Wheres the receipts. Games do not use 100% FP16 RPM.
Its all misleadinf because its theory. Noone has done it yet on any game.
-
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
- Thread starter ethomaz
- Start date
Wheres the receipts. Games do not use 100% FP16 RPM.
Its all misleadinf because its theory. Noone has done it yet on any game.
ethomaz
Banned
That is to be see with the future games using it... that indeed the point of discussion of this thread and not the denial of the raw flops of FP16 in these new cards.
Receipts for what? It is a fact and you can look at official specs for any card with double rate half-precision.
Now you want performance comparison? Wait the first games using FP16 to draw a picture but remember FP32 is not used 100% by games too that makes you claim dumb.
My main point is that it seems they use Fp16 only on thinga that dont need as much precision or detail not for the entire games calculations etc.
So far the only claim is 14% increase in FPS in Wolfenstein. Which is awesome
martino
Member
why use a 50% example and not less?
This is the problem with you and fp16.. the implication.
TheThreadsThatBindUs
Member
The ignorance of this post is that no game even uses 100% fp32 calculations. There's alot more to videogames than just floating point calculations.
TheThreadsThatBindUs
Member
No... you're arguing against a problem with what AMD publishes and then blaming it on onQ123.
I'd quit now if I was you.
TheThreadsThatBindUs
Member
Sebbbi A from Beyond3D (AKA an authority on the subject):
Sebbbi A from Beyond3D said:
It's nothing like ES RAM, however. ESRAM usage couldn't be avoided on Xbox 360. FP16 usage can. FP16 usage with hardware supporting RPM is merely an optimization route for improving performance in some areas of the graphics rendering pipeline. Some games will benefit from it more than others, as will some developers (i.e. not all graphics programmers are made equal).
You mean to tell me that you been doing these drive by posts for months & this is your understanding of things?
martino
Member
what i blame on onQ123 is what he hopes and imply from fp16. Not the way AMD publishes theorical number.
if you don't why 4.2 to 6.3 is a convinient example...it's because you don't want ti see it.
Sad Affleck
Member
Honest question: Didn't this clue you in that you are doing something wrong? Was it really that 'people don't want to understand' or that you presented the information in such a manner as to cause confusion? I'll say it again: Quoting specs and theoretical numbers may be ok in an environment where everyone is a developer and understands the information immediately without needing an explanation. GAF is not such an environment. You should have known that quoting specs out of context and without any explanation as to what these specs mean could (and did) derail entire threads as users latch on to the familiar teraflop figure and fight each other over what it actually means. That's on you because you did a poor job explaining what these specs mean.
So you think that would equal to me saying PS4 Pro is more powerful than Xbox One X?
going from fp32 to fp16 is a compromise to begin with & plus Xbox One has more & faster memory. you're fighting with your own insecurities my post had nothing to do with Xbox One X.
I used 50% because it's simple & right there in the middle to give a good example to go on.
TheThreadsThatBindUs
Member
This is your problem, right here. You're interpreting mal-intent based on nothing.
Go back and read everything that onQ123 posted and he consistently and clearly qualified his statements by claiming each time that he's talking about a "theoretical max." metric.
Clearly you aren't grasping this because the very 4.2 TFLOPs figure you quote in this very post for PS4 Pro is also a "theoretical max." metric... just in 32 bit precision as opposed to 16 bit precision. It's just as unrealistic as the 2x fp16 metric, since NO GPU micro-architecture is 100% efficient.
What you're essentially arguing here is that onQ123 should instead try to pull a number out of his arse to estimate maximum fp16 performance, instead of using the published theoretical figure that the actual hardware designer, AMD, themselves released.... do you not see now how ridiculous you're sounding?
AMD themselves published that the Vega micro-architecture with RPM delivers 2x fp16 ops per clock than 32bit ops per clock (as per the image in my previous post).
If you have an argument about these numbers being disingenuous or deceptive then blame AMD.
No because just like this thread it's people throwing BS all over the place & making a lot of noise while pointing at the person who is giving a factual statement as if he did something wrong.
Mods don't always fact check they just come in & try to clean up the thread
TheThreadsThatBindUs
Member
Base your criticisms on his posts in this thread and you'll see that nowhere has he quoted the 2x fp16 rate without clarifying that it's a theoretical maximum.
Why are you trying to crucify this guy for making correct statements based on something he said in the past in another thread that he's already learned his lesson for?
