View Full Version : 3Dmark05 High and better IQ...
http://www.beyond3d.com/forum/viewtopic.php?t=16734&postdays=0&postorder=asc&start=0
just to keep you guys upto speed with what is being discussed there...
so without the mist performance is increased? i know one thing be it fog or mist theres always a performance hit, maybe ati left the mist out to increase performance.
I'd wait for some sort of confirmation though before reaching that conclusion. Watching a movie of part of the scene captured with fraps, it's real obvious the ATI side isn't rendering nearly as much mist but maybe something is messed up with my ATI config.
Toaster
10-03-04, 03:50 PM
No, one technique is not noisier. ATI is doing the proper soft shadows and Nvidia is using DST so the results are much sharper and therefor incorrect. DST does not adhere to DX9 specs and the outputted image is incorrect, there is no arguement there. At the moment Nvidia is gaining 10% - 15% in performance because of this.
NV40 is right, there is no standard way of doing shadows. The reason they differ is because on the ati boards the pcf has to be done in a fragmentshader while the NV cards have dedicated HW support for PCF. Both perform 4 texture lookups, do the depth compaire for each of the 4 texels and average the result (with NV's pcf doing a linear filter on the result). The reason that the shadows look softer on ati-boards is because the 4 samples are taken from a slightly larger area (advantage of doing it in a shader instead of FF).
DST is a texture format like 3Dc, nothing more nothing less.
Because I'm not very familiar with DX I'm also wondering if the default R32F texture format is a depthtexture or not because if it's not then DST has another advantage because it can be rendered to in 32x0 mode.
This link (http://msdn.microsoft.com/library/default.asp?url=/library/en-us/directx9_m/directx/direct3d/gettingstarted/direct3dsurfaces/surfaceformats.asp) doesn't mention it with the depthbuffer formats but it also doesn't say it can't be used as a depthtexture...heh
P.S. Why didn't anyone complain about this untill 3dmark05? SplinterCell uses it and no one seemed to mind..
P.S. Why didn't anyone complain about this untill 3dmark05? SplinterCell uses it and no one seemed to mind..
since this is a synthetic benchmark... any changes to load on the cards leads to results that are not accurately comparable...
if ati is indeed doing something to reduce load it is a slightly less valid comparison... if this is something that is DST specific thats a different story... some of scali and DB's remarks on b3d seem to indicate it is non-DX related though I hope one or both reports with a less cryptic/more leymans terms analogy of whats happening lol...
otoh i find it odd that no one seems to care that dst is indeed giving a performance boost and is IHV specific in its usage...
NoWayDude
10-03-04, 04:19 PM
So, the explanation seems to be a diference in rendering when AA is requested by the Catalyst CP or the app CP.(acording to B3D forum posters)
Why is this then?
everybody seems to forget that futuremark themselves claims each HW would render diferently, that the engine they are using detects which HW is installed, in my eyes this isn't made for an apples to apples comparison between different HW
Toaster
10-03-04, 04:45 PM
since this is a synthetic benchmark... any changes to load on the cards leads to results that are not accurately comparable...
if ati is indeed doing something to reduce load it is a slightly less valid comparison... if this is something that is DST specific thats a different story... some of scali and DB's remarks on b3d seem to indicate it is non-DX related though I hope one or both reports with a less cryptic/more leymans terms analogy of whats happening lol...
otoh i find it odd that no one seems to care that dst is indeed giving a performance boost and is IHV specific in its usage...
Well when using the DST option it's NV that somewhat "reduces" the workload because it has specialised HW support so that the ati cards need to run a lenghtier pixelshader for a similar effect. (Unless I misunderstood you, haven't read the b3d thread(s) )
And regarding the synthetic benchmark workload equality, if a card has HW support for an option and the competition does not, does this mean it should not be used because the workload would no longer be exactly the same?
It's possible to get the exact same shadows as ATI's on NV hardware _and_ still use the HW assisted depth-compaire. It will be slightly slower than the current setup but it's still faster than disabling the DST option altogether.
In other words, exact same image will get rendered (as far as shadows go) but it will still use IHV specific HW functionality resulting in a bit shorter shader vs what ATI cards would do. Would this be considered fair or not??
if this is something that is DST specific thats a different story... some of scali and DB's remarks on b3d seem to indicate it is non-DX related though I hope one or both reports with a less cryptic/more leymans terms analogy of whats happening lol...
otoh i find it odd that no one seems to care that dst is indeed giving a performance boost and is IHV specific in its usage...
i personally like the use of DST in 3dmark2005.. because it give us an idea of how diferent algorithms will do the same job.. personally i like more the looks of DST... shadows are more stronger.. ATI method try to make it soft and it doesnt look that well ,because of the noisy pattern in the edges.. remember that DST can be disable..and it is the method of incomming game engines apparently.. just because ATI doesnt support it in hardware,doesnt mean that boths shouldnt be compared..ATi will need to support it in next generation hardware if they dont want to stay behind in next generation games.. both techniques are used in its simplest form.. so thats why you dont see perfection in neither one.. however i agree that non-DST shadows should be added in the benchmarks for the sake of better comparisons..
by default it is on.. the workload is not the same nv40... db and rev have explained that about as well as can be explained...
