ATi Radeon 8500 Review/Investigation (Part1)

"Greatly reduces stuttering"

Sounds like the placebo effect to me. I'm here to tell you that I still get tons of stuttering and sharp framerate drops, and that's with the '43 drivers in win2k and the '21 drivers in win98. BTW, those are both fresh loads of the OS due to a hard drive crash a couple of months ago.

I'm not interested in taking this thread further, simply put, I *KNOW* the problem is still around.

 

Then what can I say. Other people don't have your problem.

 

Bearing in mind at not one point is the frame rate drop below about 30fps, so turn off the frame rate counter, and you wont notice anything.

This looks exactly like the hard driver thrashing issue Wavey had with the 6025 set whilst he was reviewing. Also we are talking about 1 or 2 drops in each graph. Hardly a 'stutterfest'. In fact the rest of the time the 8500LE is clearly ahead of the Ti200 from 1024x768 onwards.

Actually the fact the 128meg 8500 doesn't suffer from it, points to it being a texture management issue.

 

Good review. I enjoyed it.

My commentary on current debates-
Driver revision- I don't care *what* policy a site adopts for driver revision as long as it's consistent at the time of the review and remains consistent for all reviews.

Sites that lump original CD drivers for one product, and fresh off the leaked FTP server drivers for another lack credibility. By the same token, a site that tests CD drivers vs CD drivers for one review, then most recent Official Website drivers vs same on the next- then back to the original CD methodology for a 3rd review.. also begs the question of "why the change?" So recurring consistency is of the absolute most importance in my book.

On OGSS/SmoothVision and the like for the 8500, here's how I see it-

The product "rumors" before actual release conflicted with the final delivered product. The "rumors" and tradeshow commentary (some from this site) included:
* multiple sample buffers of same resolution as framebuffer
* a jittered, non-uniform sampling pattern
* a random displacement of the jittered pattern based on some factor

It seems the delivered product only comes with:
* a jittered, non-uniform sampling pattern
* the random displacement of the jittered pattern only functions in special circumstances.

From the Whitepaper, SmoothVision has a possible set of sample patterns that basically resembles:
1-----2
--3-4--
-------
-5-6---
7-----8


The whitepaper also documents that, for example, 4x SmoothVision can use sample positions from the above grid that include sample positions:
(1,2,5,8) or (2,6,7,8) or (1,2,3,7) or (2,4,7,8).

And lastly, the whitepaper also documents, albeit indirectly, that the 8500 does not use multiple sample buffers but instead prefers a single, superscaled sample buffer.

From a developer email shared at Rage3D, a support person from ATI replied that when fog is enabled (in the context of the query, this was in Direct3D), the random sample pattern is no longer "random" but falls back to use a single, fixed pattern for the full screen. So, hypothetically- with FOG enabled in Direct3D, of those 4 possible sample choices, it would seem to use only one of those for the full display.

To indentify the difference, I used a game that allows a static replay frame to be captured in the same spot as well as the removal of fog. The sampling method used is visibly different, albeit somewhat subtle and needs magnification to see the difference:
http://shark_food.tripod.com/fog/nfs.html

The 400% shots are thumbnails. I urge those interested to right-click/open in new window the 400% shots and compare the left and right edges of the car to see the differences in the sampling. (especially the side windows). Fog-off is superior when closely inspected at magnification.

In conclusion, I don't see any evidence that disputes the white paper unless I am missing something. I don't believe the white paper nor the output results would suggest RGSS, but they do use a "random" jittered sample pattern- it's not like the typical 4 sample ordered but instead uses a slightly different sampling pattern. A simple comparison of the 8500 in 4xQSV versus, say, a Radeon 64DDR or GeForce Ultra in 4xOGSS shows this quite apparently.

Again, if I'm missing something here, please do tell.

(EDIT- had to disable smilies from all the 8) marks in the sample patterns were sunglass wearing smilies. heh)

 

Well in the present drivers the fog does force it back to and ordered grid pattern. In the first SV drivers fog had no effect. I'm not convinced that the fog is a hardware issue as surely it would affect ogl too?

Also this:

1-----2
--3-4--
-------
-5-6---
7-----8

is a possible set. This is the programable part of SV.

Also its only 2x and 4x modes of d3d that are non ordered atm. The rest look just like OGSS.

 

Damn, I forgot to login.