• We've upgraded our forums. Please post any issues/requests in this thread.

HD 5xxx Image Quality problem--could it be fixed or not?

Status
Not open for further replies.

Bo_Fox

New Member
Joined
May 29, 2009
Messages
480 (0.15/day)
Likes
56
Location
Barack Hussein Obama-Biden's Nation
System Name Flame Vortec Fatal1ty (rig1), UV Tourmaline Confexia (rig2)
Processor 2 x Core i7's 4+Gigahertzzies
Motherboard BL00DR4G3 and DFI UT-X58 T3eH8
Cooling Thermalright IFX-14 (better than TRUE) 2x push-push, Customized TT Big Typhoon
Memory 6GB OCZ DDR3-1600 CAS7-7-7-1T, 6GB for 2nd rig
Video Card(s) 8800GTX for "free" S3D (mtbs3d.com), 4870 1GB, HDTV Wonder (DRM-free)
Storage WD RE3 1TB, Caviar Black 1TB 7.2k, 500GB 7.2k, Raptor X 10k
Display(s) Sony GDM-FW900 24" CRT oc'ed to 2560x1600@68Hz, Dell 2405FPW 24" PVA (HDCP-free)
Case custom gutted-out painted black case, silver UV case, lots of aesthetics-souped stuff
Audio Device(s) Sonar X-Fi MB, Bernstein audio riser.. what??
Power Supply OCZ Fatal1ty 700W, Iceberg 680W, Fortron Booster X3 300W for GPU
Software 2 partitions WinXP-32 on 2 drives per rig, 2 of Vista64 on 2 drives per rig
Benchmark Scores 5.9 Vista Experience Index... yay!!! What??? :)
#1
HD 5xxx Image Quality Issue--could it be fixed or not?

See here: http://forums.techpowerup.com/showthread.php?t=106814&highlight=HD+5xxx+Image+Quality




According to this excellent article on this issue here: http://alienbabeltech.com/main/?p=12648&page=2 I am still stuck with my 4870 1GB until the IQ problem is resolved. It's still there with 10.6 Catalyst drivers on the 5870--see (click on 720p and then FULLSCREEN to see how BIG the blurred part of end of the tunnel is, yet the problem is still profound at tiny 240p compressed resolution)

http://www.youtube.com/watch?v=uYd_LLRbGHU

Now, see how the 4870 doesn't have this problem, with the same tunnel in Half Life 2:
http://www.youtube.com/watch?v=Z_vICZsMTQE


Also, PCGamesHardware.com points out the differences in the links below but the examples are rather poor with the two lines hardly detectable in doo-doo brown textures. The first one is taken on GTX 480 and the second on 5870--
To easily compare, please open both links in new tabs, so that you could switch back and forth to see the differences:

http://www.pcgameshardware.com/aid,743498/Geforce-GTX-480-and-GTX-470-reviewed-Fermi-performance-benchmarks/Reviews/&menu=browser&image_id=1269592&article_id=743498&page=1&show=original

http://www.pcgameshardware.com/aid,743498/Geforce-GTX-480-and-GTX-470-reviewed-Fermi-performance-benchmarks/Reviews/&menu=browser&image_id=1269593&article_id=743498&page=1&show=original

The image that shows the difference in light contrast when both are compared is here:
http://www.pcgameshardware.com/aid,743498/Geforce-GTX-480-and-GTX-470-reviewed-Fermi-performance-benchmarks/Reviews/&menu=browser&image_id=1269995&article_id=743498&page=2&show=original
(I cannot paste the image here since it says "No Deeplinking")


In the picture below, on the left, it's rendered in hardware on 5870 (there is a sharp boundary where everything just blurs into
solid gray--lots of detail is lost right at the boundary), and on the right, it's rendered in software mode, which is the reference AF image with full trilinear:

It's a new program from 3DCenter--details here: http://www.3dcenter.org/3dtools/filter-tester-en

Hopefully Microsoft will make it a requirement for DX11.1 or DX12 that the AF quality matches the reference IQ with full trilinear. "Brilinear" was an issue several years ago, and it's 2010.. hello? Quoted from SEVERAL years ago by Tomshardware:
""Anisotropic filtering always functions in interaction with bilinear or
trilinear filtering. The bilinear variation is logically faster but has one
disadvantage: the visible edge on the mipmap transitions. The best result is
achieved with the more computing-intensive trilinear anisotropic filtering, in
which mipmap transitions are calculated from the neighboring textures."

from http://www.tomshardware.com/reviews/ati,819-5.html

The problem here is the undersampling of the base textures whenever the new AF algorithm is being applied on 5xxx cards for a strongly bilinear effect at certain distances (lines of demarcation, or sharp breaks in the smoothness of texture transition). There is no way to disable it on 5xxx cards (changing Catalyst A.I. settings does not get rid of the breaks, nor does sharpening the LOD to negative values) but it can be somewhat mitigated with SSAA.

As one guy said very well: "I can imagine a bug related to texture caching. Is it possible that a part of texture cache is unaccessible due to a design bug? The remaining capacity wouldn't be sufficient for full trilinear, but replacing one stage with bilinear would make the capacity sufficient. I'm not saying this is the case of Evergreen, it's only an example."

I did some more research on this and found a german site suggesting that it could be a hardware "bug":

So we have information on hand that the filter is still flickering stronger than on current Geforce graphic cards, which relates to under-sampling. By this, although ATI saves two frequencies of the texture units and so gains higher performance, the screen quality has to suffer.
http://www.hardware-infos.com/news.php?news=3208&sprache=1

So it seems to be un-fixable after all, I would conclude. (Hopefully I'm wrong.) Oh well.. *crossing fingers* that S.I. will utilize full trilinear in combination with AF--I really look forward to buying it especially if it has upped memory bandwidth.
 
Last edited by a moderator:
Status
Not open for further replies.