Predict: The Next Generation Console Tech

[Lycan]

The 1 billion figure was in a way a concoction of the article's author (read : speculation). Sony CTO didn't provide any such number at all...:smile:

 

[antwan]

The 1 billion figure was in a way a concoction of the article's author (read : speculation). Sony CTO didn't provide any such number at all...:smile:

is the 'fact' that Sony will use an AMD solution not speculation as well?

 

[Rangers]

say if the $1B SOC is real & knowing that the CPU & GPU in the SOC is AMD wouldn't that mean a lot of this money is going into whatever else is in this SOC like the co-processor? which is already been shown to be a part of AMD's SOC design letting 3rd parties add their own processors to the SOC design.

what if there is a Ray-Tracing co-processor in the SOC wouldn't that be a really big advantage that a company like Sony would want bad enough to put a lot of money into?

Here come a bunch of nonsense speculation about how Sony is obviously building a supercomputer for PS4 because of the missing 1 billion. Cell all over again

 

[anexanhume]

Here come a bunch of nonsense speculation about how Sony is obviously building a supercomputer for PS4 because of the missing 1 billion. Cell all over again

Seems rather crazy, doesn't it? Especially when the 360 essentially benefited from some of the Cell engineering. I'd imagine Sony doesn't want to repeat that again either.

 

[Lycan]

is the 'fact' that Sony will use an AMD solution not speculation as well?

Indeed. But all rumours are converging, and where there is smoke, there is fire.
By contrast, the 1 billion figure was a number conjured up by the writer of an article. Explicitely referred to as guesswork...:smile:

 

[ERP]

Seems rather crazy, doesn't it? Especially when the 360 essentially benefited from some of the Cell engineering. I'd imagine Sony doesn't want to repeat that again either.

I assume you are referring to the use of processors similar to the PPU's.
They were existing IBM designs that pre existed Sony's involvement.
I's my understanding that IBM pretty much offered them to everyone including Apple and Nintendo.

 

[cteam]

Here come a bunch of nonsense speculation about how Sony is obviously building a supercomputer for PS4 because of the missing 1 billion. Cell all over again

even with sony financial figures many still believes sony
will create a Beast .

again all point to PS4 will go safe route or delayed
inline with recent IGN developer survey

 

[onQ]

even with sony financial figures many still believes sony
will create a Beast .

again all point to PS4 will go safe route or delayed
inline with recent IGN developer survey

in my book 10X the PS3 is a beast

 

[XpiderMX]

in my book 10X the PS3 is a beast

nVidia say current GPUs are 24x Xbox 360

 

[bkilian]

nVidia say current GPUs are 24x Xbox 360

At 4x the power usage

 

[Rangers]

Could be referring to 690 dual GPU as well. In pure teraflops would need to.

I'm interested in Sweeney's "5k teraflops to simulate reality" thing.

Dual 7970 ghz edition could get you 8.6 TF.

4 of then in CrossfireX or whatever could get you up to 36, theoretically.

Couple more die shrinks assuming 2X flops scaling (which hasnt held in recent times, EG 6970>7970) maybe you get up to ~140...

If you assume we've got only 4 shrinks left, you get to 560...

Still an order of magnitude short but getting there, sort of...1 more potential shrink gets you to 1k...using 500mm of die instead of 360 could get you some more...

ANYWAYS

 

[SedentaryJourney]

in my book 10X the PS3 is a beast

Well in pure FLOPs waving, you have 230.4 GFLOPS for CELL and another 400.4 for RSX, so 10x that would be a 6.308 TFLOP AMD SoC.

 

[pjbliverpool]

Could be referring to 690 dual GPU as well. In pure teraflops would need to.

No there not. A single 680 would easily achieve that level of actual improvement over Xenos. Taking it to be the same speed as the 7970GE which itself has over 18x the paper flops of xenos at(according to ERP I think, or one of the major devs here anyway) twice the efficiency then 24x actually sounds conservative. It may have been made in reference to the 580 or even 480.

Regardless though 10x in console terms would be a massive improvement allowing for graphics not seen today on any platform.

 

[Rangers]

Yeah, I was just talking about flops.

 

[SKYSONY]

even with sony financial figures many still believes sony
will create a Beast .

again all point to PS4 will go safe route or delayed
inline with recent IGN developer survey

What recent IGN survey?.

Mind the safe route doesn´t have to mean weak hardware. Sony could perfectly launch a console with a powerful hardware, maybe not the best one, but definitely not weak. I´m sure Sony wants to compete with MV in terms of dates an price.

 

[RedVi]

The motion/whatever controller is the wild card. They could have a very powerful system at $399 break even or small profit this time around, but if they bundle an expensive controller, it will end up like the BluRay drive was for sony in PS3. That was the main reason they had to sell at a loss and most would agree that it was a necessity to win the format war. A controller is arguably a similar necessity, although I don't see them subsidising it like they did BluRay, instead the hardware will suffer a little.

If MS have 8GB of moderate speed RAM and 30-50MB very fast eDRAM and Sony only have 2GB of fast GDDR5, then MS will be at a pretty big advantage IMO. That will be where the battle is won given current spec rumours IMO. 4GB for Sony would definitely be enough to compete considering it will be at least twice the speed. A bundled fancy controller may not allow for this though.

 

[Rootax]

Twice the speed doesn't compensate lack of size... ?

If you have to stock 5go of textures, 8go of slower stuff is better than 4gb of faster stuff, right... ?

Anyway 8go seems a lot in a console to me

 

[Shifty Geezer]

5 GBs of textures stored in RAM at any one time is damned inefficient texturing! See the thread on virtual texturing for details. Textures aren't the only thing to use a lot of RAM on though, especially if you're aiming for a multitasking monster.

 

[Rodéric]

5 GBs of textures stored in RAM at any one time is damned inefficient texturing! See the thread on virtual texturing for details. Textures aren't the only thing to use a lot of RAM on though, especially if you're aiming for a multitasking monster.

That's exactly why all new consoles should have mandatory installation of games on a SSD.
All of a sudden you have virtual texturing (and other resources) working flawlessly.
A shame none of the console manufacturers seems to have gone that way...

 

[sebbbi]

If you have to stock 5go of textures, 8go of slower stuff is better than 4gb of faster stuff, right... ?

Depends on access patterns and how much slower the bigger memory is.

For example, lets assume the double sized memory has half the bandwidth (simple easy example). This of course means that we can access half the memory during a single frame. For example the faster (smaller) memory could for example offer us 500 MB of accessible memory per frame, and the slower (larger) memory would offer us 250 MB per frame.

The important thing: You can actually access twice as much texture data per frame with the smaller twice as fast memory.

If you have a good texture streaming system (for example virtual texturing), you would always prefer to have the faster (and smaller) memory. Of course textures are only a part of the memory consumption of a game. However in current generation games, textures tend to be a pretty big part of it. Larger memory gives developers freedom to utilize techniques that need larger (sparsely accessed) data sets (that do not have predictable access patterns). Larger memory of course also makes development easier (streaming everything on demand with good prediction is a quite hard problem to solve perfectly for all cases).