• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Inside Playstation 4 Pro: How sony made the first 4k games console

cooldawn

Member
Gotta sell those new tv models.
Right...but it's not that simple.

TV technology would have progressed anyway but more importantly Sony seems hell-bent on pushing console technology. Look back to the PS3; Polyphony Digital Inc. had technological desires beyond capability (1080p mode).

I believe this is just Sony saying 'look at how clever we are'. Sony have always been know to push the boundaries but sometimes they want to push it too far. It's the same here so I think the 4K messaging is purely for gamers i.e. a 4K capable PS4 Pro won't add an exponential amount of sales to it's own 4K TV division.

4K TV's will organically become more popular anyway.

The PS4 Pro version of Paragon is only going up to 1080p and pushing fidelity as much as it can at 60fps. Probably a rare case but it's at least one example of it.
For now, paragon is the technical showcase every ps4 pro version of games should strive for.

900p -> 1080p bump, maybe even higher?
60fps
Added graphical effects. I noticed per object motion blur, not sure if the ps4 version has this.
More detailed environments (just look at the rocks and added grass)

ps4fasrf.png

ps4proz4sm9.png
I know it's illustrative but it doesn't look all that great anyway i.e. that's not a standard I'd use...but then again I have yet to seen anything I would say is vastly improved, or at least improved to a degree that gives an impression of the additional capability.

I hope it is a sign other developers can improve 1080p titles though, especially if it only takes 0.03% of resources to do so.

That is like asking for a PS5, right now. Not happening. Sony still believes in the PS5, and have made clear since the Pro was first leaked that the Pro was not intended to be that, but a PC like option, where the same games as the PS4 are run at a higher fidelity.

At the very most, you'll see 1080p enhanced titles of already made games, where higher graphical settings that already exist on PC are enabled on console, like Tomb Raider is doing.
Sorry, I don't understand why developing a full-on 1080p title for the PS4 Pro is anything like developing for an entirely new platform. I've tried to understand the context by reading through it a couple ion times but I can't see it : (

The ID buffer & other changers make me feel the same way but I think devs will still learn new techniques even with PS4 Pro being more of a enhanced PS4.
Yeah, I'm hoping developers resources are made easy so they can better manage and implement improvements to a PS4 Pro version.

Could a game be developed for the PS4 Pro and then drop code for the PS4 version to keep performance parity?

Something similar could happen to some extent, by pushing the Pro as a lead platform and rendering games on the standard PS4 at lower resolutions or by utilizing dynamic resolutions. A lot more can be done with the new hardware.
Oh...there it is.

The more frequent new consoles are released the less ground-up developed titles you will see...
This is what I'm afraid of and why I don't like this iterative strategy because I feel when a fixed platform starts to move about it's less important. At least with clear generational gaps there's a defined purpose to a platform.
 

onQ123

Member
All rendering is happening in parallel unless we're talking about a rendering process requiring specific order. With temporal accumulation you're using previous frame data for parts of the current frame, so this is the process which can't be happening in parallel since you have to render the previous frame before you will be able to use this data in the current one.

And this filling of blanks is unlikely to happen in parallel to the actual rendering of current frame quads either as you most likely need the current frame data to perform proper temporal reconstruction of missing parts using the previous frame data.

So you have:
A. Render 1st frame checkerboard of odd quads
B. Render 2nd frame checkerboard of even quads
C. Combine 2nd frame with 1st one adjusting the 1st one by motion vectors / ID buffer / whatever
D. Present
E. Render 3rd frame checkerboard of odd quads
F. Combine 3rd frame with 2nd one adjusting the 2nd one by bla bla bla
G. Present
H. Etc.

This isn't parallel at all.

And ID buffer doesn't really sound like something new, geometry ids like this are typical on TBDRs and I'm pretty sure that NV is using them in Maxwell+ for their TBIR approach - and this is quite probably the basis for MFAA they have in Maxwell+. It should help with temporal accumulation for sure but it's not magic, and there's certainly no "re-rendering" of pixels since the whole idea of checkerboarding rendering is in splitting the rendering between two frames since PS4Pro isn't powerful enough to handle 4K rendering in one frame.

I'm saying that the pixels needed to fill in the blank is being rerendered in parallel with the new pixels.