I don't get why you guys are so intent on targetting this guy...
You're seriously derailing this thread.
I didnt mean every calculation I meant where fp32 would be used , insert packed fp16 instead.
I also been mentioning that your quotes aee all theory, in which you agree, im just argueing that the theory means little until FP16 packed is heavily used and shown to have a huge boost in a game vs using fp32.
Which hasnt happened yet. So far the best thing weve got is here, a 14% fps boost. Which is a great thing.
I just think people are going to take your 2x boost and run wild with it and be misinformed on it all.
Ill stop posting about your guys quotes because im probably derailing the thread as much as I feel you are as well nor do I want to be banned for this.
Each & every time that I have posted the specs I have always had context to show that it's fp16 or fp32 I'm not sure what's misinforming about that.
kirankara78
Member
From what I've managed to find on the topic, fp16 usage will be only relevant in certain situations and isn't a magical increase all round.
http://www.gamersnexus.net/...
"Rapid-packed math was shown effectively doubling the amount of hair strands that could be rendered per second. This is done with precision switching, where Vega is able to push FP16 rather than FP32 for reduced precision and increased speed. Precision switching isnt deployable in every aspect of gaming youd have potential errors in data presentation but its a fine use case for hair rendering. Hair doesnt need to be precise, particularly with strands numbering in the tens of thousands. AMD mentioned that RPM has already been used in consoles and high-end compute GPUs, but that they were attempting to bring this feature into the PC gaming world. We continue to be dubious about gaming applicationsAMD can develop the technology, but its up to software developers to actually make use of it."
http://www.gamersnexus.net/...
"Rapid-packed math was shown effectively doubling the amount of hair strands that could be rendered per second. This is done with precision switching, where Vega is able to push FP16 rather than FP32 for reduced precision and increased speed. Precision switching isnt deployable in every aspect of gaming youd have potential errors in data presentation but its a fine use case for hair rendering. Hair doesnt need to be precise, particularly with strands numbering in the tens of thousands. AMD mentioned that RPM has already been used in consoles and high-end compute GPUs, but that they were attempting to bring this feature into the PC gaming world. We continue to be dubious about gaming applicationsAMD can develop the technology, but its up to software developers to actually make use of it."
Sad Affleck
Member
If you feel this way I apologize and I'm out.
From the F1 2017 PS4 Pro enhancements blog post
Side note: I think it's time for a RPM/fp16 PS4 Pro thread that can stay open without the trolling. It seems that the trolls are doing a good job because there is no thread left open to post about PS4 Pro using fp16.
Side note: I think it's time for a RPM/fp16 PS4 Pro thread that can stay open without the trolling. It seems that the trolls are doing a good job because there is no thread left open to post about PS4 Pro using fp16.
PLASTICA-MAN
Member
I was going to post this here since I found out about this right now. It seems F1 2017 is the first game to release using F16 RPM which allowed them besides improving the resolution to add better reflections and improved shadows.
We also need a thread for games that will be using it on PS4 Pro.
Good luck keeping that tread open lol people are really acting like kids when it come to something they don't want to understand. This isn't some fake made up stuff it's real but yet people act like someone is claiming secret sauce & fairy dust is inside of the PS4 Pro.
But on the topic of F1 2017 I wonder how the end result using CBR & higher settings look compared to what it would have looked if they went for native 4K with the same settings as base PS4.
They showed the ROTR hair demo I think that was a good example besides AMD don't make games so waiting for the games that take advantage of the hardware is the best bet.
Sidenote: PS4 Pro & Xbox One X GPU is separated by 4 more CU's + higher clock rate on Xbox One X & PS4 Pro having features like RPM from a GPU generation ahead of Xbox One X GPU I think the difference between games will be smaller than what is expected from the 4.2TF vs 6TF fp32 numbers. My guess is that the extra/faster Ram is going to be the thing that set Xbox One X apart with better textures while PS4 Pro might have better effects or even better fps while using CBR or a lower resolution.
I remember the dev or someone said that F1 2017 was going to be native 4K on Xbox One X but in the end they went with CBR also I wonder if that was because of the fact that PS4 Pro was able to use CBR & have higher settings than the OG PS4 which probably made them decide that it would be better to use CBR & higher settings vs native 4K with the same settings as the base hardware?
LordOfChaos
Member
AMDs figures for FP16 performance increase. Not a magic performance doubler as some liked to bill it, but definitely a nice addition for no visual degradation (and more physics).
