Iria

Source: PC Perspective
________
Today, NVIDIA is introducing the enthusiast community to Quad SLI once again. Only this time, they want you to go buy it and build yourself; no longer is Quad SLI being delegated to the boutique system builder. NVIDIA's reasoning for not wanting to put Quad SLI in the hands of enthusiasts immediately was due to the software questions and instability that existed initially while the driver software was being tweaked. As of today, NVIDIA has readied the driver for the public and we have been spending the last couple of weeks playing with it and the technology behind Quad SLI in order to share our experiences with you.

BWX

Most every Directx9 game can't even run any faster on the quad SLI than a 7900GTX SLI or GX2 single at the same IQ settings because of that buffer bug.... That really puts a hamper on things.

I'd like to see 16xAF being used too, if I have a 5000 dollar gaming rig I'm not going to be running 8xAF, especially Nvidia AF.. which is pretty weak as it is.

__________________

Zardon

yeah 8af is a weird choice indeed, I agree.

One other thing, you said you found NV AF "weak"? can you notice much difference between it and the ATI AF? I do know NV AA isn't in the same league as ATI's implementation ......

BWX

Actually in a whole bunch of racing sims that are very popular, all using a similar GFX engine, the AF and texture filtering is actually pretty bad with Nvidia. It is more noticeable because of the lines on the road and long viewing distances I think. In some games it isn't nearly as noticeable.

Games that really show how bad the AF is, even on the absolute highest settings and trilinear with all optimizations off.. (things go down hill fast when using any optimizations):

GTR, GTR2 demo, rFactor, GT Legends. ATI does a beautiful job filtering the textures in these DX9 games, but Nvidia, even at 16x AF w/ no optimizations at all leaves quite a bit to be desired. There is texture shimmering, blurred textures, and or texture aliasing that ATI cards don't have.

I'll have to dig up some screens. Other games that are more optimized for Nvidia don't seem to have this problem. Maybe it's only a DirectX thing too, I don't remember much difference in any OGL game.


EDIT
I found some screens- I didn't take these, but I can vouch for them because I see the same thing. When you have a very fast card like 7900 GTX etc.. it isn't that bad because you can crank up AF to 16x and turn all optimizations off..

But still, I would rather use ATI 8xAF with AI and optimizations than Nvidia 16xAF with all optimizations turned off in NV CP. And a lot of people are trying to make due with slower cards and just have to settle for much worse IQ and more shimmering, etc.. Only in certain games though.













The texture shimmering is much more apparent when actually playing the game though.

__________________

Zardon

quite a few rendering differences there indeed, I notice huge variances in the first set of images on the nearest tree to the left, it actually looks like the ATI image isnt rendering any shadows at all....

BWX

With that game though, the GFX are all extremely configurable, and the shadows and texture settings and all other GFX settings have like 3 or 4 different levels. Plus that top screen is not a track that came with the game, it was converted from another game, that's why the tree textures are so bad. In the bottom shot, that is an example of a stock track, very nice textures and everything.


Nvidia just has a blurrier look to everything.. Compared to my old X800XL, the 7900GTX can;t hold a candle to the sharpness of the GFX. But still, the 7900 is so much faster that I can run way higher resolution and settings in all games so overall I get way better IQ.. but not if I ran both at the same settings. Same with quad SLI kind of.. you can run higher filtering compared to the other cards.



I'll show you some shots I can get with my 7900 with those games though.. Especially when reduced in size, they still look awesome.

user made content- custom car and custom track.. not made at all by the game devs.. just modders. Same as game above - rFactor..


GTR2 demo.. has rain (rF doesn't), night to day transitions (like rF).. Going to be a cool game.


See, settings cranked with Nvidia looks nice.


but even here- in a screen
that is reduced in size, was originally higher resolution, 16xAA, you can see the line blurring. ATI would not do that.


Same here- even though it looks good, you can see the difference, the blurring.. (I don't have an ATI card to test now, but I know what it would look like)




So there is a difference.. but with some of the games he tested, 8xAF might not look much different than ATI 8xAF.. It really depends I guess.

__________________

BWX

This is interesting I found a screen I took when that track came out in almost the same place.. It's Le mans BTW. This is full sized 1600 x 1200, 16xAF - just jpeg reduced quality..



But on a 19" CRT, that amount of blurring is not that bad.


I also have a bunch of screens from COD2, Prey, HL2, Fear, etc with 8xAF and 16AF that look great.. So there ya go.

__________________

WaltC

I'm not sure I'd call that splotching a "shadow"... Take a look at the lack of detail of the tree bark in the nV pic contrasted with the ATi screen...and then follow that down to the differences in detail (and shadow?) of the metal guardrail below and in front of the tree--the ATi screen seems to show a lot more guardrail detail than the nV pic, so I'm wondering if the differences as pictured here might stem from a texture LOD setting for the nV defaults that is a good deal lower (blurrier) than the ATi defaults as captured, in addition to much less precise AF filtering being done in the nV screen shot. Of course, although you can see moire in a screen shot, shimmering is something that only appears when the game is running--shimmering won't appear in a single screen shot--has to be seen in-game to be "appreciated"...

