Trilinear Filtering Comparison(R420 vs R360 vs NV40 vs NV38)

hehe, ya... it worked for me. I was kinda loosing it there.. lol.

Actual Location: video2/images/r420xt/r420-anis0x.jpg

Wrong Location:video2/images/r420xt/r420-anis00-00.jpg

 

If you compare these two pictures

http://www.ixbt.com/video2/images/r420xt/r420-anis16-00.jpg
http://www.ixbt.com/video2/images/nv40/nv40-anis16-00.jpg

You can see that the r420 image has heavier image compression applied, it's also 100k less than the NV40 image, which is a bit odd.

BTW you will probably have to copy and paste the links.

EDIT: it's the same with some of the other SS as well, the ATI SS's have more aggresive compression.

 

Interesting... :?

 

If you're using IE (maybe other browsers too) you can click+drag the link to the address bar or to the IE title bar.

 

The r420 ss really looks like pure bilinear.
If I understood correctly what I gathered here today r420 use same filtering algorithm with r360, right?
But r360 ss looks just fine. It doesn't make sense.

By the way, I thought r420 & r360's implement apply adaptive trilinear only with aniso filtering, and use full trilinear without aniso, was I wrong?

 

I think it's the rv360 that should be the same. Although Ati's PR was of another opinion when Dave asked them (as in, rv360 should'nt be capable of this optimization).

 

Wow. I missed big V.

 

I think this might be a more advanced version of what Dave saw in the 9600, maybe? If I remember, when texture slider is down one notch on the 9600, only then did you get brilinear.

 

Your point is valid, although I don't think the way the MIP transitions are poor in the R420 shots is caused by bad compression... it seems to follow along the points in the geometry that a MIP transition would. Also, the quality of the rest of the image is right on par with the NV40 shot.

Still, it should be properly re-done with the same compression settings on all images involved.

 

I checked the headers of both files. It seems the quantization is higher on the r420 file (which looks like: 06 04 05 06 ... for L and 07 07 07 0A 08 ... for C). The nv40 file is lower (02 01 01 01 ... for L and 02 02 02 02 ... for C). Lower quantization values result in less noises.

I think someone should do this experiment again, with lossless compressed files for better comparison (and of course, correct settings).

If someone could make a movie it would be much better, but the bandwidth requirement may be just too prohibitive.

 

In the anistropic filtering shots the banding issue seems to disappear (and quality is on par or better than the R360) and I scanned around some other R420 TF shots and they don't show similar results. Maybe someone should ask the ixbt guys to run that again on their x800 or maybe someone else could?

 

Just because there is a size difference in the JPG doesn't mean the same settings weren't used.

If you use a program like Photoshop, and say set the JPG quality to "10," it will use as much compression as it can to achieve a quality "10" JPG image. The smooth gradient in the NV40 image was likely more difficult to compress without artifacting than the less smooth gradient in the ATI image, so you got a larger filesize on the NV40 image. I've saved different high resolution JPGs with the same resolution at the same compression level and have gotten sizes between 900kb and 3.5mb depending on the content of the image.

 

You really shouldn't make stuff up...

 

As noted above by pcchen, this doesn't appear to be the case.

You've taken a screenshot at the same location and gotten a file that was 4x larger? I think not.

The disparity in file size is much too great here just to be compression variance, they clearly must have used a different setting. Using lossy compression for IQ comparisons is not ideal, especially when you are looking for very small variances. Using different levels of compression for comparitive IQ screenshots is just plain stupid.

 

Anyone with a X800 that can verify if the X-Bit labs screenshot are taken with the correct settings ?

 

Use PNG for screenshots. Lossless compression, files should be around 1-2MB depending on resolution.

 

[OT]

Sorry to bust in like that, but it's 3PM already and no chat?

[/OT]

 

It's 10 mins before that, your clock's living in the future

 

Damn it! I just added 6 hours to my time, not very accurate. Thanks

 

Didn't I read somewhere it would be open an hour earlier for people to join up? Or does that mean nothing is going to happen until 4pm EST?

EDIT: nm it just opened