PDA

View Full Version : Far Cry, Nv3x, verses FP16, verses FP32 (56k Warning)


Pages : [1] 2 3 4

ChrisRay
06-28-04, 04:18 PM
I have heard alot of people have been discrediting far cry issues for the Nv3x path to be a FP16 precision issue. Well From my experiences it's not. Please Follow me here :) Several Comparison need to be made. All shots were done @ 1024x768 with 4x AA/16x Performance AF.

Pipe Scene: This has been a popular comparision because it seems to show one pipe being "More Shiny" Than the other, So I thought investigating whether this is actually a FP16 issue or not would be relevent.


NV3x Mode


http://www.nvnews.net/articles/chrisray_iq/NV3xPipes.png



FP16 R300 Mode


http://www.nvnews.net/articles/chrisray_iq/FP16pipes.png




FP32 R300 Mode

http://www.nvnews.net/articles/chrisray_iq/FP32pipes.png





Pipe Scene Conclusion: Moving from NV3x Mode to FP16 R300 mode has the most dramatic increase in Quality. Small differences can be seen moving from FP16 r300 mode to FP32. Unfortunately it seems the SM 3.0 path has less "Shiny" pipes than the FP32 and FP16 path. There is a large performance enhancement but there IQ is that of the Nv3x path








Floor Scene: The dreaded banded floor scene. Many have believed that the issues with the floor scene revolve around FP16 precision. This does not appear to be the case.



NV3x Mode


http://www.nvnews.net/articles/chrisray_iq/NV3xfloor.png






R300 FP16 Mode


http://www.nvnews.net/articles/chrisray_iq/Fp16floor.png






R300 FP32 Mode



http://www.nvnews.net/articles/chrisray_iq/FP32floor.png






NV3x Mode


http://www.nvnews.net/articles/chrisray_iq/Nv3xcooler.png



R300 FP16 Mode


http://www.nvnews.net/articles/chrisray_iq/fp16cooler.png




R300 FP32 Mode



http://www.nvnews.net/articles/chrisray_iq/Fp32cooler.png





Floor Conclusion: As you can see the differences between FP 16 and FP32 seem to be rather marginal. However the differences between NV3x mode to FP16 are instrumental. The NV40 SM 3.0 Path also looks identical in this case.



P.S. Special thanks to Jakup and Fivefeet, for inspiring me to investigate this issue further.

jAkUp
06-28-04, 04:32 PM
P.S. Special thanks to Jakup and Fivefeet, for inspiring me to investigate this issue further.

Nice shots and comparisions!!! And an extra w00t for mentioning me!! :clap:

Jedi Rainman
06-28-04, 04:39 PM
Good comparisons

ragejg
06-28-04, 04:50 PM
Thanks for the good short sweet summarization, Chris, :thumbsup:

fivefeet8
06-28-04, 05:05 PM
What type of performance difference are you seeing Chris? From the using the Nv3x or R300 paths. It seems a lot of people are stating that running the R300 path results in a 20-30 percent performance loss. I just haven't seen any concrete proof of it. And from what I'm seeing on my end, there doesn't seem to be that much of a performance hit at all going from the Nv3x path to the R300 path.

ChrisRay
06-28-04, 05:14 PM
What type of performance difference are you seeing Chris? From the using the Nv3x or R300 paths. It seems a lot of people are stating that running the R300 path results in a 20-30 percent performance loss. I just haven't seen any concrete proof of it. And from what I'm seeing on my end, there doesn't seem to be that much of a performance hit at all going from the Nv3x path to the R300 path.


My performance differences have been marginal at best, As a matter of fact FP16 Mode seems faster than the Nv3x mode to me. Thought I have only compared these "two" spots and levels.


Fp32 mode seems to be roughly equivalent to Nv3x mode for me in "these" levels.

fivefeet8
06-28-04, 05:30 PM
hmm... Ok Chris, looking at your screenshots it seems you didn't force FULL High Quality Lighting. The glowing glass doesn't look like it should. It should look like this:

http://www.nvnews.net/vbulletin/showpost.php?p=352368&postcount=12

To get the FULL effects, you need to decompress(winzip) both the shaders1.pak and shaders.pak files into a 'shaders' folder. Then compress that "shaders" folder into a new shaders.zip(rename to shaders.pak). Place the new shaders.pak file into the FCdata folder. Delete the old shaders1.pak, or move it somewhere out of the Far Cry directory.

