nicer. 360 is the main console for iD development now.

- Skeptical about the Revolution/controller. Carmack said something about it being interesting, but that 90% of the ideas will probably be crap. Another iD guy said they will give it a chance, looking into all the consoles.

- Geoff asked about the drama at id (the founder being fired or whatever) but they kind of avoided the question saying that they are just concentrated on making great games.

- Will continue to make FPS's. Left the door open in the future for more genres, but said they are concentrated on FPS's.

- Wants to release the source code to Commander Keen, but they weren't as organized back then and they can only find bits and pieces of it.

[source: http://www.ga-forum.com/showthread.php?p=2106865 ]