• 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.

Wolfenstein II and Far Cry 5 will support FP16 Rapid Packed Math

ethomaz

ethomaz

Banned
AMD announced the specs, price and release date of the incoming GPUs based in Vega (GCN 5) yesterday and at the same time two games were confirmed to support the FP16 Rapid Packed Math:

Wolfenstein II: The New Colossus

Far Cry 5

Can FP16 Rapid Packed Math become a thing in the future and more games start to support it?

I'm really curious to see if this feature will show some increase performance over the actual nVidia GPU or even a future comparison with these games running on Vega GPU with FP16 Rapid Packed Math enabled/disabled.
 
T

thill1985

Banned
I'm skeptical of its implementation seeing as there are no real details afaik yet other than DICE's use of it to speed up the resolving of a checkerboard setup. How difficult is it for devs to thoroughly implement it outside of edge scenarios like resolving a checkerboard? In light of that, how likely is it that devs will use it more broadly going forward if only Pro and the highest end PC rigs can utilize it? I want more info, and not just from the marketing team at AMD's marketing partners.
 
R

RoboPlato

I'd be in the dick
Cool. Wonder if the PS4 Pro versions of these games will utilize it as well.

thill1985 said:
I'm skeptical of its implementation seeing as there are no real details afaik yet other than DICE's use of it to speed up the resolving of a checkerboard setup. How difficult is it for devs to thoroughly implement it outside of edge scenarios like resolving a checkerboard? In light of that, how likely is it that devs will use it more broadly going forward if only Pro and the highest end PC rigs can utilize it? I want more info, and not just from the marketing team at AMD's marketing partners.
Click to expand...

It's been utilized in more small games than large games so far on the Pro. Mantis Burn Racing was the first to talk about it. Said it was a big part of how that game got to native 4k on the Pro.
 
ethomaz

ethomaz

Banned
I think if more big games start to use it then even nVidia will migrate this feature of HPC cards to desktop.

I really expect Volta desktop to feature 2xFP16 like Volta/Pascal HPC.
 
I

Inuhanyou

Believes Dragon Quest is a franchise managed by Sony
Is FP16 really a tangible trick? Like is it similar to checkerboarding or lower resolution graphical effects when it comes to utilizing certain rendering tricks?
 
ethomaz

ethomaz

Banned
Inuhanyou said:
Is FP16 really a tangible trick? Like is it similar to checkerboarding or lower resolution graphical effects when it comes to utilizing certain rendering tricks?
Click to expand...
I don't think it is just a trick... there are parts of the rendering process that didn't require 32bits maths.

I think it is more like optimization of hardware resources... if you don't need 32bits then you don't have loss of quality and the result will be identical... while checkerboard there are loss of the quality and that is what I can call a trick.
 
R

RoboPlato

I'd be in the dick
Inuhanyou said:
Is FP16 really a tangible trick? Like is it similar to checkerboarding or lower resolution graphical effects when it comes to utilizing certain rendering tricks?
Click to expand...

It's not quite the same as those. Instead of being an alternate method they just do fewer calculations on processes that don't need to have as high precision for the final result to look good. It's an efficiency boost, similar to how GPU compute fills in the gaps in GPU utilization.
 
ethomaz

ethomaz

Banned
RoboPlato said:
Cool. Wonder if the PS4 Pro versions of these games will utilize it as well.
Click to expand...
If it is already coded in the game engine for Vega GPUs on PC then I don't see any reason for it not be enabled on Pro. So I expect Wolfenstein II and Far Cry 5 uses this feature on Pro.
 
V

VariantX

Member
St3v3 said:
And what does that mean for gamers?
Click to expand...

My best guess is that calculations that don't need high precision numbers can be done in FP16 and done in roughly half the time as it would be done in FP32. So you could be doing more of those low precision things in the same space of time. I remember there was a thread here with some graphical examples on this stuff not too long ago specifically with textures.
 
D

dogen

Member
BruceCLea said:
What does it do
Click to expand...