AMDs figures for FP16 performance increase. Not a magic performance doubler as some liked to bill it, but definitely a nice addition for no visual degradation (and more physics).
Who said it was a magic performance doubler?
if everything could be done using fp16 & it was no other bottlenecks maybe but that's not the case & the benefits will depend on what can be moved to fp16 without a noticeable drop in quality.
dr_rus
Member
This image was posted and discussed previously.
Also - the jury's still out on the "no visual degradation" part.
AMDs figures for FP16 performance increase. Not a magic performance doubler as some liked to bill it, but definitely a nice addition for no visual degradation (and more physics).
And that's for only those specific rendering processes not the whole gpu rendering frame. So that 20% faster may only be for a process that's 5% of the entire rendering frame so it's a 1% increase in performance
And that's for only those specific rendering processes not the whole gpu rendering frame. So that 20% faster may only be for a process that's 5% of the entire rendering frame so it's a 1% increase in performance
Is that 1% of China?
anothertech
Member
Hope this helps with framerates. Get a little more umph outta these jaguars. lol
AMDs figures for FP16 performance increase. Not a magic performance doubler as some liked to bill it, but definitely a nice addition for no visual degradation (and more physics).
PS4 Pro 8.4TF confirmed?
/s
And that's for only those specific rendering processes not the whole gpu rendering frame. So that 20% faster may only be for a process that's 5% of the entire rendering frame so it's a 1% increase in performance
Or you know, the idTech6 demo quoted earlier which had 14% fps improvement with fp16 over fp32.
Onq123 redeemed.
Sounds like adding in RPM to the GPU was some great future proofing by Sony.
Negotiator
Member
No need to downplay RPM just because X doesn't have it. X will still be the most powerful console until PS5 comes out.And that's for only those specific rendering processes not the whole gpu rendering frame. So that 20% faster may only be for a process that's 5% of the entire rendering frame so it's a 1% increase in performance
TheThreadsThatBindUs
Member
Not really.
Obviously, developers are only going to use fp16 and mixed precision where it makes sense in the rendering pipeline (i.e. where the additional precision of 32bit floats isn't necessary).
No dev is going to sacrifice visual fidelity of graphical effects for the higher performance, otherwise all console games would be 60fps.
So I think its more accurate to claim that the jury's still out on the degree of performance gain mixed precision opens up in future games, or on the extent to which the graphics rendering pipeline can leverage mixed precision for the increased performance.
I think it's more than RPM that's part of PS4 Pro but not Xbox One X like IWD & Primitive Shaders
”The work distributor in PS4 Pro is very advanced," he claimed. ”Not only does it have the fairly dramatic tesselation improvements from Polaris [AMD's GPU architecture], it also has some post-Polaris functionality that accelerates rendering of scenes with very small objects. ”
Edit:
Next-generation geometry engine
To meet the needs of both professional graphics and gaming
applications, the geometry engines in ”Vega" have been
tuned for higher polygon throughput by adding new fast
paths through the hardware and by avoiding unnecessary
processing. This next-generation geometry (NGG) path is
much more flexible and programmable than before.
To highlight one of the innovations in the new geometry
engine, primitive shaders are a key element in its ability to
achieve much higher polygon throughput per transistor.
Previous hardware mapped quite closely to the standard
Direct3D rendering pipeline, with several stages including
input assembly, vertex shading, hull shading, tessellation,
domain shading, and geometry shading. Given the wide
variety of rendering technologies now being implemented
by developers, however, including all of these stages isn't
always the most e cient way of doing things. Each stage
has various restrictions on inputs and outputs that may
have been necessary for earlier GPU designs, but such
restrictions aren't always needed on today's more flexible
hardware.
”Vega's" new primitive shader support allows some parts of
the geometry processing pipeline to be combined and
replaced with a new, highly e cient shader type. These
flexible, general-purpose shaders can be launched very
quickly, enabling more than four times the peak primitive
cull rate per clock cycle.
In a typical scene, around half of the geometry will be
discarded through various techniques such as frustum
culling, back-face culling, and small-primitive culling. The
faster these primitives are discarded, the faster the GPU
can start rendering the visible geometry. Furthermore,
traditional geometry pipelines discard primitives after
vertex processing is completed, which can waste computing
resources and create bottlenecks when storing a large batch
of unnecessary attributes. Primitive shaders enable early
culling to save those resources.