The original shaders1.pak and shaders.pak are located in the FCdata folder in the Far Cry directory.

photophreak314
06-28-04, 05:31 PM
So crytek should make the FP16 R300 mode the default for NV3X in patch 1.2. That would solve all the IQ issues, and have virtually no performance loss. If they were smart.

jAkUp
06-28-04, 05:33 PM
Yes, I also noticed that his glass didnt look quite like mine once I forced the R3XX mode... I figured maybe I just did it wrong..heh

ChrisRay
06-28-04, 05:35 PM
hmm... Ok Chris, looking at your screenshots it seems you didn't force FULL High Quality Lighting. The glowing glass doesn't look like it should. It should look like this:

http://www.nvnews.net/vbulletin/showpost.php?p=352368&postcount=12

To get the FULL effects, you need to decompress(winzip) both the shaders1.pak and shaders.pak files into a 'shaders' folder. Then compress that "shaders" folder into a new shaders.zip(rename to shaders.pak). Place the new shaders.pak file into the FCdata folder. Delete the old shaders1.pak, or move it somewhere out of the Far Cry directory.

The original shaders1.pak and shaders.pak are located in the FCdata folder in the Far Cry directory.



Thats interesting, I merely used the FP16/FP32 Patch from Tommti Systems to illustrate the differences between the FP precision modes.

Is Forcing Higher Lighting going to work on the Nv3x pathway? because if not, is that a good way to illustrate the difference between FP16/FP32?

I'm curious your thoughts on this because. The main purpose of this test was to show that FP16 was not the cause of the artifacts seen in the Nv3x mode.

Dont mis interpret the results of these tests as a performance comparison of any kind. They were meant to show that the Nv3x Mode, is responcible for Lighting issues in Far Cry. Not FP16.

fivefeet8
06-28-04, 05:42 PM
Thats interesting, I merely used the FP16/FP32 Patch from Tommti Systems to illustrate the differences between the FP precision modes.

Is Forcing Higher Lighting going to work on the Nv3x pathway? because if not, is that a good way to illustrate the difference between FP16/FP32?


I'm not sure either, but that is the only way to get the glass and maybe some other effects to look like it does on a R300. But who really knows what it's suppose to look like. Maybe the Nv3x has it right.

Anyways, I do agree with you from my own experience with Far Cry. FP16 wasn't the culprit of those banding issues. I do wonder what it is though that causes them.

Hmm. Maybe if we tried the FP16 R300 mode on a real Nv3x.

ChrisRay
06-28-04, 05:45 PM
I'm not sure either, but that is the only way to get the glass and maybe some other effects to look like it does on a R300. But who really knows what it's suppose to look like. Maybe the Nv3x has it right.

Anyways, I do agree with you from my own experience with Far Cry. FP16 wasn't the culprit of those banding issues. I do wonder what it is though that causes them.

Hmm. Maybe if we tried the FP16 R300 mode on a real Nv3x.



Ok, So long as we're understanding each other on the intention of the test, I'd have to "look" into the higher lighting. But as you can see from just these results. FP16 isnt causing the results for the Floor.

At moment. Not trying to "immitate" The r300 mode fully. Just distinguish the difference between FP16, FP32, and Nv3x :)

anzak
06-28-04, 05:48 PM
Ok, So long as we're understanding each other on the intention of the test, I'd have to "look" into the higher lighting. But as you can see from just these results. FP16 isnt causing the results for the Floor.

At moment. Not trying to "immitate" The r300 mode fully. Just distinguish the difference between FP16, FP32, and Nv3x :)

What about the glass in jakup's shots? Could that be caused by very high lighting?

ChrisRay
06-28-04, 05:50 PM
What about the glass in jakup's shots?