Say that the 1920x2160 pixels from the fixed function pipeline was 1 & 3 but the compute rendering using the ID buffer for color propagation & data from the previous frame was 2 & 4. this would be happening in parallel

(Not Sony's patent)

SCALABLE GEOMETRY PROCESSING WITHIN A CHECKERBOARD MULTI-GPU CONFIGURATION

Read more: http://www.patentsencyclopedia.com/app/20140306949#ixzz4OIB8Q5S6
20140306949_03.png
 

pottuvoi

Banned
I'm saying that the pixels needed to fill in the blank is being rerendered in parallel with the new pixels.

Say that the 1920x2160 pixels from the fixed function pipeline was 1 & 3 but the compute rendering using the ID buffer for color propagation & data from the previous frame was 2 & 4. this would be happening in parallel
Yes, post processing pass can happen when developers choose and when it fits their pipeline.
 

dr_rus

Member
I'm saying that the pixels needed to fill in the blank is being rerendered in parallel with the new pixels.

Say that the 1920x2160 pixels from the fixed function pipeline was 1 & 3 but the compute rendering using the ID buffer for color propagation & data from the previous frame was 2 & 4. this would be happening in parallel

(Not Sony's patent)

SCALABLE GEOMETRY PROCESSING WITHIN A CHECKERBOARD MULTI-GPU CONFIGURATION

Read more: http://www.patentsencyclopedia.com/app/20140306949#ixzz4OIB8Q5S6
20140306949_03.png

Nothing is being "re-rendered", previous frame info is adjusted spatially by motion vectors and whatever they use and then used in the same form it was present in the previous frame.

This adjustment can happen any time but if it's using current frame data for determining motion vectors then it will happen after current frame quads will be rendered.

This patent is irrelevant, there are no multiple GPUs in PS4Pro.
 
http://www.anandtech.com/show/8460/amd-radeon-r9-285-review/3
http://images.hardwarecanucks.com/image//skymtl/GPU/RX-480-ABC/RX-480-ABC-6.PNG
http://images.anandtech.com/doci/10446/P11.png

1) Improved tesselation probably explains why only the Pro version of RoTR has tesselation enabled. KZ-SF also had it on OG PS4 though...

2) Polaris delta color compression offers a 40% boost. That means that the 218GB/s figure is the equivalent of 305GB/s for the old GCN 1.1 design.

It says it doesn't support more than 256GB/s though, which is weird, considering the fact that Scorpio sports a 320GB/s pipe.
 

dr_rus

Member
It says it doesn't support more than 256GB/s though, which is weird, considering the fact that Scorpio sports a 320GB/s pipe.

The bandwidth figure is for the Polaris 10 GPU. Scorpio is unlikely to even be based on Polaris but if it will - nothing stopping them from using a wider bus.
 
The bandwidth figure is for the Polaris 10 GPU. Scorpio is unlikely to even be based on Polaris but if it will - nothing stopping them from using a wider bus.
I thought that the (de)compression engine didn't support more than 256GB/s of mem bw.

If Scorpio supports delta color compression, then we're talking about 448GB/s of effective bandwidth...
 
Maybe but MS need to say more than 6TF because if they are using a Vega GPU 6TF could mean 3TF FP32 / 6TF FP16.
They have given us more information. They've also given us ALU count (2560) and clock speed (1172MHz). GPU ALUs are capable of two FP32 operations per cycle, hence 2560 x 1172 x 2 = 6,000,640 MFLOPS = 6TF. Unless the cores used are only capable of 1 FLOP per cycle (the probability of which is zero), it's 6TF FP32.
 

onQ123

Member
They have given us more information. They've also given us ALU count (2560) and clock speed (1172MHz). GPU ALUs are capable of two FP32 operations per cycle, hence 2560 x 1172 x 2 = 6,000,640 MFLOPS = 6TF. Unless the cores used are only capable of 1 FLOP per cycle (the probability of which is zero), it's 6TF FP32.

did you miss the part that says "if they are using a Vega GPU"


Also why did you purposely remove the rest of the quote?

Maybe but MS need to say more than 6TF because if they are using a Vega GPU 6TF could mean 3TF FP32 / 6TF FP16 or it could mean 6TF FP32 / 12TF FP32.
 
Top Bottom