IHV specific opts are a bit of a wash when discussing a synthetic benchmark... ESPECIALLY if they are on by default and users of the free version can't turn it off afaik...
i personally like the use of DST in 3dmark2005..
Indeed nice for Nv card owners, but not by default. Futuremark should have made a checkerbox where you can put it on, not "off". :lame:
And btw, a little bit off-topic but it seems that Nv uses DST too as a default in FarCry. Check the shadows on the gun:
http://www.xbitlabs.com/articles/video/display/farcry30_4.html
"ATI RADEON X800-series produces smoother shadows in FarCry when compared to NVIDIA GeForce 6800-series."
it looks to me that ATI and Nvidia effects (even when the frames are the same ) are out of syncronization by 1 second or so.. that might explain why obvious diference in position of the fire or smoke at times .. in some shots..
just like here...
this time using ATI screenshot provided by [APPLE] at ...6xaa/8xaf (http://members.chello.nl/h.appel1/6.jpg )
Nvidia screeshot by Bjorn3d 4xaa/8xAF (http://www.************/photos/showphoto.php/photo/33/sort/1/cat/503/page/1)
the sync argument could be true however if it was true the ship position would be off slightly too. open up the bjorn one and then the frame other ones to see what im talking about the ship slightly moves up right. also the smoke in frame 1830 ati is smaller then the nvidia one but still showing a light on the aft of the ship
anyways the missing shadow or light source has been marked, so as i said either ati is not rendering the shadow correctly or nvidia is not rendering the light correctly. theres also the difference of the smoke.
in anycase would anyone who has a nvidia card be willing to take frames 1830 and 1832? of gt3?
bjorn reference 1831 gt3 (http://home.comcast.net/~ntxawghaam/nv_gt3_aa_1831.jpg)
same frame ati 4x8x (http://home.comcast.net/~ntxawghaam/ati4x8xgt3_1831.jpg)
ati 4x8x frame 1830 (http://home.comcast.net/~ntxawghaam/ati4x8xgt3_1830.jpg)
ati 4x8x frame 1832 (http://home.comcast.net/~ntxawghaam/ati4x8xgt3_1832.jpg)
Okay, here is one where it's a little more obvious that, at least with my ATI config, they're not rendering all the mist. Taken from frame 1410 of the Canyon Flight game test.
ati...............................reference image...............nvidia
http://img26.exs.cx/img26/2586/ati_807_gt3_1410.th.jpg (http://img26.exs.cx/my.php?loc=img26&image=ati_807_gt3_1410.jpg) http://img26.exs.cx/img26/9236/ref_image_gt3_1410.th.jpg (http://img26.exs.cx/my.php?loc=img26&image=ref_image_gt3_1410.jpg) http://img26.exs.cx/img26/7656/nv_6681_gt3_1410.th.jpg (http://img26.exs.cx/my.php?loc=img26&image=nv_6681_gt3_1410.jpg)
ChrisRay
10-03-04, 06:41 PM
Okay, here is one where it's a little more obvious that, at least with my ATI config, they're not rendering all the mist. Taken from frame 1410 of the Canyon Flight game test.
ati...............................reference image...............nvidia
http://img26.exs.cx/img26/2586/ati_807_gt3_1410.th.jpg (http://img26.exs.cx/my.php?loc=img26&image=ati_807_gt3_1410.jpg) http://img26.exs.cx/img26/9236/ref_image_gt3_1410.th.jpg (http://img26.exs.cx/my.php?loc=img26&image=ref_image_gt3_1410.jpg) http://img26.exs.cx/img26/7656/nv_6681_gt3_1410.th.jpg (http://img26.exs.cx/my.php?loc=img26&image=nv_6681_gt3_1410.jpg)
I am gonna upload these pics to my website and do a comparison, I need to download the hotfix ect, I wanna see this for myself.
Okay, here is one where it's a little more obvious that, at least with my ATI config, they're not rendering all the mist. Taken from frame 1410 of the Canyon Flight game test.
thats a nice example... clearly shows what you are talking about... :thumbsup:
post it up on b3d if you have not done so already...
ChrisRay
10-03-04, 06:45 PM
thats a nice example... clearly shows what you are talking about... :thumbsup:
post it up on b3d if you have not done so already...
I wonder if there is a difference in the drivers ect, Wondering if I can replicate this.
Chris
I wonder if there is a difference in the drivers ect, Wondering if I can replicate this.
Chris
from what I am gathering from scali's comments... the refrast image may well be done with dst as well... which would explain why the nvidia shot looks like the refrast...