I'm not sure I understand the question? :)


Ok *edit* I understand your question. Ya the glass seems to be caused by "Very High LIghting" which isnt enabled in this comparison.

Skinner
06-28-04, 06:31 PM
I have the impression it still doesn't look as good as with a 9800/X800, the shinyness is what I'm talking about. I remember someone took this shot with a X800 aslo..

Waffles
06-28-04, 06:45 PM
EXCELLENT work Chris. This pretty much shatters the theory that FP16 causes the banding.

I can't see any difference between FP16 and FP32 in r300 mode, other than the torpedo looks like it has just an Extra bit of shine to it in fp32 mode. Anyone else? I'd just take the extra speed.

anzak
06-28-04, 06:48 PM
I have the impression it still doesn't look as good as with a 9800/X800, the shinyness is what I'm talking about. I remember someone took this shot with a X800 aslo..

I have a Radeon 9600XT which produces the same image as the 9800/X800. The R300 modes with FP32 looks just like the IQ on my card.

Skinner
06-28-04, 07:04 PM
Well maybe its my imagination, if I find the x800 shot in the same place I'll post it. That's if my XFX6800U doesn't arrives first.

-=DVS=-
06-28-04, 07:17 PM
Glass looks as it should in Chris PICS , same look on Radeon cards , there was a small bug in Nvidia driver when you froce R300 mode glass looks different (buggy)

Glass look as it should
http://www.nvnews.net/vbulletin/attachment.php?attachmentid=6647

Driver shader bug by Forceing NV3x/4x to R300 , but not all driver got that bug.
http://www.nvnews.net/vbulletin/attachment.php?attachmentid=6648

Hope that cleared some things.

Look exacly as in Chris pic on Radeons with correct floor and glass.
http://www.nvnews.net/articles/chrisray_iq/Fp16floor.png

ChrisRay
06-28-04, 07:20 PM
Hope that cleared some things.


Appreciate it, It definately does :) But are you sure? your top pic looks as if it has the light banding present for the Nv3x mode.

Clay
06-28-04, 07:27 PM
Glass looks as it should in Chris PICS , same look on Radeon cards , there was a small bug in Nvidia driver when you froce R300 mode glass looks different (buggy)
Hope that cleared some things.
Can you provide some links to back this up? I'm not doubting you personally, I've just heard/read to the contrary and haven't seen any real proof of those claims either.

Thanks. :)

bmg
06-28-04, 07:35 PM
Would it be possible to explain how to get the two R300 modes (16/32)? I'm familiar with the Tommti Systems fix as well as the "shaders" method. What's the "correct" way to get FP16 and FP32 R300 mode?

-=DVS=-
06-28-04, 07:39 PM
I used pics from nvnews for easyer explanation , i don't see why someone would think im makeing this bug thing up , its not something fanboyish :rolleyes: , i don't keep webpages to back my knowledge , but obvioulsy drivers Chris use are good ones , so just try other set and you will see for your self, by the way witch part of proof you guys asking for NV40 or that Radeon 9XXX runs farcry floor without banding and correct glass , as shown in above pics ?

ChrisRay
06-28-04, 07:43 PM
Would it be possible to explain how to get the two R300 modes (16/32)? I'm familiar with the Tommti Systems fix as well as the "shaders" method. What's the "correct" way to get FP16 and FP32 R300 mode?


That is the correct way, I use the tommti system and it seems to be producing the correct results.

Clay
06-28-04, 07:43 PM
I used pics from nvnews for easyer explanation , i don't see why someone would think im makeing this bug thing up , its not something fanboyish :rolleyes: , i don't keep webpages to back my knowledge , but obvioulsy drivers Chris use are good ones , so just try other set and you will see for your self, by the way witch part of proof you guys asking for NV40 or that Radeon 9XXX runs farcry floor without banding and correct glass , as shown in above pics ?Easy man, nobody is suggesting that you are making this up. :ORDER: :)

Anyway, I was asking about the glass being rendered correctly (or not) in the first screenshot. The second screenshot is what I had been led to believe was the "correct" rendering.

Again, I'm not suggesting anything here. I'm just trying to find proof that confirms this question one way or the other.