GeordieMark
Member
What?
-
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 the Scorpio Engine: the processor architecture deep dive
- Thread starter Space_nut
- Start date
What?
Here you go
Scorpio deep dive and 27 pages of PS4 and the fable of the FP16
Ooh, are those half quack or single quack
... or are they double quack?(Thanks for cheering me up
)Don't forget iphones, ipads and android.
I can kinda see this happening. I wouldn't be surprised either way.
I think he wants an Xbox Switch.
transformer
Member
No need to be cautious. Turn 10 already has Forza tech running 4K/60 with ultra settings on Scorpio. Still time to optimize as well. No need to deviate from 4K/60 at this point.
LordOfChaos
Member
I don't think FP16 shaders are much of a programming challenge like say SPUs were, it's more of a "how much do we write cookie cutter code for everyone vs tailor one platform to its strengths". Certainly exclusives will take advantage of rapid packed math, while cross platform may have fairly portable shader programs between them.
dr_rus
Member
A large pool of programmers and artists are working with FP16 shaders right now. In fact, it's probably a larger pool than that of them working with FP32 on modern consoles and PCs.
Pasedo
Member
What's the point of making just another more powerful Xbox and mimik the PS4 Pro. They should add something different like giving it a different form factor such as a built in screen. This would mean they can compete and take share from the rising gaming laptop market. Perhaps it will be called the Xbox Portable?
Here you go
Scorpio deep dive and 27 pages of PS4 and the fable of the FP16
Now I want Super Rub-a-Dub 2
Fabtacular
Member
A large pool of programmers and artists are working with FP16 shaders right now. In fact, it's probably a larger pool than that of them working with FP32 on modern consoles and PCs.
PS4 Pro is definitely going to OWN the mobile port market.
tiredheadcrab
Member
I doubt there will be a Minecraft 2.
I mean unless they are overhauling the platform to allow for bigger building features and such.
Could be. But I wouldn't bet on it.
I mean unless they are overhauling the platform to allow for bigger building features and such.
Could be. But I wouldn't bet on it.
oldergamer
Member
Hahaha, wow, this is getting entertaining more and more.
ethomaz
Banned
Before 2006-2007 with the Unified Shaders most work was done in FP16... to be fair Radeon before Unified Shaders only have FP24 support.... nVidia has FP16 and FP32 but FP32 was crap at least in NV30.
My point was more that for a game as recent as GTA5 (last gen) there's already a lot of FP16 code and knowledge what not to try to implement in FP16.
And GTA5 was ported to fp32 machines, XBO and PS4. So why would they again change the engine substantially for one system that is already performing better than its smaller brother by running the same code?
Because the special sauce in the dGPU is f16 and 8.4 teraflops of double secret power.
On the real, they wouldn't.
To keep players engaged with a 4k patch? Build an internal prototype for a RDR2 Pro version? Invest the knowledge directly into the engine?
Changing shader code doesn't have to be a "substantial" change. If the assets work with the new code there seems to be not much affected down the line. (compared to - for example - changes in the AI). This basically stuff Nvidia and AMD did (do?) for benchmark titles at driver-level on PC.
I was actually getting to the point that using FP16 is not some crazy dive into the past for most devs (as it would for a pure PC developer) and GTA was just an example.
What in gods name am i reading?!
RedAssedApe
Banned
I think he wants a Surface pro 4?
RedAssedApe
Banned
Here you go
Scorpio deep dive and 27 pages of PS4 and the fable of the FP16
Is this like the ps3 duck demo? it looks so real though.
Pasedo
Member
Exactly. People are already trying to make the damn thing portable. But yes hear me out all you skeptics. They're definitely becoming more serious as a hardware player and disrupting the laptop market. They're already doing it with the surface. I'm in the business myself and know that gaming PC's, especially laptops are experiencing a boom. Why not then leverage the Xbox brand to go after this. Traditional laptop gamers can still get all the same 3rd party games plus the added benefit of console optimisation. If this thing say costs around $800 AUD that's an extremely competitive price against an equivalent gaming laptop. Imagine this announcement at E3. 6TFs of power.... Etc... And the when they reveal the box with the top part opening up to show a built in screen and they say it's also got a lithium battery battery so it can be plugged to the TV or played on the go. Don't you think it'll be kinda cool? Plus Xbox is in a better position to do this because of the Windows ecosystem. So why not leverage on that as well. Have I convinced anyone yet???