I'd class this Quad SLI review in the "fair-average" category because of the following:

(1) The author states "Each card features two G71 cores and two separate 512 MB frame buffers, for a total of 2 GB of frame buffer memory in a quad SLI computer. Excessive much?," and a couple of times he reiterates "2 gigs of memory" in the review without precisely and accurately explaining that only 512mbs of video memory is accessible by a Quad SLI setup at any time. If he sometines is talking about videoram and sometimes is talking about system ram, he needed to be precise when doing so, so that his readers would know what he was talking about and when.

512mbs of video ram is "all" that is accessible by a 3d game, regardless of there being a "total" of 2 gigs of 3d memory inside a QuadSLI system. The total of videoram used in QuadSLI is irrelevant since no amount above 512mbs is accessible by a game. Since there's nothing insufficient about 512mbs of videoram, I don't think there's any reason for a reviewer to be shy about explaining this. Of course, "2 gigs" sounds better, but...there really aren't 2 gigs of dedicated videoram available to any of the gpus running in QuadSLI.

(2) It would have been nice for the author to have used these components inside a case as even most "enthusiasts" are unlikely to setup everything in an open-air rack as was done here. Then the reviewer could have looked at niggling little things like heat and noise, for instance, that go along with all of this. Enthusiasts often "sweat the small stuff," if you know what I mean...

(3) I know a lot of "enthusasts" who, despite being enthusiasts, do not have unlimited budgets for their enthusiasms, and so it would have been nice to see the reviewer price the entire system he used as opposed to simply the parts he assumes upgraders will have to buy and translate into their current rigs to support QuadSLI--although I do not recall him anywhere adding in the price of his "1 KW" TurboCool PSU. (Pardons if I overlooked this.)

A major component lacking in these reviews is a price/performance comparison--ie, a "substantial" 18-30% frame-rate performance increase doesn't look so "substantial" after all if it requires a 200%-400% greater expenditure of funds, does it? I think looking at price-performance ratios for both Crossfire and SLI provides crucial information for the enthusiast.

(4) I think he skirted a primary issue about SLI (or CrossFire.) Neither approach is supported directly by 3d APIs, and as such nV or ATi/AMD multi-gpu environments have to be literally hacked into the APIs/OS's they run under. He skirts the issue when he talks about some D3d limitations in regards to when and how it is possible to hack into and around the D3d API to setup QuadSLI AFR buffers, but he never directly addresses the issue that Crossfire and SLI create--that they aren't supported by the APIs, but rather by proprietary driver hacks that seek to bend the APIs into doing things they were never designed to do. I think this explains the relative poverty of the extent of his game testing. IE, I would like to see such reviews test at least HALF of the "hundreds of games" that both ATi/AMD and nV state their multi-gpu approaches "support." I mean, why just look at half-a-dozen games and let it go at that? How does that tell the whole story?

(5) Image Quality, that old, dependable standby. IQ is certainly something that enthusiasts look at when choosing 3d solutions. I have to agree with BWX here. A review which essentially produces bar charts showing frame-rates for a few high resolutions doesn't provide much in the way of IQ information--at least, not on the order of screen shots of the kind that BWX has placed here. Such screenshots demonstrably prove that IQ is certainly far from being "subjective" as the differences shown in even BWX's example screenshots are very objective in the kinds of IQ differences they can illuminate. It's always been my opinion that without objective demonstrations of IQ inside a review, that all frame-rate bar charts then become completely subjective themselves, because you don't know what kind of IQ levels you'll get when you get the frame-rates the bar charts illustrate. Talking loosely about perceived IQ is just not the same as producing informative screenshots which illuminate your opinions.

Frankly, in summary, I like the "single-card" dual-gpu approach nV has created, and I would choose this solution any day over two separate cards running in two separate PCIe slots--regardless of the number of total gpus that the IHV attempts to leverage. But I am still not sold on the multi-gpu approach to the degree that I am tempted to buy it, whether it's made by ATi/AMD or nV. Also, I think the single-card, multi-gpu approaches have the best chance of receiving direct support from upcoming versions of D3d, if indeed any multi-gpu approach will receive such API support in the future.

BWX

Check out the pic in post 7... and read the post too.. it explains it. That's just the way NV renders it..

Yes that is a lower quality textured track.. but it happens in higher quality, higher resolution tracks too. That track I think came from a game called F1C, I can't be sure though. A newer version came out and everyone uses that one now.

But what you are seeing is just the way it's rendered.. Yeah it does look blurry, but that's because in that GFX engine, that is the way NV does it.

Also that shimmering can be seen in screenshots. I have a lot of experience with this game. If you are trying to say those pics are faked or doctored, you would be wrong. ATI just looks that much different. They are both taken with the same in game settings. (highest possible setting in post 7 though)

That shimmering is worse when optimizations are enabled, but it still present when all optimizations are disabled. What you see there is what you get. The track with the shimmering is a high quality track made by the devs who made the game too, not like the first one, and you can also see the lines look a little better and less blurry. It's called Toban.

http://www.rfactor.net/index.php

__________________

dipstick

Having used an X1800/x1900 for almost a year, going back to nv I noticed the better AA more so than the worse AF. But I will be honest that unless you put things under a microscope you will not notice any IQ differences at all. Shimmering is not even close to what it was back in the 7800GTX days and Quad is simply amazing when it works. Playing FEAR with a minimum of 60+fps at max everything is a great experience.

BWX

That Texture clamp setting set to "on" is needed for me to get rid of the shimmering in most games.. But it works.
(not lack of AA shimmering, but what looks like the entire screen shimmering) I never noticed that with ATI.. the last one I had was an X800XL..

__________________