Tripolygon
Banned
I'm just gonna chuck this up to simple misunderstanding.
This is how digital foundry does pixel counting if you're interested.
http://www.eurogamer.net/articles/digitalfoundry-2016-teach-yourself-pixel-counting
-
Hey Guest. Check out your NeoGAF Wrapped 2025 results here!
Mark Cerny on The Verge (GPU, elite controller, etc)
- Thread starter LordOfChaos
- Start date
This is how digital foundry does pixel counting if you're interested.
http://www.eurogamer.net/articles/digitalfoundry-2016-teach-yourself-pixel-counting
Sad Affleck
Member
It is decently capable hardware and first-party developers have the advantage of only coding for one platform. The key point is that the hardware is performing as expected based on the announced specs. Before release the expectations for PS4 games were 1080p/30 fps for most games and that is exactly what we got. For PS4 Pro the expectations based on announced specs are 4K-ish/30fps and that's what we'll get. Don't believe in secret sauces and marketing.
No where am I saying that Checkerboard rendering is the same as native 4K .
The FP16 not being 8.4TF comment don't make sense because it is 8.4TF the problem is people having a set idea of what a TF is without taking into account that FP32 is not the only thing it's used for. It doesn't matter how angry people get because of their lack of understanding PS4 Pro is 8.4TF FP16 that's not up for debate it's a fact.
That's who you remind me of.
Checkerboard rendering is a great way to construct a 4k image, but it's not native and will always be inferior to native 4k.
Your comment tells me that you didn't even understand what you just read because it has nothing to do with my comment that you quoted
I'm not sure if you're implying that dragonelite or you misunderstand how pixel counting is done but that article describes exactly what dragonelite said in more laymen terms. Dragonelite is correct how pixel counting is currently done.
Do you see a game being developed entirely with FP16 calculations?
The problem is not what we do or do not take into account. The problem is that you refuse to acknowledge the industry standard of how performance is measured.
No one is denying that the Pro is capable of FP16 calculations, that is indeed a fact. It's just not the most important performance metric.
They don't calculate Flops by how many is being done they just give you the theoretical peak number of floating point operations that can be done in a second.
The answer to that in FP16 is 8.4TF,
Again TF isn't calculated by how many flops is being done but by how many could be achieved in theory. that number in FP16 is 8.4TF
Me saying that is not me telling you that 8.4TF FP16 is the same as 8.4TF FP32.
LordOfChaos
Member
Yup, the same architecture features that let it outdo even AMDs larger Piledriver cores per-clock also mean it was bound to relatively low clocks. 2.1GHz is right up against where the highest clocked desktop Jaguar part is, and that's where the CPU isn't sharing a thermal load with a PS4 Pro sized GPU.
https://en.wikipedia.org/wiki/Jaguar_(microarchitecture)#Desktop
Sony must have been in a pickle between wanting the Pro to be another PS4 and ensure full compatibility, and as best address the CPU being the weak spot already as best they could without breaking compatibility. So a 30% upclock was what they could manage, even if it went past Jaguars sweet spot for efficiency.
Still find that odd. Who is doing clock dependant programming anymore?
And if someone ties physics to framerate, it would usually be a locked framerate anyways. Wish the CPU at least was kept at the higher clock for regular PS4 games, maybe it could smooth a number of them out (like Bloodborne)
I never said that's how TF is calculated.
You're just being obtuse on purpose, ignoring my point and question because you know it doesn't support your narrative.
It's impossible to have a conversation with you when you pivot and spin like a politician.
You said I refuse to acknowledge the industry standard of how performance is measured & I'm telling you that me saying that PS4 Pro is 8.4TF FP16 is not me refusing to acknowledge the industry standard. A game being made using mixed precision or not using FP16 at all does not change how many TF the console has.
Negotiator
Banned
Apart from AI pathfinding, what other shaders can be ported to FP16 without losing precision?
Tripolygon
Banned
I think you will always loose precision, its just a matter of how precise you want to be with your calculation.
Negotiator
Banned
Not all shaders require 32-bit precision though, right? Am I wrong about AI pathfinding?
Tripolygon
Banned
Correct. As for AI pathfinding i'm as dumb as a rock so i have no idea.
-ImaginaryInsider
Member
You would think making the claim that a 2017 console being half the specifically mentioned teraflops while also somehow being less powerful than a 2016 console would require some kind of well thought-out explanation.
To me, extraordinary claims require extraordinary proof, and it seems like conclusions were jumped to and doggedly defended without adequate explanation as to why MS would need to do all this in the first place.