RedAssedApe
Banned
lol...I dunno if my post had the intended effect/reaction
Pasedo
Member
I mean they need a better USP to the PS4 Pro than just more TFs. It won't cut the mustard. Look at users on here dismissing it already with all the FP16 talk. And probably the biggest clue to all of this is that Scorpio or rather scorpion, are known as 'mobile' predators. Hence Xbox Portable will be a mobile beast. Ha!
Found something worth reading regarding bandwidth in computers as it was discussed also a few days ago:
https://fgiesen.wordpress.com/2017/04/11/memory-bandwidth/
https://fgiesen.wordpress.com/2017/04/11/memory-bandwidth/
Hesemonni
Banned
I'd love to know the story behind this picture.
Your serious! Your mental gymnastics would make Mary Lou Retton proud!
Jaguar Victory
Member
I read it but it's not really relevant except for for giving us a history lesson on bandwidth.
I thought it was interesting especially how important high amount of memory bandwidth actually is.
So developers are going to make the change to make their shaders that results in requiring a large regression test run for a console which has a fraction of the user base just for a few % in performance?
For first party, I can see this happening, for everyone else, no.
anothertech
Member
Just when I thought this thread couldn't get any better.
Memorabilia
Member
I came in here expecting to read insights on the Scorpio engine and, instead, I'm reading endless debates about how so-called FP16 secret sauce makes PS4 Pro "better" and fever dreams about Scorpio being a Surface Gaming Edition. WTF are people smoking and where can I get some?
oldergamer
Member
Same shit happened to all the other Scorpio threads. Only new info can save them.
They increased the TLBs.
That's all.
So you are excited about Scorpio then.
The same thing happens in every thread where there's a lot of enthusiasm but not much in the way of concrete information. What little information is available gets dissected ad infinitum and tangents creep in when people who don't like the conversation decide to disrupt it.
The instigating article here reiterated a lot we already knew and clarified only a few points, such as differences in Vega's influence on the design from their primary competitor's approach. Not that you'd know it from reading the complaints, but an exploration of what omitting double rate FP16 implies is about as on-topic as the thread ever got.
No, some might just use FP16 in some instances and deploy that optimized code on all platforms. Be it true as true FP16 or up-converted to FP32.
Again, I'm coming from an idea how much R&D that would require and it seems to me at last, that many current rendering techniques must have been already implemented in FP16 and that's there is still senior staff that knows what absolutely requires FP32 or knows some tricks to sidestep a lack of precision.
oldergamer
Member
STOP talking about FP 16, it's not going to be used like you think it will. It lacks the precision to be as useful.
Thank you, good Sir
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
LukasTaves
Member
30% on a 1-2ms pass. That in no way means the 30% gains can be had across the board. Also, they didn't elaborated whether the gains came from the twice flop rate or the lower bandwidth usage (considering other real world scenarios the latter is way more likely), since Pro is bandwidth bound way faster than it is math bound for the intended resolution it's very likely that at least the bandwidth savings were the biggest factor in the increased performance, and that would be a saving on pretty much all platforms not just pro.
LukasTaves
Member
Liabe Brave
Member
You're mistaken about what's being referenced there. The 30% you refer to is from the Frostbite presentation, and is about gains during CBR calculations. The 35% estimate I repeated instead came from a dev on Beyond3D, and referred to their entire pixel shader pipeline. This is still not 35% overall, but it's not negligible either. And in the rest of this thread other developers and technically-minded folks gave estimations for overall benefit that were nonzero, and ranged into double digits. In some titles it may be considerably lower, or absent, but it's clearly not meaningless across the board.
In the thread I was told that there's not any bandwidth savings, but that packed math instead reduces register pressure. If that's incorrect, could you explain in detail what's actually occurring?
In any case, you must be correct about there being benefits to FP16 even without double-rate calculation. After all, devs are currently using FP16 extensively on platforms without the higher rate. But it doesn't follow that adding RPM has no additional benefit.
Am I misreading this or did you just pick up this conversation out of the blue from three months ago?
Colbert
Banned
I think the discussion is so much theoretical. We should postpone it until we actually have a game that allows us to switch between usage of FP16/FP32 mixed and FP32 only to measure the differences on frame times. Otherwise it will always a back and forth of "opinions".