AMD's 360 and Wii GPU Guru Speaks: Future Game Hardware To Combine CPU, GPU

Both the Xbox 360 and the Wii have graphics processing units designed by AMD. After the recent roundtable discussion on the future of console gaming chips, Gamasutra had a chance to sit down with Bob Feldstein, VP of strategic development at the company, and Jon Carvill, PR for graphics. The discussion touched on the philosophy behind console GPU development, as well as the present and future of the game industry through the eyes of one of its most important collaborators.

Bob Feldstein, provided some insight on the future of console gaming chips, speculating that GPUs and CPUs may someday be a single entity:

"I'll say that what I see the future as is a combined CPU/GPU, where the delineations are not there anymore. You combine them into a chip; you have compute elements of both, and jobs get kind of -- threads get kind of put on the resources that are best suited for them. GPU-like resources, vector engines for problems that simulate the real world, graphics, and physics. And for sequential problems, about other parts of this, you know, you've got your typical sequential processor."

Read Full Story >>
The story is too old to be commented.
Vojkan4870d ago

"Future Game Hardware To Combine CPU, GPU" - funny, Sony is doing that with CELL, but ask xbots and they will tell you its otherwise.

cooke154870d ago

It has not worked too well with Cell though. It was a good idea but poorly designed. As you can see PS3 seems to have the worse version of multiplats. Im gonna get flamed for this but its the truth, developers are even saying it.

Xi4870d ago

the ps3 had an nvidia card in it. Ie a gpu, keep trying though.

BrotherNick4870d ago (Edited 4870d ago )

xi is right :P but sometimes people just don't know about stuff they aren't particularly interested in. People work too hard to bash the consoles lol.

Xi4870d ago

the cell is a general purpose cpu, in that it can preform many functions of both a gpu and cpu, but a jack of all trades is a master of none.

However what he's talking about is a multicore processing unit that has both types of di on it.

Ju4870d ago

The CELL was a first attempt and way ahead of its time. Obviously it was not good enough on the GPU side to compete against a discrete chip at the time.

What that AMD guy's talking about is simply the next step on a possible CELL roadmap. This can be achieved to add discrete shader elements to the CELL core or improve (and widen) the SPUs to catch up to stream processing bandwidth (either due to widening the instruction set or increase the number of SPUs on die).

The disadvantage that IBM (STI) have is, that they have not the background AMD has in graphics processing (due to the take over of ATI). So, either they'll buy one company (NVidea - they'll never sell, though) or they try to get that knowledge from somewhere else. But, yeah, a next gen CELL could look like that - as well as a future AMD gaming processor (with an x86 ISA).

Xi4870d ago

I think he's implying that they'd have a specialized chipset such that it'd work like the spe's in the cell but specialized so that one would work moreso like a gpu and the others like a cpu, probably with a buffer. The threading would be a mess to deal with.

Ju4870d ago

Nah, they have nothing - yet. It just "down the road". He just takes a guess, that the GPU will eventually melt into the CPU, and then have stream processing and sequential processing close together. My imagination is not big enough to be able to guess how he thinks a HW piece can distribute between sequential and streaming, but he thinks, ATI used automated HW threading all the time (simply because all shaders used a more HL approach and you usually don't hit the metal, which allows the driver to handle the threading).
The CELL is a step in that direction. The SPUs were meant to replace the stream processors in the GpGPU, but have less parallel units, but more complex ISAs to make up for that. The design is still 6 years old, I'd guess. In that time, GpGPUs made a huge step forward, which the SPUs couldn't keep up with. That's all. But eventually it might merge into something similar. Imagine 1000 spu like copros on the CPU die, plus four general purpose cores. Perfect for a gaming console. You wouldn't replace that piece of HW in the 10 years live cycle.

For the people worried about MS-Word (or office), such a CPU would run cycles around anything what's available today - and would probably be dirt cheap. If you really feel the urge to upgrade because this might be outdated again soon, well, yeah...why would you? Buy a new MB. The funny thing is, as soon as they'll have that tech, it will be over. Nobody will upgrade anymore after they'll release that.

+ Show (4) more repliesLast reply 4870d ago
mighty_douche4870d ago (Edited 4870d ago )

if the 2 were to be combined it would just be to rip off consumers. look at it this way, if you do alot of word processing and have multiple windows open at once you may want a decend quad/duo cpu but have no need for a high end gpu. doing it this way means you'll only be able to buy a high end cpu with a high end gpu built in, what if you dont want that gpu?

also, this would make upgrading far more expensive, as things stand now, i can upgrapde my cpu, wait a few months till i have more money then upgrade my gpu, rather than being forced into doing both at the same time...

plus makes all current mobo's, gpu's, cpu's obsolete! which = alot of unhappy pc user, me included.

sume it up... STUPID IDEA!

EDIT @ Xi, cheers for the reply dude, i hope it stays that way as i dont wanna be upgrading anytime soon! completely missed the word 'console' oops, my bad.

Xi4870d ago

pc are obviously going multi core as cpu's can't get much faster than the 5 op's per clock cycle they're at now.

season0074870d ago

2nd...thosearent multi plats
PC, PC variant, PS3? more like just 2 platforms to me

Bonsai12144870d ago

basically what the cell was designed to do but it fell short. and by short, i mean "unable to put textures onto established geometry." because thats the main purpose of the RSX, is to texture things. everything else is handled by cell. so they're almost getting there.

Ju4870d ago (Edited 4870d ago )

Its not that bad. I am experimenting with a SW renderer (just a bit research) which is supposed to run geometry on 2 SPUs and the shader on the other 4 - in parallel. The geometry engine is designed to partition a vertex array and transform each partition at the same time. The shader runs in parallel. The vertex array is double buffered. That works completely transparent, which means the app running on the PPU still does its glVertex() (or actuall pretty similar), but that's where it ends.
I have no idea how far the shader unit can go (beyond flat shading :) - but I'll find out for sure. This would scale to any number of SPUs. So, give me 24, and we're talking real performance here.

Edit: That's funny. I got a disagree. Seems I have an imaginary friend who knows better then me, what I'm doing.

Show all comments (20)
The story is too old to be commented.