Tom's GF4 anisotropic scores in his Parhelia review

I did notice this when I first read the review but as I don't really take notice of the numbers but rather the objective description I didnt take it on board.

I am running the 29.80's which seem to a little faster. But aniso doesn't seem much faster.

I can only think that maybe the LOD setting was lowered on the GF to make a difference or maybe they have the new det 5's, which are supposed to offer a significant improvement in aniso speed, and were lying about using 29.42's.

Or maybe they goofed up the scores.

This is going to cause more mud slinging @ THG :-?
 
John,

From the top of my head the L8/4x score doesn´t sound exaggerated since he used a 2.2 P4, as long as he used "performance mode". Performance with aniso on drops drastically if you use "quality mode". I´m using 29.42 WHQL with Rivatuner RC11 myself, so I´ll check out the differences between L2 and L8 aniso when I get home.

Could it be that drivers flip between performance and quality mode when no 3rd party tweaking utility is used? If yes then the performance delta between L2/quality and L8/performance could make sense. (only guestimating here).
 
I did a little benchmarking betweem the Quality/Performance modes, and it was not very significant...Maybe like 6-7 FPS, but not huge...
 
I ran some GF3 scores for comparason:

Q3A 1.17, demo001, no sound, texture/color depth @ 32/32, high geometry, max texture slider, no tweaks, 29.60 drivers, P4 2.4GHz (533), PC2700, i845G chipset, AGP4X, FW @ SBA enabled, GF3 64MB

1600x1200
Norm 98.4fps
Quin anis L2 34.8fps
Quin anis L8 29.6fps
4x anis L2 41.7fps *
4x anis L8 34.5fps *
* 2x anis L2 & 2x anis L8 give the exact same scores, so the driver is probably falling back

1024x768
Norm 205.9fps
Quin anis L2 80.7fps
Quin anis L8 67.4fps
4x anis L2 71.5fps
4x anis L8 61.9fps

I highlighted the only res I play Q3A in. The other scores don't matter to me at all.
 
Typedef Enum said:
I did a little benchmarking betweem the Quality/Performance modes, and it was not very significant...Maybe like 6-7 FPS, but not huge...
Click to expand...

6-7fps?

29.42 WHQL
Rivatuner RC11
demo001/1024x768x32/LODcurveError@30000/max detail
4xOGMS

64tap quality optimisations: 73.9 fps
64tap performance optimisations: 86.9 fps

16tap quality optimisations: 82.3 fps
16tap performance optimisations: 92.1 fps

Assume there is a possibility that the drivers with no third party tweaking utility involved switch by themselves to "performance" mode in highest aniso settings how irrational is my theory since the performance delta as shown above is ~87 vs 82 fps?

13 fps might not represent a "huge" drop but drastically is not exactly exaggerated considering it's a healthy 15% drop.

PS: I reran each test 3 times.
 
Interesting...

When I made the original post, I did not enable FSAA, and the difference was about 6 FPS.

I just re-ran the same Q3 benchmark, but did the same thing as you: 8x AF + 4xFSAA, and this time, the difference was ~ 10 FPS.

I didn't really try to distinguish one from the other yet, so I'm not sure what the tradeoff really looks like...
 
Although I'm not completely sure I think quality mode filters lightmaps. I use performance anyway. If that theory should have any substance then I'd like to have a performance/quality option for D3D too.
 
I've run some comparisons on my GeForce4 Ti 4200:


1024x768x32, 4x FSAA:
2-degree aniso: 84.5 fps
8-degree aniso: 75.8 fps

1280x1024x32, 4x FSAA:
2-degree aniso: 47.8 fps
8-degree aniso: 44.0 fps

As you can see, those benchmarks are not at all unfeasible for a Ti 4600.

One thing you have to remember is that the GeForce3 and 4 cards all have reduced performance hits for aniso or FSAA when both are used together (since they impact different parts of the graphics card...FSAA impacts memory bandwidth, and aniso impacts fillrate). Additionally, higher degrees of anisotropic are used less, which means less of an additional performance hit from increasing the degree of aniso.

With 4xs FSAA, there is a significantly larger hit with anisotropic filtering, as the half-supersampling puts a significant hit on fillrate.

Note: all benchmarks were done with max quality settings in Quake3, and no other driver options tweaked except anisotropy and AA.

Update: Oh, yeah, almost forgot. All benchmarks done with 29.42 drivers.
 
Driver 28.32
1024x768x32, 4x FSAA:
2-degree aniso: 101 fps
8-degree aniso: 79.6 fps

1280x1024x32, 4x FSAA:
2-degree aniso: 58.8 fps
8-degree aniso: 47.9 fps

Driver 29.42
1024x768x32, 4x FSAA:
2-degree aniso: 106.6fps
8-degree aniso: 95.1 fps

1280x1024x32, 4x FSAA:
2-degree aniso: 61.1 fps
8-degree aniso: 56 fps

Vtek 4600, P4 2.53, 512MB PC800 RDRAM.
 
I've just done a quick test on the SS:SE scene I normall use and it would appear the LOD on the newer drivers has been increased quite significantly.
 
DaveBaumann said:
I've just done a quick test on the SS:SE scene I normall use and it would appear the LOD on the newer drivers has been increased quite significantly.
Click to expand...

As increased for a sharper picture?
 
DaveBaumann said:
I've just done a quick test on the SS:SE scene I normall use and it would appear the LOD on the newer drivers has been increased quite significantly.
Click to expand...

Hmmm I'm quite sure that 28.32 didn't show any performance/quality optimisations in Rivatuner. Could you recheck your 29.42 scores, yet this time with quality vs performance aniso?
 
It just looks to me from your scores that with 28.32 on default quality mode is used for 64-tap, in the contrary to 29.42 which looks more like performance mode.

What doesn't make sense here is that looking at your screenshots it's apparent that 29.42 has offset LOD quite a bit further than 28.32. Normally more aggressive (negated) LOD settings hurt performance they don't increase them.
 
Ailuros said:
What doesn't make sense here is that looking at your screenshots it's apparent that 29.42 has offset LOD quite a bit further than 28.32. Normally more aggressive (negated) LOD settings hurt performance they don't increase them.
Click to expand...

Exactly! Which, like I said earlier, is why I'm very skeptical as to what Nvidia is doing with these drivers, especially since I've seen posts stating that it's extremely hard to notice any visual differences between 4x and 8x aniso (something I've also noticed myself). And texture aliasing has definitely been increased (very noticeable in NWN) with these latest drivers.
 
So...does nV still AF textures not on the horizontal or vertical, or have they adopted ATi's adaptive method? Who wants to check out a rotating colum or room or two? :)
 
Hrm after some fooling around with Geforce4s I've noticed that oftentimes, anisotropic filtering will be disabled when running a benchmark with AA+anisotropic filtering. Defaulting to a lower setting. Tried Quake3 last night , for example, 1280x1024 2x FSAA 32tap and 64tap scored exactly the same, using the Nvidia control panel. Using Rivatuner , however there was the expected % drop in performance going from 4x to 8x .
 
You must log in or register to reply here.
Back
Top