View Full Version : do you still think it's only 3dmark ?
How do you feel looking at these numbers?:
http://www.anandtech.com/video/showdoc.html?i=1821&p=27
And knowing that 4xAA is the most used AA seting in game review:
http://www.hothardware.com/hh_files/S&V/gffx5900u(3).shtml
(...also look at quake3 numbers on page 5....)
and AF ??? thoose loosy explanation on the blury/buffer subject ?
(and no explanation of influence of post processing on FPS)
Does AF/AA and co work the same in the synthetic AF/AA tests (thoose all colored tests with the right shapes where you clearly see AA and AF in action) and in games where specific "skip/replace" codes are applied?
Where are Nvidia and ATI going ????? Where are we going ???
solofly
05-24-03, 06:41 PM
Read...
http://www.hothardware.com/hh_files/S&V/r9800256mb_gffx5900upd(2).shtml
rellingsen
05-24-03, 07:33 PM
Thanks for the links! After I seen the diff. in the AA quality of the 2 cards I know I won't be getting a 5900 Ultra even if it is 20-25% faster with AA on. What a blurry mess the 5900 screens are. Makes me wonder how they obtained the 20%+ lead. HH even captured the screens correctly with HyperSnap and Post Proc capture enabled. To me AA is important as I play all my game with at least 2X AA with it mostly set at 4X. I will sacrifice 10FPS not to have that blur. What a diff. when I opened them up in seperate windows and flipped back and forth between them. Look at the landing above the gun and even the tiles in the floor. Links to full pics.
5900 2XAA
http://www.hothardware.com/reviews/images/fx5900/nv35utaashots/nv352xaaut.htm
9800 2XAA
http://www.hothardware.com/reviews/images/fx5900/r9800pro/9800utaashots/R9800ut2xaa.htm
5900 4XAA
http://www.hothardware.com/reviews/images/fx5900/nv35utaashots/nv354xaaut.htm
9800 4XAA
http://www.hothardware.com/reviews/images/fx5900/r9800pro/9800utaashots/R9800ut4xaa.htm
5900 6XAA
http://www.hothardware.com/reviews/images/fx5900/nv35utaashots/nv356xaaut.htm
9800 6XAA
http://www.hothardware.com/reviews/images/fx5900/r9800pro/9800utaashots/R9800ut6xaa.htm
It's not finished yet...
a little Doom3 now (since their comment only give credits to games like doom3 and ut):
http://www.beyond3d.com/forum/viewtopic.php?t=5959
http://www.beyond3d.com/forum/viewtopic.php?t=5956
Doom3 scores higher with AA turned on ?????
Originally posted by rellingsen
What a blurry mess the 5900 screens are.
You think those are bad? Take a close look at these:
No AA
http://www.nvnews.net/previews/geforce_fx_5900_ultra/images/q3_screenshot_1.png
4X AA
http://www.nvnews.net/previews/geforce_fx_5900_ultra/images/q3_screenshot_2.png
Originally posted by Zurble
How do you feel looking at these numbers?:
http://www.anandtech.com/video/showdoc.html?i=1821&p=27
I feel like he needs to re-run those benchmarks on the NV35. They are incorrect.
rellingsen
05-24-03, 08:10 PM
Boy it gets worse with every pic I see. :eek: I guess we can write off anybody using AA with their new $500. 5900 Ultra cards. Well at least if they have a decent monitor they can play at 1600x1200 and reduce some of the jaggies. :p
Perhaps anand did it wrong, or perhaps he used a driver version with "specific optimization", anyway there is enought evidence for me to be upset against both nvidia and ati (remember the quack issue).
Like DSC suggested, the best thing to do is to "voice our displeasure"...
http://www.nvnews.net/vbulletin/showthread.php?s=&threadid=12336
...to both of them
Originally posted by rellingsen
Boy it gets worse with every pic I see. :eek: I guess we can write off anybody using AA with their new $500. 5900 Ultra cards.
I forgot to mention an important point about the 4X AA image. That's what it looked like when I took a screenshot. What I saw on-screen looked more like the image with no AA.
rellingsen
05-24-03, 08:38 PM
So HyperSnap is not capturing all the post processing? Those images do seem really bad. I don't know a lot about 3D rendering but is it going thru some kind of sharpening proc. in post processing? A lot of reviewers have been saying negative things about NVidia's AA. So it's really not that bad?
Edit: for clarity and errors
vampireuk
05-24-03, 08:46 PM
Mike had a good post up a few weeks back that pointed this out in regards to the AA on the 5900. To put it simply a capture does not do it any justice:)
How do post-processing remove the blur ???
Does this mean AA is applied in post processing for 4xAA (only?) ????
i am not an expert, but it seems a bit confusing
Originally posted by rellingsen
So HyperSnap is not capturing all the post processing? Those images do seem really bad. It don't know a lot about 3D rendering but is it going thru some kind of sharpening proc. in the post filter? A lot of reviewers have been saying negative things about NVidia's AA. So it's really not that bad?
