Is a PPU in Xbox360 still possible?
- Thread starter skilzygw
- Start date
With MPU, I mean something even more generalized than a CPU: a multi-processing unit.
While physics aren't the only growing requirement, they're a big enough part that they could deserve special attention in the system design budget when they finally get their due alongside game logic and game graphics.
While physics aren't the only growing requirement, they're a big enough part that they could deserve special attention in the system design budget when they finally get their due alongside game logic and game graphics.
A CPU is used to coordinate the system and run the application's logic. At some point in the future, though, the benefits from specializing silicon for tasks like graphics and physics will yield less benefit than using the chip budget for more programmable power, and a chip with a lot of general computational power which can be applied efficiently to different types of workloads, an MPU, will replace the need for explicit GPUs and CPUs.
It's just a name thing, really. As graphics get more and more advanced, graphics chips will have to get more and more flexible to allow the new things to be done. Eventually, they'll have to be such a flexible processor that the distinction between them and CPUs will disappear, and this ultra-flexible processor that results will be best for doing graphics or any other kind of work. That will be the MPU.
These seems to come in cycles in computing. Int he 80s, custom chips totally outclassed single processor computing for games. Then the PC caught up with raw power entering 3D. This was changed to hardware acceleration for graphics and sound, and now physics.Lazy8s said:
In the future it looks like Cell or something similar will eventually offer a generalised solution, a mighty processor that solves everything with a software solution. I wonder if a super-scalable architecture would ever be replaced by custom solutions, or will that represent the ultimate in economics? PS4 can be 4 Cells. PS5 can be 16. No need to design custom chips anymore...
pc999:
I was pointing out that a CPU which has to handle other major processing jobs like physics or graphics is more aptly termed a multi-processing unit since some systems could have a CPU but still handle physics and graphics with a PPU and GPU, respectively.
Shifty Geezer:
Your idea was correct; I was just trying to clarify the terminology... sorry, didn't mean to be confusing.
I was pointing out that a CPU which has to handle other major processing jobs like physics or graphics is more aptly termed a multi-processing unit since some systems could have a CPU but still handle physics and graphics with a PPU and GPU, respectively.
Shifty Geezer:
Wouldn't ever want to regress in flexibility for the jobs the processor is already focusing on, so there'd be no need to go back again for that. New kinds of tasks might always come along, though, which could start out benefitting more from specialization.
randycat99
Veteran
I'll break the ice and say that, imo, the general CPU has finally reached a point in history where it will have to permanently relinguish certain dedicated tasks to specialized processing units forever. We are already seeing the precursor indications of this in that proposing a bigger, more powerful, higher clock general CPU is only yielding more unweildy, inefficient designs. So the only direction is to go laterally with clustered general processors. Its duties will gravitate more to general housekeeping and looping the general framework of a software, rather than doing the hardcore number crunching work. The very intensive computation work will be done on specialize processors or stripped-down for speed co-processor clusters. True, the number crunching tasks have historically moved back'n'forth between general CPU and dedicated silicon many times, but I predict this will be the last time. As soon as the software and OS limitations of seamlessly managing an agressively scaling architecture have been overcome (which we are seeing the precursors to now), there simply won't be the impetus to develop a bigger, more powerful single CPU to handle the whole shebang. Architectural scaleability will be far less costly to develop and support, rather than invent a new, larger mega-CPU.
So we'll see CPU and clustered FPU units, just like Cell? What about Intel and AMD? Intel are working on this strategy aren't they. Makes the most sense, unless cutom hardware can still outclass massive programmable floting-point power in some things.
randycat99
Veteran
Well, I won't go so far as to call them "clustered FPU's". 
That might not be doing them justice for what they ultimately do. They'll be small, fast, but not entirely "dumb", either. The crucial component is you'll have a team of them get the work done, naturally.
As for Intel/AMD, I think they will probably go the multicore route (in this case being, their best mega-CPU with duplicity applied) for a while. Then maybe we will see more and more development in the SIMD area (to the point where it can be more autonomous than a simple, "dumb" SIMD unit). Then they'll introduce multiple SIMD's on-core, and this will be heralded as a "great advancement" in desktop technology (except, by then it will have been old news from the respect of other technology circles because that will have been done since 2005/2006). Well, that's my prediction for the future, but by no means am I good at predictions...
That might not be doing them justice for what they ultimately do. They'll be small, fast, but not entirely "dumb", either. The crucial component is you'll have a team of them get the work done, naturally.As for Intel/AMD, I think they will probably go the multicore route (in this case being, their best mega-CPU with duplicity applied) for a while. Then maybe we will see more and more development in the SIMD area (to the point where it can be more autonomous than a simple, "dumb" SIMD unit). Then they'll introduce multiple SIMD's on-core, and this will be heralded as a "great advancement" in desktop technology (except, by then it will have been old news from the respect of other technology circles because that will have been done since 2005/2006). Well, that's my prediction for the future, but by no means am I good at predictions...
I was agreeing that fully programmable silicon will eventually make the best use of chip area for all non-fixed processing requirements. For tasks that have predefined requirements and which wouldn't benefit from any flexibility, specialized silicon would use area more effectively.
As for the chip's arrangement, to use its area most effectively across the whole chip, processing has to be de-centralized in order to be better parallelized for full coverage. The degree of parallelization has to be kept in balance with manageability, though.
As for the chip's arrangement, to use its area most effectively across the whole chip, processing has to be de-centralized in order to be better parallelized for full coverage. The degree of parallelization has to be kept in balance with manageability, though.
Similar threads
