dahuman said:
|
So like people mentioned early why doesn't Sony eliminate the GPU and put two Cell processors in the PS4 like they originally were going to do? Surely the Cell is cheaper now to get good benifits out of it.
not even the PS1/N64 peaked because people could still find ways to make the games look better etc. / Everygame that runs slow with fps below 30 peaked the cell when the CPU load reached 100% even if its for 1 nanosecond.
So there is no real peaking of this technology at all because you cant specify what peaked is.
Is it A)reaching the limit of the consoles power or B) is it making better stuff untill it doesnt work anymore?
The first is a stupid assumption since a ton of games reached that limit with inefficient coding, the second will not happen because it will always be possible to make the games look better with better devs and better coding (+ you cant run a console on 100% CPU 100% GPU for 10h+ it will die.)
Cell is Skynet, it was found by Sony engineers in Arnold's arm back in the 80s.
Gilgamesh said:
So like people mentioned early why doesn't Sony eliminate the GPU and put two Cell processors in the PS4 like they originally were going to do? Surely the Cell is cheaper now to get good benifits out of it. |
Like I posted before, modern GPGPU makes it irrelevant, 2 Cells can't hope to compete with something like the Wii U even.
VGKing said:
Which explains why PS3 exclusives look amazing despite its GPU being worse than the Xbox 360s. |
yeah, you know, 360 shouldn't be in this conversation, just saying. PS3 exclusives look nice, and that's fine, I don't know why 360 has to come in to the discussion at all.
Soleron said:
The Cell would fare much better than the GFLOPS number suggests. The GPU achieves this with a bunch of brute force FP units - they have very limited memory access and can't work together on one thread or take normal-looking code. It's like you have a computer on every asteroid in the solar system and they can only communicate by flashing a single LED to each other. Cell was much more CPU-like in memory access, many fewer threads were needed. Cell was good at FP-heavy normal code. Unfortuantely game computing is INT-heavy and game graphics work best on a GPU. It's still THE best at what it does best, but it was never suitable for a game console. Also remember the Cell was designed at 90nm. We will have 28nm in 2013, which will have ~24x the transistor density and so 24x the performance. Allowing for losses, Sony could put 16 Cells for the PS4 CPU and still be able to sell it. |
Yeah, I always wondered why they'd use it for a gaming machine, it's obviously more flexible with code compared to a GPGPU yes, but it was weird as hell for them to use it in a gaming machine. BTW, 28nm process would make the stupid thing expensive again, I like the AMD APU idea much better lol.
Gilgamesh said:
So like people mentioned early why doesn't Sony eliminate the GPU and put two Cell processors in the PS4 like they originally were going to do? Surely the Cell is cheaper now to get good benifits out of it. |
A cell does not behave like a GPU at all, if that makes more sense to you...
When the idea of using two cells in the PS3 was laid out, your typical GPU was a pretty "weak" thing, it had a very few shader groups running at a few hundred MHz at best. So using a 8-core SPU chip at several GHz was "a lot faster" than any existing GPU, a t least on paper (and apparently they tried and couldn't get it to perform reasonably for whatever reasons - hence the GT7800 GPU tacked into the PS3 in the end). Times have changed since. Even a basically mid-range APU like the A10 has a GPU with almost 100 shader groups, running close to 1GHz. You can't stuff so many SPUs in a cell to compete with that nowadays.
The correct path to the PS4 would have been an improved cell chip, 3 enhanced PPUs and 4-6 enhanced SPUs (no need for more SPUs, as the GPU now does the graphics stuff) at 28/32nm (basically leading to a 150mm^2 chip, not much bigger than the cell chip now), and a modern powerful GPU. However, with IBM/Toshiba leaving the cell playfield, they would have had to find ways to actually redesign the thing on their own. They chose to listen to AMD ("we are building the superduperchips") and are now sitting in front of a train wreck, as AMD hasn't really built rockets into their CPUs. It doesn't help either that suddenly everybody thinks how "GPGPU" solves everything and how good an A10 fits into that hype. Forget the hype about GPGPU, it is completely meaningless when talking about game consoles.
A game console has to run a game on a powerful CPU and make pretty pictures with a powerful GPU, there is no place here for the GP in a GPGPU (unless your game is about multiplying large matrices and solving linear equations - I definitley doubt such a game would sell). If we look at the latest rumours, this is what the XBoxNext has, a powerful CPU (FX8xxx) and a modern GPU (HD7xxx). PS4 lands in nowhere land with an A10 (even with a second 6670GPU tacked on).
dahuman said:
Like I posted before, modern GPGPU makes it irrelevant, 2 Cells can't hope to compete with something like the Wii U even. |
umm, no, but hey, why not post.
Someone actually posted the correct answer, so I'll just let you work it out on your own.
drkohler said:
A cell does not behave like a GPU at all, if that makes more sense to you... When the idea of using two cells in the PS3 was laid out, your typical GPU was a pretty "weak" thing, it had a very few shader groups running at a few hundred MHz at best. So using a 8-core SPU chip at several GHz was "a lot faster" than any existing GPU, a t least on paper (and apparently they tried and couldn't get it to perform reasonably for whatever reasons - hence the GT7800 GPU tacked into the PS3 in the end). Times have changed since. Even a basically mid-range APU like the A10 has a GPU with almost 100 shader groups, running close to 1GHz. You can't stuff so many SPUs in a cell to compete with that nowadays. The correct path to the PS4 would have been an improved cell chip, 3 enhanced PPUs and 4-6 enhanced SPUs (no need for more SPUs, as the GPU now does the graphics stuff) at 28/32nm (basically leading to a 150mm^2 chip, not much bigger than the cell chip now), and a modern powerful GPU. However, with IBM/Toshiba leaving the cell playfield, they would have had to find ways to actually redesign the thing on their own. They chose to listen to AMD ("we are building the superduperchips") and are now sitting in front of a train wreck, as AMD hasn't really built rockets into their CPUs. It doesn't help either that suddenly everybody thinks how "GPGPU" solves everything and how good an A10 fits into that hype. Forget the hype about GPGPU, it is completely meaningless when talking about game consoles. A game console has to run a game on a powerful CPU and make pretty pictures with a powerful GPU, there is no place here for the GP in a GPGPU (unless your game is about multiplying large matrices and solving linear equations - I definitley doubt such a game would sell). If we look at the latest rumours, this is what the XBoxNext has, a powerful CPU (FX8xxx) and a modern GPU (HD7xxx). PS4 lands in nowhere land with an A10 (even with a second 6670GPU tacked on). |
Would an APU be easier to design games on then a CPU+GPU console?
Pjams said:
Someone actually posted the correct answer, so I'll just let you work it out on your own. |
You know we are speaking on terms of video games and not general purpose computing right?
About Us |
Terms of Use |
Privacy Policy |
Advertise |
Staff |
Contact
Display As Desktop
Display As Mobile
© 2006-2026 VGChartz Ltd. All rights reserved.