It means you can execute 2 floating point operations at the same time on 2 different lower precision numbers, in the same amount of time as 1 instruction performed on a normal precision number. The catch is that lower precision isn't useful everywhere and can result in problems or loss of quality if not used carefully. Plus, the potential performance improvement, even specifically where it's used, is only 2x in theory because there will essentially always be other limiting factors.
 
D

DieH@rd

Banned
Good stuff! idTech6 continues to rock.

edit-
wmKVMV3.jpg
 
ethomaz

ethomaz

Banned
VariantX said:
My best guess is that calculations that don't need high precision numbers can be done in FP16 and done in roughly half the time as it would be done in FP32. So you could be doing more of those low precision things in the same space of time. I remember there was a thread here with some graphical examples on this stuff not too long ago specifically with textures.
Click to expand...
This is the actual PR example.

C5xhj9mVAAAB56b.jpg
Click to expand...
 
ethomaz

ethomaz

Banned
StrongBlackVine said:
Is this similar to the tech Cerny talk about with PS4 Pro? If so I wonder if it will be utilized.
Click to expand...
Some indies already use it on Pro and Wolfenstein II will be the first big title.

Tallshortman said:
Huh if it's available in the PS4 Pro maybe this will become widespread. Looks very promising.
Click to expand...
I think to become widespread it needs to be supported by nVidia with Volta gaming GPUs.

nVidia has near 80% of the gaming GPU market after all.
 
I

Inuhanyou

Believes Dragon Quest is a franchise managed by Sony
Tallshortman said:
Huh if it's available in the PS4 Pro maybe this will become widespread. Looks very promising.
Click to expand...

i dunno, will devs bother? I mean for the games devs are making, pro is the only mainstream machine that utilizes FP16 right?
 
I

Inuhanyou

Believes Dragon Quest is a franchise managed by Sony
Proelite said:
Switch uses it.
Click to expand...

I did say "for the games devs are making", although i probably should have clarified, my bad dude.

Switch isn't getting far cry 5 or Wolfenstein 2 or any games that will utilize this technology to a large extent to benefit Pro by having an extension of a userbase
 
D

dogen

Member
Inuhanyou said:
i dunno, will devs bother? I mean for the games devs are making, pro is the only mainstream machine that utilizes FP16 right?
Click to expand...

Anything polaris or newer (which includes scorpio) can still get higher GPU utilization from the lower register pressure. So I guess you could say scorpio has packed math, just without the rapid part.
 
P

Painguy

Member
half precision stuff has been a thing in GPU's since dx9, but for one reason or another it didn't take off the way it has been now. mostly been hidden from public eye.
 
N

Neith

Banned
All the haters are now looking at this in a different light now maybe. We finally have some big devs on board.
 
I

icecold1983

Member
should be interesting and will finally silence all the people who think fp16 is magic(nintendo fans). <15% seems realistic. should be easily enough to put vega ahead of a 1080 in such titles tho.
 
K

kirankara78

Member
Neith said:
All the haters are now looking at this in a different light now maybe. We finally have some big devs on board.
Click to expand...

I'll wait till i see results before getting too excited. Will also be interesting to see how widely adopted this actually is as it sounds like a lot of extra work
 
ethomaz

ethomaz

Banned
kirankara78 said:
I'll wait till i see results before getting too excited. Will also be interesting to see how widely adopted this actually is as it sounds like a lot of extra work
Click to expand...
What extra work?

If you you coded a game "thinking" only in FP32 yes I guess you will have extra work to convert/port some parts to FP16.

But if you already coded (new games) with FP16 in mind you will have little to no extra work I guess.

Remember mobile devs already utilize FP16 massively in games.
 
PLASTICA-MAN

PLASTICA-MAN

Member
RoboPlato said:
Cool. Wonder if the PS4 Pro versions of these games will utilize it as well.



It's been utilized in more small games than large games so far on the Pro. Mantis Burn Racing was the first to talk about it. Said it was a big part of how that game got to native 4k on the Pro.
Click to expand...

I am wondering the same too. If they really do use it on PS4 PRO they could run both games in Native 4K maxed out on PS4 PRO.
 