I noticed that screenshots taken with the NV35 were mis-representing 4X AA image quality. I even got a second and third opinion from my wife and son :)
This is an area that needs to be investigated further, but the difference was obvious in Quake 3, which is OpenGL. Haven't had a chance to investigate Direct3D. I'd also like to find out if it's related to hardware or software (drivers).
Maybe the drivers detect when Hypersnap is running and turns up the FSAA for the snapshot? After all, the screenshots taken when not using Hypersnap looks like very little or no FSAA is being applied.
Originally posted by Zurble
How do post-processing remove the blur ???
Does this mean AA is applied in post processing for 4xAA (only?) ????
i am not an expert, but it seems a bit confusing
That's why I've presented NVIDIA with my findings and asked them to elaborate.
Another possible answer could be that the FSAA is really not all that great on the GFFX and the drivers are detecting Hypersnap and applying a blur filter to make the snapshot appear to be a lot smoother than in reality.
Typedef Enum
05-24-03, 09:15 PM
Just to be clear about Quake3, in particular...
Those numbers are absolutely, positively _wrong_. Do not think for a moment that they _might_ be right, under any circumstance.
Look at any other NV35 preview, and you will see the 9800/5900 be very close to one another...
I'm just glad that Anand took the time to verify that those numbers are not only correct from a benchmarking perspective, but that they also seem reasonable, given just a touch of common sense.
It's been 2 weeks now, and I know for a fact that people have emailed him letting him know that he's flat out wrong.
Typedef Enum
05-24-03, 09:19 PM
Yeah...
I also wish people would stop saying things like,
"I know the AA isn't too great right now, so I hope this can be fixed by a driver update..."
How on earth is it that people believe you can magically morph hardware to do things at the same speed/quality via software updates, as opposed to physical hardware?
Lets put this one to rest right now...It ain't going to get any better until the next _real_ piece of hardware is released. And even then, you're going to have to hope that this issue was on their "things to fix" list, and will be addressed through hardware.
StealthHawk
05-24-03, 10:18 PM
Originally posted by Zurble
How do you feel looking at these numbers?:
http://www.anandtech.com/video/showdoc.html?i=1821&p=27
Anand's numbers are wrong. You can see other sites where the Q3 1024 4x FSAA numbers are the same as Anand's, but on other sites, as the resolution increases, the scores drop. Anand's results at 1280 and 1600 are flat out wrong.
Many of these NV35 previews are less than impressive. Tom's NV35 Doom3 FSAA benchmarks were mislabeled too, they are now correct.
Hellbinder
05-25-03, 02:08 AM
Perhaps anand did it wrong, or perhaps he used a driver version with "specific optimization", anyway there is enought evidence for me to be upset against both nvidia and ati (remember the quack issue).
You need to read my thread on Quak in the other cards forum.
Besides wich you are talking about ONE event from ATi going on 2 years ago, coumpared to multiple infractions from Nvidia. Like Constant ones. In that thread im talking about I posted a link to show what Nvidias GF2 Drivers looked like in Quake 3. A completel Blurry mess compared to ATi. That was 3 years ago. The pattern continues year after year.
This is why i get a little miffed when i read posts Trying to Suggest that the *Quak* issue is even in the same Country as Nvidias constant tactics.
One last Time Specific *Optimizations* for games are NOT wrong. Period end of statement. As long as they dont sacrifice Quality for speed (in general). Cheating is simply not the same thing as Optomizing. In my book Cutting Corners hacking just to win benchmarks in *anything* is Cheating.
On this one you are wrong. Optimization can only occurs in routines without changing specifications (read tim sweeney response). When you can run DX9/Opengl card independant routines faster without changing specs at all, this is optimization.
When you change routines specs, you cheat whatever you notice or not the difference.
Many don't see the difference between 1024 2xAA 8xAF and 1280 4xAA 16xAF, so if ATI or Nvidia decide to replace one by the other saying it is not noticiable is realy bad faith.
As soon as there is some sort of code that detect what software is running in any driver, IT IS CHEAT. and the "bug correction"/"optimization" thing is only an excuse.
If NVIDIA or ATI thinks a game is not programmed correctly for their hardware, they should release an independant patch "UT2003_NV35.EXE" and NETHER do it in drivers.
As a company game i could sue them for copyright violation otherwise.
Originally posted by Zurble
How do post-processing remove the blur ???
Does this mean AA is applied in post processing for 4xAA (only?) ????
i am not an expert, but it seems a bit confusing
and where do this blur comes from ?
Originally posted by MikeC
I feel like he needs to re-run those benchmarks on the NV35. They are incorrect.
Yeah we had a thread and folks were asking them about it.... after a few days with no answer they came back with
"We're looking into it....."
After a few mopre days of NOTHING ( didn't figure on anything anyways), there was more talk.....
They deleted the thread, so I started another asking where it went.
Those Numberrs are FOUL and they KNOW it.
Yet they do nothing and say NOTHING.
Pretty sad
vBulletin® v3.7.1, Copyright ©2000-2012, Jelsoft Enterprises Ltd.