Predict: The Next Generation Console Tech

Status
Not open for further replies.
On one hand we have an insider like bkilian telling us not to expect 680 levels of transistors and flops because of power and other constraints, on the other we have tales of 400mm^2+ SoCs with additional processors and other features. How can you do a 400mm die and still sell the box for console prices? And if you can, why use cheap DDR3 with a narrow bus?

p.s. can someone explain what a "blitter" would do in a modern gpu design?

bkilian also said that flops (and PC things) are not as important in consoles. And he said too that we must try to stop to compare PC components with console hardware, consoles does not need all the PC things.
 
On one hand we have an insider like bkilian telling us not to expect 680 levels of transistors and flops because of power and other constraints, on the other we have tales of 400mm^2+ SoCs with additional processors and other features. How can you do a 400mm die and still sell the box for console prices? And if you can, why use cheap DDR3 with a narrow bus?

p.s. can someone explain what a "blitter" would do in a modern gpu design?

I am wondering, isn't "blitter" technically GPU?
 
bkilian also said that flops (and PC things) are not as important in consoles. And he said too that we must try to stop to compare PC components with console hardware, consoles does not need all the PC things.
That goes without saying, and it's not was I was implying. I am asking if you are adding 100+ mm of fixed function or limited programability hardware to aid a cpu/gpu, wouldn't those resources be better spent on more CUs and more memory bandwidth? Bandaids have diminishing returns.

As for secret sauce, where does it say that just because the tablet/laptop version of a Jaguar CU has 2MB L2 (probably limited by that form factor's TDP) that it also has to be like that in a custom SoC? What if the rumored eDRAM was a much larger L2, shared between the Jaguar cores and GCN CUs?
 
That goes without saying, and it's not was I was implying. I am asking if you are adding 100+ mm of fixed function or limited programability hardware to aid a cpu/gpu, wouldn't those resources be better spent on more CUs and more memory bandwidth? Bandaids have diminishing returns.

As for secret sauce, where does it say that just because the tablet/laptop version of a Jaguar CU has 2MB L2 that it also has to be like that in a custom SoC? What if the rumored eDRAM was a much larger L2, shared between the Jaguar cores and GCN CUs?

All those details are what we need to know xD, and I guess the "leakers" don't know any details apart of core number and memory size.
 
Even if this "secret sauce" exists, I question whether it can universally be applied, what it's limitations are and what burden it places on developers for its use. Are you forced to use ray tracing? Does it only work with forward renderers, or maybe it's limited to deferred rendering? Is it like the 3DS where you get a few hard coded shader effects, but will that mean the Durango gets left behind if Orbis devs start creating more naturalistic or artistically designed shaders that don't fit the fixed function's expected parameters? Will every Durango game have the same "Durango" look? Won't it be a hassle for programmers who made their displeasure at needing to manage hetergeneus cores on the PS3 known far and wide.

The number one rule to understanding all of these questions. Have we heard any negative grumblings from developers that Durango is hard to program for?

Source: http://www.ign.com/articles/2012/06/01/the-next-generation-according-to-game-developers

From a hardware perspective, nearly 80% of respondents said Microsoft’s next console is the easiest to work with, and the overwhelming majority suspect it will be the sales leader over the next five years.

Microsoft makes development tools for the PC and for consoles and have done this for years. I guess they know what they are doing. Shrug.
 
p.s. can someone explain what a "blitter" would do in a modern gpu design?
No idea. The blitter was one of the helper chips in the Amiga (along with the copper, and the paula). It's job was to move large chunks of memory around without involving the CPU. Generally used to copy sprites to the screen. For completeness sake, the copper manipulated the palette registers, and paula was the sound chip, capable of 4 8-bit channels (or with some fiddling, stereo 14 bit) and floppy drive controller. :)
 
How many surveyed were hands on with both platform at that point? How many had only used the MS kits and just assumed they were better? Was any of the functionality from the additional, exotic hardware components even exposed to them at the time of the survey?



I'm sure the tools and environment will be of a high quality. But if the underlying hardware of Durango is more complex than Orbis, that's not something tools will magically solve. So either it will be up to programmers to manage these new features, or their effect will be highly limited because they are so automated.

I can't answer any of those questions and I am sure you won't get answers until much latter with NDAs are up. However, I can say this......

If Microsoft has good documentation with sample code I am sure developers love this stuff. They love being able to take advantage of new features and so if it's easy to use in their code with plenty of examples and documentation and it helps them evolve their game then why not?

Great tools, great environments, easy to debug and test, and of course plenty of documentation with samples helps a lot.

So, I am not sure where you are going with this. We haven't heard any complaints and we will see again after this years GDC (I wish it was in Feb like it used to be in years past).
 
I'm simply questioning the blind faith so many people have developed, literally overnight, in rumored custom miracle technologies in Durango. Your blithe "MS has historically good development tools" response does not actually address my long list of questions above. We should be at the beginning of the discussion. We know next to nothing about whether these hardware enhancements exist, what they actually do, how useful they will be in real world scenarios, their cost in terms of programming complexity, their flexibility to be used in different kinds of engines and scenarios, etc, etc, etc.
 
To layman like me it sounds like a controlled leak just to see where it comes out.

Seems a really high level leak. All the the talk about Blitter, DSPs, and other custom special sauce reads like the Atari Jaguar design philosophy. John Mathieson was the lead engineer on those designs and he is a high level designer at Nvidia.
 
I'm simply questioning the blind faith so many people have developed, literally overnight, in rumored custom miracle technologies in Durango. Your blithe "MS has historically good development tools" response does not actually address my long list of questions above. We should be at the beginning of the discussion. We know next to nothing about whether these hardware enhancements exist, what they actually do, how useful they will be in real world scenarios, their cost in terms of programming complexity, their flexibility to be used in different kinds of engines and scenarios, etc, etc, etc.
This isn't the thread to discuss how new techs could be used. It's just about identifying what components are in there.
 
That goes without saying, and it's not was I was implying. I am asking if you are adding 100+ mm of fixed function or limited programability hardware to aid a cpu/gpu, wouldn't those resources be better spent on more CUs and more memory bandwidth? Bandaids have diminishing returns.
You're better off discussing that here: What custom hardware features could benefit a console? Custom hardware != 'bandaid'!
 
A bit blitter, in the Amiga sense, wouldn't make any kind of sense.

A flexible DMA engine with complex source/destination address generation to move data around (with swizzle support) might make sense

Cheers
 
yukon.jpg
This matches pretty well with the 3 dsp information.
1) HW audiovisual acceleration
2) HW scaler (and something more?)
3) HW Kinect computation acceleration
4) for the extra mystery sauce, edram/esram.

Those 3 PPC-cores for backward compatability, if they're in the final design, could be one place where the mm2's partially disappear.
 
A bit blitter, in the Amiga sense, wouldn't make any kind of sense.

A flexible DMA engine with complex source/destination address generation to move data around (with swizzle support) might make sense

Cheers

I'm not exactly knowledgeable when it comes to the finer details on computer architectures (CPU, GPU, DSP, FPGA etc.), but 'swizzle support' sounds like someone is yanking my chain :p.

"Call now to get this low price on this high quality flexible DMA engine with swizzle support!" The sales presenter said with a boisterous tone, "And if you call right now we'll deliver one extra, for FREE!!" After which they bring the TellSell phone number on screen. :p

Naw, just messing around. Just to be a little serious, what is swizzle support in your case? Or are you yanking our chain? :p
 
Status
Not open for further replies.
Back
Top