S

Sad Affleck

Member
Neith said:
All the haters are now looking at this in a different light now maybe. We finally have some big devs on board.
Click to expand...

There are no haters of the technology itself. What there are haters of, and I consider myself one of them, is the attempts by specific posters (onQ123, specifically) to present this technology (and any technology really) in a way that suits his console war agenda. The technology itself has some worthwhile benefits and I hope it is used extensively in the future.
 
H

Henrar

Member
stryke said:
Is RPM a special technique FP16 can do or is it a AMD catchphrase to use interchangeably with FP16?
Click to expand...

RPM means that the GPU can do FP16 math at twice the speed of FP32 computations in one cycle. GPUs without RPM can do FP16 math, but they don't have any performance advantage.
 
ethomaz

ethomaz

Banned
stryke said:
Is RPM a special technique FP16 can do or is it a AMD catchphrase to use interchangeably with FP16?
Click to expand...
It is the ability to run 2x FP16 instructions in one FP32 unit... that means you delivery twice the number of instructions per clock with FP16 compared to FP32.

Basically all GPU can run FP16 maths but actually only Vega GPUs, PS4 Pro's GPU, mobile GPUs (almost all... that includes any major smartphone, tablet and Switch) and HPC's GPUs run it twice faster.

RPM itself is only a fancy word for marketing this feature in Vega... you can call it Double-rate FP16 too.
 
P

pottuvoi

Banned
ethomaz said:
What extra work?

If you you coded a game "thinking" only in FP32 yes I guess you will have extra work to convert/port some parts to FP16.

But if you already coded (new games) with FP16 in mind you will have little to no extra work I guess.

Remember mobile devs already utilize FP16 massively in games.
Click to expand...
One pretty much has to have up to date fp32 version of shader as a reference to use when seeking precision errors.
 
ethomaz

ethomaz

Banned
pottuvoi said:
One pretty much has to have up to date fp32 version of shader as a reference to use when seeking precision errors.
Click to expand...
Precision errors?

You know before code if a variable will require FP32 or not.

In code terms is just change "single" to "half" in the declaration of the variable type and use some convertion functions in some required parts.

It is not something any programmers is not used already.

PS. Most engines will already switch based on what target hardware you choose without any trouble to developers.
 
jobrro

jobrro

Member
Will be interesting to see how the Pro (with FP16 at double rate) stacks up with the raw power advantage of the Xbox One X (FP32 only).

Have no idea what to expect in the real world gaming performance from FP16 at double rate, not sure how much it can be applied to in game engine.

I am not really expecting something huge but looking forward to the Digital Foundry and NX Gamer analysis.
 
R

riflen

Member
Everything old is new again. The first PC 3D graphics accelerators came about precisely because some enterprising people who worked in 3D graphics simulation in military and professional markets, realised that if you didn't need such precise maths, you could do some pretty fast 3D rendering for comparatively cheap cost on PC.

In the past we've had FP16 hardware support (NV30 I think, in 2003), more performance if you don't need the precision. I imagine it wasn't used extensively enough and was not included in subsequent designs. Now, here we are again. I imagine the rendering world has changed enough that we now maybe have more use for this.

Also, crucially, FP16 is useful in network training, which is a major market for the GPU designers today. This is likely the main reason for this feature appearing again in consumer GPUs.
 
You must log in or register to reply here.

Similar threads

Nvidia has a new graphics Control panel
Analysis 2 3
123
8K
YeulEmeralda replied
Tom Warren - Sony’s PS5 Pro is real and Sony is asking developers to get games ready over the summer, with a push for ray-tracing support
News Hardware Hype 2 3 4
186
11K
DanielG165 replied
Coming Soon to Game Pass: Far Cry 6, Remnant II, SteamWorld Build, and More
37
2K
Jesb replied
AMD dodges question on anti-consumer behavior regarding their bundled games. (Update: article is selective)
Drama Clickbait Astroturfing 2 3 4
199
9K
Reallink replied
Diablo IV, Redfall, Forza Horizon 5 and more getting DLSS 3.0
26
3K
SlimySnake replied
Top Bottom