Much like Jeff, I think he reached a very juicy conclusion first, felt the need to be the first one on GAF to post about it for that "GAF cred", but then never bothered to work backwards as to any really legitimate reason for a company to do this past the vague technicality that they could.
To me, extraordinary claims require extraordinary proof, and it seems like conclusions were jumped to and doggedly defended without adequate explanation as to why MS would need to do all this in the first place.
Much like Jeff, I think he reached a very juicy conclusion first, felt the need to be the first one on GAF to post about it for that "GAF cred", but then never bothered to work backwards as to any really legitimate reason for a company to do this past the vague technicality that they could.
Just because you say it doesn't make it true. Everyone here knows that FP32 is the industry standard. Again you're being obtuse to push your narrative.
Unless a game uses only FP16 calculations, it is a misleading measurement to be throwing around.
You are only spinning things in a way that you can say you were right. On a real tech forum you'd be laughed out of every thread for pushing this silly agenda.
Checkerboard rendering as presented by the Rainbow 6 Team seems a lot more advanced than a fixed rule like bilinear filtering.
To interpolate between a blank checkerboard pixels and it's neighbours it looks at the the direction camera moves, the color values and the depth difference to its neighbours. It also looks at the previous frame's values for all three measures.
It then weights these inputs and determines the final color.
For reference bilinear filtering (=regular scaling) looks only at the neighbouring pixel colors.
So checkerboard rendering has a lot more information to predict the color a pixel will have and it seems at least plausible that it is does a lot better.
antyk
Member
I doubt DF uses the opaque buffer - or any internal graphics pipeline buffer, really - to pixel count. They use what they have access to, i.e. the screen captures and count the stairs on hard geometry lines (obviously not shadows, particles, SSR-s, etc. because those can be of lower resolution to save on bandwidth).
This is silly. 8.4TF FP16 is a spec & that spec doesn't change depending on how it's used it's always going to be the same.
What do you want me to tell you that it's not 8.4TF FP16?
That was actually a rhetorical question.
They use opaque geometry, which would be the hard geometry you mention here. I believe that is what dragonelite meant.
It's a useless spec without actual context to how it'll be used. It's about as relevant as saying the 360 is a 1TF machine like MS originally claimed. You can spin the numbers so that it's technically correct but it's still misleading in the end.
Negotiator
Banned
Damn... I thought only Sony/Nvidia was misleading people with the 1.8TF bullcrap.
http://www.overclockers.com/forums/showthread.php/424359-Xbox-360-1-teraflop
There is context I have never said it was 8.4TF without showing that I'm talking about FP16.
So what's your real beef?
Damn... I thought only Sony/Nvidia was misleading people with the 1.8TF bullcrap.
http://www.overclockers.com/forums/showthread.php/424359-Xbox-360-1-teraflop
They both bullshitted around that time, trying to one up each other with meaningless statistics
Damn... I thought only Sony/Nvidia was misleading people with the 1.8TF bullcrap.
http://www.overclockers.com/forums/showthread.php/424359-Xbox-360-1-teraflop
The funny thing is this only proves my point that you can't just go by the TF number without context which was the point of my post a few months ago that people still haven't got over yet.
No one has ever said a Flops number then explained that it only count if you use it.
Why is this different today? what are you looking for?
Negotiator
Banned
Were they counting non-programmable (i.e. ROPs) flops or something?
False advertising at its finest... I'm glad we're past that bullshit.
Your quote from Cerny didn't say software had to be redesigned. He specifically cited testing as important to ensure that running on a platform with different performance characteristics doesn't expose bugs. So to rephrase your question "who is writing software with bugs anymore?" Answer: everyone who writes non-trivial software, especially anything multi-threaded.
It will certainly be interesting to see if Sony decides that it's worth testing and whitelisting specific games. For now, their approach seems to amount to letting developers decide for themselves whether to put the time and energy into ensuring that their existing titles work well in Pro mode, and while they're at it perhaps putting minimal effort into at least bumping the title's resolution.
Negotiator
Banned
The way I see it, it could potentially break low-level coded (libGNM) games.
Games written in the DX11-esque API (libGNMX) may be able to handle increased clocks.
Can a PS4 dev chime in on this issue?
I'm done, trying to have an actual discussion with you is impossible. Like anyone here who's smarter than me, I'm not going to engage in this nonsense any longer. Have fun living in your bubble of irrelevant statistics.
Ok & have a nice day I'm still not sure of any point that you was trying to make but ok bye.