I don't quite follow his comments, but db's witty one-liner seems to point in that direction...
/me shrugs
I have tried with cat 4.7's and 4.9's... no difference i can see in anything...
I haven't tried the new uber drivers though...
ChrisRay
10-03-04, 06:56 PM
from what I am gathering from scali's comments... the refrast image may well be done with dst as well... which would explain why the nvidia shot looks like the refrast...
I don't quite follow his comments, but db's witty one-liner seems to point in that direction...
/me shrugs
I have tried with cat 4.7's and 4.9's... no difference i can see in anything...
I haven't tried the new uber drivers though...
Care to point me to thread? I doubt the fog would have anything to do with DST..
Its still get in business time and I need some things pointed out to me ;)
its just what I gathered from the thread... considering I am not as informed as you and others on this subject I may well have it wrong :D
http://www.beyond3d.com/forum/viewtopic.php?t=16734&postdays=0&postorder=asc&start=40
thats the 3rd page of 4 for me... a little ways down there are comments by scali and DB about the refrast...
I went back to the 4.7s and I've tried the 4.9s but had the same result.
Sazar, are you saying that frame 1410 with your config looks like the reference and nvidia shot? If so, it almost has to be a config problem with my system. I haven't noticed a difference with the 4.7s, 4.9s or 8.07s. So, it's either something that has been there a while or maybe something is just screwed up with my system. All the other IQ shots I've tried look fine though so I don't know. I have the ATI CP set to app defaults and I'm just using the defaults for the IQ screenshot (except I'm changing which frame I want) -- if that helps.
Anyway, I'd prefer that someone else duplicate what I'm seeing before taking it elsewhere just in case it's me that is the problem. :)
Oh, I'm still using 1.0 at the moment also.
I only have the free version m8... there is no way for me to freeze frame it @ that point... :(
if I could, I woulda provided a screen cap... sounds weird considering its almost a slide show on my card anyways lol... but its true :D
ChrisRay
10-03-04, 07:03 PM
I went back to the 4.7s and I've tried the 4.9s but had the same result.
Sazar, are you saying that frame 1410 with your config looks like the reference and nvidia shot? If so, it almost has to be a config problem with my system. I haven't noticed a difference with the 4.7s, 4.9s or 8.07s. So, it's either something that has been there a while or maybe something is just screwed up with my system. All the other IQ shots look fine though so I don't know. I have the ATI CP set to app defaults and I'm just using the defaults for the IQ screenshot (except I changing which frame I want) -- if that helps.
Anyway, I'd prefer that someone else duplicate what I'm seeing before taking it elsewhere just in case it's me that is the problem. :)
Oh, I'm still using 1.0 at the moment also.
WOrking on It OWA, I just need about 15 more minutes, Its a question of getting software installed on Second machine :) And then converting Bmp to .png ect. Also need to install it on Primary machine again, So I am guessing its gonna take me about 20 Mins ish to get some kind of verification,
Because I have to run it my 9800 Pro and then my 6800 card.
ChrisRay
10-03-04, 07:05 PM
Indeed nice for Nv card owners, but not by default. Futuremark should have made a checkerbox where you can put it on, not "off". :lame:
And btw, a little bit off-topic but it seems that Nv uses DST too as a default in FarCry. Check the shadows on the gun:
http://www.xbitlabs.com/articles/video/display/farcry30_4.html
"ATI RADEON X800-series produces smoother shadows in FarCry when compared to NVIDIA GeForce 6800-series."
That was with old driver set, its been fixed in future drivers.
I only have the free version m8... there is no way for me to freeze frame it @ that point... :(
if I could, I woulda provided a screen cap... sounds weird considering its almost a slide show on my card anyways lol... but its true :D
Slideshow on my 6800 modded to 16x1/6 Vertex Units, So dont feel bad, heh
ChrisRay
10-03-04, 07:31 PM
You have been vindicated a bit OWA, Theres a huge discrepency, I am gonna post your and my findings over at b3d, to see if it can be explained in a moment.
My 6800.
http://members.cox.net/omega1979/6800canyon.PNG
Radeon 9800 Pro
http://members.cox.net/omega1979/aticanyon.PNG
Thanks for checking into it. I was starting to get a little nervous that it might be just me. :)
ChrisRay
10-03-04, 07:45 PM
Thanks for checking into it. I was starting to get a little nervous that it might be just me. :)
You are definately not crazy. If there is such an error, We are both are able to replicate it.
Forgive me if i quoted you at b3d, But I felt it was worth looking into. I opted for higher quality .png files.
Also guys please dont link these pictures across various forums. They were uploaded for the Nvnews/Beyond3d threads. Not to link around the internet like some floozy :)
vBulletin® v3.7.1, Copyright ©2000-2012, Jelsoft Enterprises Ltd.