When enough is enough (AF quality on g70)

anybody tried/has the "shimmering fix" drivers?
http://forums.anandtech.com/messageview.aspx?catid=31&threadid=1676468

 

78.03 was sent out to a bunch of people yesterday. Definitely makes things much better in HQ mode, at first glance, with pretty much bugger all performance difference.

 

That's good news and some fast action from nvidia.

 

Where is the 7800 series IQ-wise with these new drivers vs. 6800 series? (in high quality, that is)

 

It's pretty close. I don't want to stick my neck out and say the filtering is identical since I can't be 100% sure (yet, since I'm not sure of the new filter ratios when sampling across MIP boundaries), but IQ is definitely improved to the point where they could likely drop it on nZone without issues with possible extra tweakage for a WHQL/official release.

 

Flashing textures?

 

That'll be an app-specific bug for them to fix whereas this issue is a global fixup for everything.

 

Sounds good, thanks!

I hope someone takes the time to look into the performance differences over a number of apps once a final driver is released.

 

Since I'm not from England, am I right in assuming that means pretty much no difference?

 

yes

 

I have these drivers too and they are marked to improve high quality. Like Ryss I find it pretty comparable to the Nv40 cards in High Quality mode. I wouldnt go far as to call them exact. But the difference is night and day to those sensitive to shimmering as is.

 

I'm going to be pretty disappointed if all nVidia does is improve the HQ mode to NV40 levels. The quality mode is a major problem, in some games (ie. just try UT2004 on ONS-Primeval there is a massive 30%+ performance hit going from quality to high quality) because the aniso filter optimization cannot be enabled in High Quality. To this day my profile on my 6800 GT was only 2X AF HQ in UT2004, and only 6X AF on my 7800 GTX, way too low for a 2 year old game on the supposed latest and greatest graphics card.

 

I don't buy a 30% hit with AF enabled on 7800GTX.

When I enable HQ in a special texture fetch bound level in UT, I get a 11% hit and a ~5% hit with classic levels. (whatever the driver is)

 

Try ONS-Primeval, most of the other maps there isn't too much of a hit, but ONS-Primeval the performance hit is massive going to HQ. Or you can try ONS-Primeval in Quality mode with the Ansio mip optimization disabled and enabled, and see the same performance hit, the framerate hit is so massive you don't even need a timedemo (although a timedemo on Primeval will show it too), you can see it just hosting a sever with no one on it and walking around looking at fraps.

The performance on 4X AA, 16X AF HQ is so bad that the framerates are bellow the 30s almost constantly, and this is without anyone else on the map.

hmm... just tried this myself, on my 7800 GTX and the performance hit on it goign from Q to HQ isn't that big, but on the 6800 GT it is massive, maybe it is CPU limited in Umark, because the framerate difference is more noticeable walking around in fraps than in Umark for me. But on the 6800 it dips bellow the 30s, on the 7800 it stays in the 70-100 range on HQ. Nevertheless if you try walking around on ONS-Primeval using fraps, going from Q to HQ yeilds roughly a 30% performance hit, although on a 7800 the framerates stay playable unlike on a 6800.

 

Yes PM. Thats something most people dont seem aware of. The 7800GTX line loses alot less at HQ than the Geforce 6800 line. Its not really something I have been able to explain. Even with the new drivers with virtually no performance hit compared to the old drivers. The 7800 line loses alot less than the 6800 line in HQ mode.

 

Primeval is one of the absolute worst case scenarios in UT2k4. I've nothing against worst case scenarios, rather the contrary, but they should reflect in game performance more than in one spot. Primeval is not only an exeption for the entire game, but especially for all the other ONS maps.

 

What's weird is I can't reproduce the framerate hit in Umark on a 7800 GTX. In Umark the difference between Quality and High Quality is only about 1 fps. But when I walk around in fraps, there is consitently a 20-30 fps difference between quality and high quality. Maybe it's because on Primeval the view distance is huge, and when you use Umark the 3rd person perspective is facing the ground. But if you walk around on Primeval using fraps there is an undeniable framerate 20-30% hit, maybe I should try recording my own first person timedemo on Primeval.

 

Set all settings to maximum in the game with one insignificant aspect to performance to a slightly lower setting and let UMark run with the game's own ini and not it's own max quality ini. It'll change to a completely different timedemo scenario

 

It was the 3rd perspective or the realtime AI calculations in botmatches that was the problem.

Anyways here are my results using my own ONS-Primeval timedemo recorded in first person on a 7800 GTX. First person is more what you would get in gameplay so I feel this is more accurate than botmatches. Quality was tested with (trilinear optimzation off, aniso filter optimization on, aniso sample optimization off)

High Quality: 99.430864

Quality: 139.988691

what this tells me is that botmatches suck for testing AF performance, because when I used the UMark botmatches it was 87.750137 on High Quality and 89.507195 on Quality.

 

7800 has 8 more texture units than 6800...So, the benefits of any quality/performance tradeoff optimization that cuts the texture quality to decrease texture filtering cycles are higher on 6800 than 7800.

These sort of optimizations might make some sense at 6200 level...but a $500 card having to stoop to lower quality levels sucks IMHO