By using this site, you agree to our Privacy Policy and our Terms of Use. Close
JustBeingReal on 24 November 2014
Edit Thread
walsufnir said:
JustBeingReal said:
walsufnir said:
 


But you can't use it if other resources or channels are already congested. I don't say it can't be used but the GPU is restricted and dependant of other resources and these are also used by CPU and by GPU contexts so we will have to see to what extent devs will be able to use it but I am not sure if it is that much. There will be a trade-off between using the GPU for graphics or GPGPU, you can't have both in that case without penalting the other.

 

That's the point of the way the whole APU has been designed, developers can easily queue up compute commands, have them waiting for when GPU hardware is free to process it. The GPU is never completely saturated dealing with graphical tasks, there's always tonnes of times when Stream Processors are sitting idle, so this is when AI and Physics are handled. Remember I said efficiency?

You should be sure that developers can use this to a great level of extent, because that's what they asked for when Sony did all of their research in what hardware to put in PS4.

There's no need for a trade off with this because it's part of the system's design to use resources as efficiently as possible.


Waiting for the GPU hardware to be free to process data? In a game where devs are fighting for milliseconds to get their stuff done (ie. computed)? How would that work? You need your data fast and in a predictable way - what if you need the data now and can't wait for another bunch of milliseconds?

I am not sure of anything until a dev talks freely about it and provides data on how and when the GPU was free enough to process data for a decent amount of time when it is not limiting the render tasks and what can be accomplished when doing so.

Btw, how would you know to which extend PS4's GPU is currently utilized? There are clearly games that are not as stable as some wish so why wouldn't devs use the hardware not to 100%? This makes no sense at all.

 

It works because even in those instances where the GPU is under heavy load there are always Stream Processors free, the developers plan for this stuff in the development process.

It's part of the GPU hardware, not all of it in it's entirety. What you have to remember is that it's never the whole of the GPU doing one task, the GPU is running mulitple tasks simultaneously. Games are designed around all of their aspects from graphics to compute, you're acting like it's a one or the other type situation and completely missing the point that graphics never take up all of GPU's time. It's all a matter of proper resource management, when the game's being developed.

 

You should be sure that this is a great thing, if you did research about it, it actually becomes self evident.

 

This is simple, it's efficient use of resources, not wasting it when it could be put to use.

The reason I know this stuff is from research and common sense. The reason games aren't as stable as they could be is because of a lack of optimization, in the design of the games and how optimized the development tools are at this point in time. This new SDK directly addresses the physics side of shared compute load between the CPU and GPU.