vg247-PS4: new kits shipping now, AMD A10 used as base, final version next summer

RoboPlato

I'd be in the dick
Quick question about IO RAM, do we know the configurations it could come in? If it is potentially used in the final system, could there maybe be a little bit extra for OS stuff on top of a 4Gb target for games?
 
Not really. In a longer run (and consoles are a long living products) it's better to have a UMA architecture with simplier board than a NUMA architecture with several memory buses/types on board. Memory prices will go down, board complexity is fixed cost mostly.


It's not weird for a platform holder with financial troubles to shoot for the cheapest setup possible.


There are some things you can do in a closed console hardware dedicated to just one main goal and intended to work in a rather small range of screen resolutions. Off chip eDRAM is one such thing, large on chip memory is another. It all comes down to having more bandwidth for less money anyway and that's probably the only place where we can expect something that's not avialable in a high end PC space.
On a side note, I don't believe that even with these customisations next gen consoles will be able to beat high end PCs of 2013 in raw processing power.

UMA might be better but using 1/4 of your GDDR5 memory for the OS is just a fools' choice.
It's paying a cost with no return, no one would make such a stupid choice.
Also Sony's financial situation has absolutely nothing to do with PS4, the memory they're going to put it they're not giving it for free, it really depends on the product they want to make (also it's not even the gaming division which is losing money).
Having 2GB of memory fully dedicated to games would be one thing, having 2GB of total memory is an other, that is just not enough, even people with little tech knowledge understand that, go figure the top engineers in the world :D
Also of course there can be hardware optimizations specific to a closed architecture, but they won't do magic as trasforming a relatively shitty 1.2Teraflops GPU (if that's the case) into high end stuff like the GTX680, also adding more things and different hardware modules as 'helpers' put a burden onto developers to understand how to use those things properly leading to very different results depending on programming skills, budget and other factors. A recent example was the free anti-aliasing using tiles and the embedded RAM which the 360 was supposed to have, that was an ''optimization'' too and almost no one bothered or mastered it.

Quick question about IO RAM, do we know the configurations it could come in? If it is potentially used in the final system, could there maybe be a little bit extra for OS stuff on top of a 4Gb target for games?
That's what I'm thinking as well, they can have a single pool of fast memory which acts as UMA just for games but there will be a side pool of memory that no one has mentioned so far for the OS and multitasking applications. It's the only thing that makes sense to solve the issue of balacing capacity/bandwidth/costs.
 
Simple explaination please:

What are general benefits of APU with all mentioned in comparison to (i.e.) Intel i_ processor and dedicated gpu in a closed system?
 
Rumor is a rumor, we still don't know the bus width they're using and which kind of modules they will be using.
The former of course has been already decided, the latter should be finalized really late depending on costs and contracts with suppliers.
Also using a 1024 bus with DDR is kinda insane, it misses the point of using DDR in first place.

Right now unfortunately we still haven't clear infos on the memory system or at least something that makes sense (and both 2GB of GDDR5 and 4GB GDDR5 as total memory don't make sense for different reasons).
It's just speculation on my part but I think that the best and most obvious thing to do is to have a memory split which still is unified for game programming purposes.
OS datas, background and non-gaming applications datas will go into a smaller pool (2GB) of cheap DDR3 memory over a 128 bit bus accessed by the CPU.
All the game datas will go into a unified bigger pool of GDDR5 memory which is accessed by both the CPU and GPU through a wider bus (256bit at least).
Still we'll see what they come up with =)
Yes, I am most interested in bus and memory layout.

Followed by any dedicated hardware for:
* small/no latency natural input @ 60Hz (voice, motion, biometric, etc)
* media processing (e.g., H.265, audio, image and video processing)
* being able to do all of the above with 3D graphics at the same time

At system level
* Any breakthrough in handling 3D data now that the CPU and GPU are closer, with ample memory (e.g., how easy to generate 3D visual with little developer work, can we generate accurate polygon model based on depth camera input on-the-fly ? 1080p x 2 players x 3D x 30/60fps with destructible/morphable world ? ...)
* ease of development

Others:
* Meaningful controller enhancements
 
Simple explaination please:

What are general benefits of APU with all mentioned in comparison to (i.e.) Intel i_ processor and discreet gpu in a closed system?
I believe that the big bonus is that since it's on this same package (like the cpu and gpu are stuck together sort of), latency is highly reduced, and hardware revisions are the stuff of dreams for the manufacturers.

Don't know if it's exactly the greatest quality from a consumer point of view though...
 
I think it can't be overstated what a waste it would be to use a chunk of that expensive GDDR5 for OS, reducing the amount available to games to perhaps 3GB. In a few years time, it'll look a lot less enticing for development. Then again, we must remember that the earliest rumour came with 2GB GDDR5. Now we can assume that that amount was prior to deduction for OS. So following that logic, the first generation/launch games were being designed with around <2GB in mind. So if anything, these devs are experiencing a surplus now. As aforementioned, it's a rosy situation that may quickly sour as developers try to get extract more from the system. It's also a reason why I think there may be more to this. I know it's Sony and they generally make the best or worst of choices but can they be really this short sighted (unless they think the memory footprint of the OS, much like this gen, can be reduced significantly)?
What do you mean by the OS? Any game will have a lot of OS calls, whether it be to the networking stack during any online game, the drives, the controller, probably even the GPU. Putting those parts of the OS in a different and much slower part of memory will cause games to grind to a screeching halt every time there's an OS call. Going lower-level on OS functions like threading and memory management makes it even more critical that it's in higher speed RAM.

If you mean background media apps (like Music Unlimited for your custom soundtracks, like on Vita), then the rumored special media processing chip(s) (I'm tempted to just refer to it as the Cell) will likely handle most of the grunt work while the memory-hogging app front-end could be swapped to local storage while it's in the background.

The main thing that's going to be tough is having a browser loaded all the time.
 
Just as a fact check - are Wii U's on one die or on two dies in one package?

APU = one die.

Two distinct dies on one package.


Simple explaination please:

What are general benefits of APU with all mentioned in comparison to (i.e.) Intel i_ processor and dedicated gpu in a closed system?
See the quote from b3d on the last page, when getting the GPU to offload some work from the CPU you then have to move that information back to the CPUs memory, which can take even longer than the computation itself. THat's why few people use it, and when they do they just use it for GPU work to avoid that penalty. On an APU with a unified memory address space, the CPU could near instantly access whatever the GPU just put there. With no memory transfer penalty, that makes offloading non-visual computations to the GPU more worthwhile.

Hasn't really panned out yet since todays APUs are so slow compared to high end discreet cards, but eventually I think that's the way to go once fabrication processes allow.
 
Hasn't really panned out yet since todays APUs are so slow compared to high end discreet cards, but eventually I think that's the way to go once fabrication processes allow.
That's why I asked, if they are generally that slower, does this advantage compensate enough? I know it can't be put in a sentence or two, of course.

Thanx for the answers, we'll see what they can do in a closed system. I really can't wait.
 
It may allow the GPU to work on and switch between multiple data sets quickly since the console may be doing multiple things at the same time, not just playing a traditional game. The CPU and other subsystems can prepare other data sets in the mean time (e.g., overlaying and changing GPU output with web browser controls or natural input control based on app/game logic). Modern OS UI already mix GPU and CPU work together today.

In Cell, I remember you could also DMA I/O data to the LocalStore. In this shared memory setup, the shared memory basically holds all the things the I/O devices, CPU and GPU are working together.
 
Yup, I think they are going to have the unveil before the E3, Sony after all is not new on revealing everything they have just before the E3.
The last years all the software was exposed some week before it.