The ”Vega" 10 GPU includes four geometry engines which
would normally be limited to a maximum throughput of
four primitives per clock, but this limit increases to more
than 17 primitives per clock when primitive shaders are
employed.⁷
Primitive shaders can operate on a variety of di erent
geometric primitives, including individual vertices,
polygons, and patch surfaces. When tessellation is enabled,
a surface shader is generated to process patches and control
points before the surface is tessellated, and the resulting
polygons are sent to the primitive shader. In this case, the
surface shader combines the vertex shading and hull
shading stages of the Direct3D graphics pipeline, while the
primitive shader replaces the domain shading and
geometry shading stages.
Primitive shaders have many potential uses beyond
high-performance geometry culling. Shadow-map
rendering is another ubiquitous process in modern engines
that could benefit greatly from the reduced processing
overhead of primitive shaders. We can envision even more
uses for this technology in the future, including deferred
vertex attribute computation, multi-view/multi-resolution
rendering, depth pre-passes, particle systems, and
full-scene graph processing and traversal on the GPU.
Primitive shaders will coexist with the standard hardware
geometry pipeline rather than replacing it. In keeping with
”Vega's" new cache hierarchy, the geometry engine can now
use the on-chip L2 cache to store vertex parameter data.
This arrangement complements the dedicated parameter
cache, which has doubled in size relative to the
prior-generation ”Polaris" architecture. This caching setup
makes the system highly tunable and allows the graphics
driver to choose the optimal path for any use case.
Combined with high-speed HBM2 memory, these
improvements help to reduce the potential for memory
bandwidth to act as a bottleneck for geometry throughput.
Another innovation of ”Vega's" NGG is improved load
balancing across multiple geometry engines. An intelligent
workload distributor (IWD) continually adjusts pipeline
settings based on the characteristics of the draw calls it
receives in order to maximize utilization.
One factor that can cause geometry engines to idle is
context switching. Context switches occur whenever the
engine changes from one render state to another, such as
when changing from a draw call for one object to that of a
di erent object with di erent material properties. The
amount of data associated with render states can be quite
large, and GPU processing can stall if it runs out of
available context storage. The IWD seeks to avoid this
performance overhead by avoiding context switches
whenever possible.
Some draw calls also include many small instances (i.e.,
they render many similar versions of a simple object). If an
instance does not include enough primitives to fill a
wavefront of 64 threads, then it cannot take full advantage
of the GPU's parallel processing capability, and some
proportion of the GPU's capacity goes unused. The IWD
can mitigate this e ect by packing multiple small instances
into a single wavefront, providing a substantial boost to
utilization.
”Vega" NCU with Rapid Packed Math
GPUs today often use more mathematical precision than
necessary for the calculations they perform. Years ago, GPU
hardware was optimized solely for processing the 32-bit
floating point operations that had become the standard for
3D graphics. However, as rendering engines have become
more sophisticated—and as the range of applications for
GPUs has extended beyond graphics processing—the value
of data types beyond FP32 has grown.
The programmable compute units at the heart of ”Vega"
GPUs have been designed to address this changing
landscape with the addition of a feature called Rapid
Packed Math. Support for 16-bit packed math doubles peak
floating-point and integer rates relative to 32-bit
operations. It also halves the register space as well as the
data movement required to process a given number of
operations. The new instruction set includes a rich mix of
16-bit floating point and integer instructions, including
FMA, MUL, ADD, MIN/MAX/MED, bit shifts, packing
operations, and many more.
For applications that can leverage this capability, Rapid
Packed Math can provide a substantial improvement in
compute throughput and energy e ciency. In the case of
specialized applications like machine learning and training,
video processing, and computer vision, 16-bit data types are
a natural fit, but there are benefits to be had for more
traditional rendering operations, as well. Modern games,
for example, use a wide range of data types in addition to
the standard FP32. Normal/direction vectors, lighting
values, HDR color values, and blend factors are some
examples of where 16-bit operations can be used.
With mixed-precision support, ”Vega" can accelerate the
operations that don't benefit from higher precision while
maintaining full precision for the ones that do. Thus, the
resulting performance increases need not come at the
expense of image quality.
KnightimeX_Legacy
Banned
Is Wolfenstein 2 going to use the same engine DOOM was using?
I think it's IDtech 6 but not sure.
DOOM looked GREAT!
Wolfy 2 using the same would be very nice.
I think it's IDtech 6 but not sure.