That attitude was why we had an near 10 year generation. Pure hubris
dragonelite
Member
Correct i thought rasterized geometry would have been explicit enough.
Tripolygon
Banned
I wanted to let this issue go by just chucking it up to a simple misunderstanding but your just insistent on something you are totally wrong about.
You said
So there is literally no room for misinterpretation on my part. You were trying to equate reprojection and checkered rendering to simple upscaling which can be done at the end of the rendering pipeline. And Native means something that isn't scaled, it is independent of screen resolution. Checkered rendering or reprojection is not generated later or done at the end like you insinuated. They are part of the rendering pipeline and not outside of it.
Secondly you said
I pointed you to digital foundry article and video on how they do it. They don't have access to the opaque buffer, all they have access to is the final render buffer that is sent out to the TV. If you said opaque geometry that would make sense. You said opaque buffer.
Edit: I reached out to Digital Foundry to clarify and if i'm wrong, I will post it in a second update.
mitchman
Gold Member
Hey, don't shoot the messenger, but this is an (old) interesting discussion of FP16 in practical use: https://forum.beyond3d.com/threads/...n-gpu-architectures.56180/page-3#post-1805861
Bits N Pieces
Member
Will Sony promote the Pro as being an 8.4TF FP16 machine?
With that reasoning, it's a 16.8TF FP8 machine. Or 33.6TF FP4 machine.
That would be really, really dumb.
Locuza
Member
Did Sony promote the PS4 activly as a 1,8TF machine?
Does it even matter?
But it's not because the ALUs can't process four FP8 operations or eight FP4 operations at once.
It's either one FP32 operation or two FP16 operations, depending on the native format everything below it will still operate with the same processing rate.
I know. But hey, who knows what the future brings.
Still, it's not that you can move everything to FP16. It will have its uses, sure (just as researching and implementing GPGPU to offload a CPU), but it will not be a mechanism that will suddenly make your game go from 30 to 60FPS or something like that.
if it would be a car engine, FP16 operations would be something like an 'extra booster' that you can use with a very specific type of fuel in a specific type of car, with a speciic type of driver, to get to a higher top speed. If you design your car around that principle, it might be a bigger benefit (first party games); if you don't, it's just a nice extra (most third party games).
Negotiator
Banned
You would be right if FP8/FP4 existed and its GPU supported them.
INT8 exists though in some nVidia cards...
So pretty much a developer could use certain pieces utilizing FP16 and some pieces utilizing FP32 if understanding this correctly. It seems like people are making this an either/or when it seems like devs could offload certain processes to be less precise and use FP32 for certain things so in a sense they could reach certain goals vs being stuck with just using one or the other.
or am i off in this case?
or am i off in this case?
Some important processes can be sped up GPU side when optimized for FP16 code, yes, and its definitely a benefit for the GPU when trying to squeeze juice out of it, but it can't be utilized on everything.
Generally, its a side benefit that generally speaking won't even be used all that much IMO, because devs won't be optimizing for pro in a vacuum. The basic utilization of FP32 will be far more than enough to get results on screen and that's generally all devs will aim for, since 99% of the development effort will be on the base PS4.
Locuza
Member
If you look at the practise you may also claim that it's not a 4,2 TF machine because the practical performance is not matching the theoretical output since the game code is hitting various bottlenecks.
It's in theory a 4,2 TF FP32 machine and a 8,4 TF FP16 machine.
Nobody should try to sell explicit performance simply because of that, it's a "stupid" number which only shows one part of the whole picture.
I don't get the arguments for or against the notion of the hardware specs.
Negotiator
Banned
That's right.
This could boost the FP32-only target of 4.2TF to 5-6TF by using mixed precision. Not bad at all.
yeah i figured it may be useful for things that dont need the extra precision like static elements etc if i understand correctly. Also understand now probally would just go for FP32 as much as possible if anything since the ps4 is the base. But very interesting they would allow for that and maybe its to help proof the system for a longer period of time.
yeah its a pretty smart design there based on that premise.
Tripolygon
Banned
That was such a ridiculous comparison, made absolutely no sense. And literally nobody is making the argument that it would make a 30fps game to 60fps.
The bolded is false. FP16 has been used in graphics rendering for a long long time. Heck even games on PS3 used it. Seriously, its not a new thing at all, all that is new is PS4 Pro and future AMD GPUs support 2 FP16 operations per clock cycle. Mobile GPUs support it.
The information came from me so they have to somehow make it wrong.