PlayStation Shading Language for PS4
GDC Europe 2013: http://schedule2013.gdceurope.com/session-id/825424
GDC Europe 2013: http://schedule2013.gdceurope.com/session-id/825424
PlayStation 4 (codename Orbis) technical hardware investigation (news and rumours)
- Thread starter Love_In_Rio
- Start date
- Status
OpenGL makes it easier for devs, but will be more restrictive than Sony's low-level libraries. I think OGL will be important for a lot of lower tier games though, like download titles. It'll make something like Unity integration much easier.
anexanhume
Veteran
My thoughts too. Goes back to Cerny harping on the much smaller time to triangle for PS4.
As others here have pointed out it is already implemented in OGL as 'Sparse Textures' and as the spec text says it was an AMD extension (which can be used by NV but they call it summat else). Of course Sony/AMD will need to extend libGCM or PSSL to support it but I'd imagine it's easier to implement in a new language than design from scratch
http://www.opengl.org/registry/specs/AMD/sparse_texture.txt
Mynd just posted this on PSU when talking to a poster named Foraeli
This statement interests me. I thought the idea that there was some division of GPU resources and\or the architecture was not suited to have those other 4 CU's do normal rendering workload like the other 14 was bunk?
Mynd seems to suggest that they're not really set up in a way to do the same kind of work as the majority of the GPU as optimally.. Due to some kind of ROP deficiency?
Again, Mynd is going off publicly released information. Not inside knowledge.
This statement interests me. I thought the idea that there was some division of GPU resources and\or the architecture was not suited to have those other 4 CU's do normal rendering workload like the other 14 was bunk?
Mynd seems to suggest that they're not really set up in a way to do the same kind of work as the majority of the GPU as optimally.. Due to some kind of ROP deficiency?
Again, Mynd is going off publicly released information. Not inside knowledge.
I think Mynd misunderstand's PS4's architecture. Those rumours were never substantiated, and everything official contradicts them. The consensus is that PS4 has 18 CUs, without any limits in functionality. At this point it's probably best if his opinions weren't posted in this forum unless he posts them directly, as you can't hold a decent discussion across two different domains. 
PS4 is supposed to have 32 ROPs. Should be more than enough relative to the CUs and system bandwidth. If anything, I thought the system bandwidth was supposed to be low for a 32 ROP part. In other words, there should be more than enough ROPs, I think.
Imo they should give up on low level optimizations that tie them to any specific hardware and compromise compatibility with future products.
Then some piece of software are actually doing a good job at exploiting resource of a system.
Imo I think they won't do it but not granting access to close the the metal is a mistake.
anexanhume
Veteran
Why? BC is no longer guaranteed anymore. Gakai is supposedly going to fill their BC gap, and who's to say future hardware won't be identical enough, or future versions of libGCM can't interpret PS4 game code (or do some dynamic recompilation)?
To discourage to-the-metal optimization is to dispense one of the advantages of consoles.
I think Mynd misunderstand's PS4's architecture. Those rumours were never substantiated, and everything official contradicts them. The consensus is that PS4 has 18 CUs, without any limits in functionality. At this point it's probably best if his opinions weren't posted in this forum unless he posts them directly, as you can't hold a decent discussion across two different domains.
Apologies, def won't happen again. I can see how that would be problematic.
Probably compute and scheduling related extensions ?
In his first interview, Cerny mentioned that developers have to code using more than one language systems to use the GPGPU efficiently. I assume that means compiling into different binaries and then bundle them together (GPU and CPU parts ?) explicitly.
Their goal is to be able to code using just one language system to exploit the GPU "fully".
I do not believe in it, and it think running Gakkai will prove incredibly costly, it is imo a loosing approach, you have to burn more power "in the cloud" than to run the software locally, there move the data around, etc.
Onlive is not doing well, Gakkai was searching for an acquisition and succeed but the business model is as unproved and risky as it get.
Well it is disputable as Andrew L did (at least in some regards), especially in a console environment where you don't have actor X that design a APi for every and then every other vendors that run its drivers on its hardware for many different confs, etc.
I would that a single actors could do something both thinner/lighter and better.
Then there is the assunmption that programers always does better, I read a presentation about ISPC Andrew gave a while ago, the compiler usually keeps up with people coding in intrinsic (really light penalty) and sometime wins.
Actually if you run something as gakkai you may want to be able to run your code without having to port code to better hardware when it gets available.
It offers greater option than the usual constrained shrink consoles manufacturers has to deal with to reduce price.
I think that the loss is worse it, cloud or not.
Last edited by a moderator:
The thing tha surprised me is the PS3 game streaming. They probably have custom built Gaikai servers for PS3 games.
Those devices you mentioned will need a controller for PS3 games. And their sheer number may require a huge infrastructure from the get go. But yes, that's likely the eventual goal, perhaps for smaller games.
Those devices you mentioned will need a controller for PS3 games. And their sheer number may require a huge infrastructure from the get go. But yes, that's likely the eventual goal, perhaps for smaller games.
They haven't said much about gaikai, it's puzzling. The 22nm cell is almost confirmed, and there's a rumor of an RSX at 28nm. It could explain why Gaikai will be deployed only in 2014. They can probably fit a LOT of these in a 42U rack, I'd guess 500 per rack, maybe using an existing x86 backplane for the service side which gaikai already developed? They'd be in high demand only for the first two years of the generation until people stop being interested in BC. Usually BC is great because the new console doesn't have enough games, it doesn't look like it's going to be the case with PS4, hopefully.
I'm a bit angry they didn't announce a BC module for PS4 and only mentioned gaikai as a solution. Oh well, my PS3 slim will fit nicely on top of the PS4, and this combo will have a space requirement that isn't any bigger than an xbone (if you do the math correctly). So I can't complain much.
I'm a bit angry they didn't announce a BC module for PS4 and only mentioned gaikai as a solution. Oh well, my PS3 slim will fit nicely on top of the PS4, and this combo will have a space requirement that isn't any bigger than an xbone (if you do the math correctly). So I can't complain much.
This sounds like a plausible theory. And seeing how small the PS4 is, now that according to Mr Fox the 22nm Cell is about to become a reality, they could attach the PS3 motherboard to the PS4 and create an entire new box from it. It's not like they don't have experience with a similar approach.
If that really is the long term plan, then you shouldn't buy any of your games on physical disks. Gaikai will be the carrot to go all digital.
- Status