DOOM looked GREAT!
Wolfy 2 using the same would be very nice.
PS4 Pro GPU seem to be basically Vega without HBM all the way down to the fact that it can swap apps in & out of the GDDR5 to the DDR3 memory.
HBCC technology can be leveraged for consumer
applications, as well. The key limitation in that space is that
most systems wont have the benefit of large amounts of
system memory (i.e., greater than 32 GB) or solid-state
storage on the graphics card. In this case, HBCC ef ectively
extends the local video memory to include a portion of
system memory. Applications will see this storage capacity
as one large memory space. If they try to access data not
currently stored in the local high-bandwidth memory, the
HBCC can cache the pages on demand, while less recently
used pages are swapped back into system memory. This
unified memory pool is known as the HBCC Memory
Segment (HMS).
HBCC technology can be leveraged for consumer
applications, as well. The key limitation in that space is that
most systems wont have the benefit of large amounts of
system memory (i.e., greater than 32 GB) or solid-state
storage on the graphics card. In this case, HBCC ef ectively
extends the local video memory to include a portion of
system memory. Applications will see this storage capacity
as one large memory space. If they try to access data not
currently stored in the local high-bandwidth memory, the
HBCC can cache the pages on demand, while less recently
used pages are swapped back into system memory. This
unified memory pool is known as the HBCC Memory
Segment (HMS).
ethomaz
Banned
Yes it is the same engine but enhanced to take advantage of FP16 RPM.
Still a huge gap between the pro and xbx. Memory is another thing I wish they would increase for higher res texture beside the lack of uhd bd drive.
dr_rus
Member
How will you tell that you're seeing lower image quality due to 16 bit math being used if there's no way to compare that to 32 bits - on PS4Pro for example? Also where the line lies in quality degradation - is some minor shimmering in places okay for a 10% performance boost?
The jury is always out on such features. The fact is that 16 bit math usage can (and did, actually, back in GeForce FX days) impact quality. Developers aren't gods and they do make mistakes more often than not. The lure of using FP/INT16 a bit more often to get ahead of competition in PC space should not be underestimated either.
IWD seems to be little more than a marketing buzzword and primitive shaders is a software (driver) construct able to run on any modern GPU. In fact, it's a solid bet that NV has them (well, not them but a similar s/w / driver optimization) running for some time now as they are somewhat related to current TBIR implementations.
Vega doesn't move programs anywhere, it uses system RAM as a slow memory pool for whatever resources a running 3D app requires. What Cerny speak of is basically an implementation detail of their NUMA OS scheduler. It's a CPU/OS thing.
My point is that PS4 Pro share these features with Vega , I'm not sure what your opinion of what IWD & Primitive Shaders is has to do with that. How did you come to the conclusion that a hardware feature that hasn't been talked about much is little more than a buzzword?
Did you quote the wrong post or something?
But what do you consider a huge gap? Some people don't even think the PS4 vs PS4 Pro is a huge gap.
kungfuian
Member
Cerny, it's possible to perform two 16-bit operations at the same time, instead of one 32-bit operation. In other words, with full floats, PS4 Pro has 4.2 teraflops of computational power. With half floats, it now has double that -- which is to say, 8.4 teraflops of computational power. As I'm sure you understand, this has the potential to radically increase the performance of games.
Cerny doesn't have a history of hyperbole or spouting marketing BS and the key word he uses here is radically. Not 'a little' or 'marginally' or any other adjective he could have used. He specifically said radically and if he says radically he probably means radically.
Cerny doesn't have a history of hyperbole or spouting marketing BS and the key word he uses here is radically. Not 'a little' or 'marginally' or any other adjective he could have used. He specifically said radically and if he says radically he probably means radically.
I mean memory in general between pro and xbx, even with the additional .5 gigs they added, 4k frame buffers ui etc already eat up a huge chunk of memory. If they designed they pro for 4k even if its checkerboarding they need much more for higher quality texture. Texture is one of the most abvious things that stands out.
dr_rus
Member
As I've said, HBCC is most certainly not in PS4Pro - and it would be pretty much useless there since this is a fixed h/w platform anyway.
IWD being a meaningless buzzword comes from the fact that there's no noticeable advantage of Vega against Polaris in a typical gaming geometry load. It is likely just a part of the whole "primitive shader" s/w optimization - which seems to be disabled in Vega for now which explains the results against Polaris.
What's "radically"? Because it will be less than 2X in any case.