The op says:
Where the review he takes his numbers for the comparison says:
Text bolded by me to bring the point out.
So even though nvidia had the "higher AA number" the quailty is lower because of the algorithms used. So if you "give" bf2 back to the one with the higher image quality and higher FPS, then you have ATI winning 4 and nividia winning 3. <shrug>
Where the review he takes his numbers for the comparison says:
Text bolded by me to bring the point out.
So even though nvidia had the "higher AA number" the quailty is lower because of the algorithms used. So if you "give" bf2 back to the one with the higher image quality and higher FPS, then you have ATI winning 4 and nividia winning 3. <shrug>
Cuular, you just quoted what I was about to quote. The review specifically states that the "Maximum Quality" tests are apples-to-oranges in some regards, since they're just measuring the highest-quality settings you can pretty much run. It is a more realistic test in some respects, but also you can't really compare the framerate results as if they were teh same settings. And there's still more to it.
Crossfire at maximum quality is clearly an anomaly. Look at BF2 for example - running Crossfire with the maximum quality gets you lower performance than running a single card. However, that is not the case with the "lower quality" (like 4xaa/8xAF) settings, where Crossfire gives you 50% performance increase over a single XT. This is definitely a problem with ingame code or a driver issue, either of which can be fixed.
Crossfire at maximum quality is clearly an anomaly. Look at BF2 for example - running Crossfire with the maximum quality gets you lower performance than running a single card. However, that is not the case with the "lower quality" (like 4xaa/8xAF) settings, where Crossfire gives you 50% performance increase over a single XT. This is definitely a problem with ingame code or a driver issue, either of which can be fixed.
You're absolutely right, Cuular, I apologize for that oversight.
For my own education, will you (or someone out there) please explain to me why ATI's 6x MSAA is better than nVidia's 8xS AA (or at least point me to where I can read about it)? Also, looking at the Black and White 2 images, there is a side-by-side comparison between ATI's and nVidia's "Maximum AA". Is this the "apples to oranges" scenario that the reviewer is referring to? I'm assuming these Maximum AA parameters are not one in the same as those used in the Quake 4 image comparisons, because nVidia's 8X AA is clearly of higher quality than ATI's 6X AA with respect to the screenshots posted in the review.
Difference in algorithm. The algorithm uses a different 'AA grid' to do the antialiasing perhaps.
So ATI doesn't do AA the same way nVidia does? Which one is more computationally intensive, and which one produces the best image quality? Does it depend on the game?
Also, xtknight: your post indicates that the question about AA algorithms was asked by Cuular, but it was me (dredd2929) that asked it. Not that big a deal to me but Cuular may not appreciate being quoted on something he didn't say.
Originally posted by: dredd2929
So ATI doesn't do AA the same way nVidia does? Which one is more computationally intensive, and which one produces the best image quality? Does it depend on the game?
Also, xtknight: your post indicates that the question about AA algorithms was asked by Cuular, but it was me (dredd2929) that asked it. Not that big a deal to me but Cuular may not appreciate being quoted on something he didn't say.
ATI's AA is generally better at the 'same' level, but NVIDIA's 8xS (which combines SSAA and MSAA) should be better than ATI's 6xAA (which is just MSAA with six samples per pixel). However, 8xS mode is generally VERY slow.
More detail:
ATI uses 'jittered-grid' antialiasing for its MSAA, which means the samples are offset slightly. This makes it *generally* look better on things that are tilted at a small angle. NVIDIA uses ordered-grid antialiasing, which can sometimes miss antialiasing objects that are slightly tilted, but does about the same on edges at a sharper angle. ATI's MSAA is generally equal, and sometimes does a significantly better job with the 'same' number of samples (ie, 4xAA versus 4xAA).
They're about equal in computational resources, but ATI has more flexbile AA capabilities (things like 6xMSAA, Temporal AA, and the SuperAA modes using Crossfire) -- but they either cannot or will not enable SSAA. NVIDIA cards can use SSAA modes for higher quality (though with an extremely large performance hit, to the point where it is generally unusable except on very old games).
In terms of comparing the numbers in the review -- AT said (if you read the beginning part of the review carefully, which apparently many people did NOT) that the 'Maximum Quality' numbers for NVIDIA used 4xAA on some games and 8xSAA on others. Since they don't indicate where each one is in use, you can't really compare between NVIDIA and ATI in those charts.
Well both SLI and Crossfire are novelty technologies at this time. It would take DDR2-1000 to supply enough bandwith for 2 GPUs, over 2xPCIe X16 whether they hop over HyperTransport on AMD or Hit the Northbridge then Memory on Intel. Problem is AMD doesn't use DDR2-1000 yet and will not likely support this for atleat a year and Intel might support DDR2-1000 before it kills off current P4 but then you have the problem of Intel just sucking as a CPU altogether for games and workstation graphics.
So the X1900XT is right now the most cost effective choice for performance unless you want to be the idiot who bought into marketing hype and got 2 cards on a tired, saturated data transit system.
Huh? Why would the video cards be talking to main memory? Any data they have to send back and forth is either done through the bridge/dongle, or sent directly over the PCIe bus.
Doesnt anybody care about picture quality.. Who cares if FPS if its over 40. I think the 1900XTX will have the best quality for a single card and for what I have heard from G71 they will still not be able to do AA and HDR at the same time, which I think is really strange.
Navida needs to address this.
Aother factor is the XBOX factor game companies will do games optimized for XBox. And Xbox also made by ATI and has 48 shaders..
So the conclusion of this is that for me ATI is the best at this time and for atleast a couple of months. I still hope Nvidia will have a better card in G71 as it will force ATIs hands to get the next card out and so on.. Got to love free enterprise.. A looking at the CRYTEK new engine we are going to need it..
Lets start saving.......
Navida needs to address this.
Aother factor is the XBOX factor game companies will do games optimized for XBox. And Xbox also made by ATI and has 48 shaders..
So the conclusion of this is that for me ATI is the best at this time and for atleast a couple of months. I still hope Nvidia will have a better card in G71 as it will force ATIs hands to get the next card out and so on.. Got to love free enterprise.. A looking at the CRYTEK new engine we are going to need it..
Lets start saving.......
there has to be time for better drivers and games that make use of the shaders better (think of it as dual core)
This thing is SLI slapped onto a single board. It's generally not considered when people talk about who has the fastest single video card. If it makes you feel better, we can say that the X1900XTX takes the crown for the fastest single GPU video card.
flexy
Diamond Member
- Sep 28, 2001
- 8,464
- 155
- 106
and X1900XT STILL has the performance crown. What is your point ? Want to "surprise" us with the fact that 7800GTX in SLI is often faster ? DUH
And.."NO" the x1900XT is not "faster on paper"...is it SO HARD to swallow the fact that right now the ATI X1900XT is the f@cken fastest card around ?
YES ? Is it ? Is it ?
Jeeezzzuss...
flexy
Diamond Member
- Sep 28, 2001
- 8,464
- 155
- 106
exactly. which is the X1900XT.
If i want to compare against SLI..which is frickin TWO cards..then please do me a favor and ALSO include X1900XT(X) in CROSSFIRE
Your statement 'nvidia is faster'..you're tlking about Nvidia in SLI ?
The problem then is the tester...but they compared SLI to the SINGLE 1900 just to have an idea...and in ONE or so games the card is actually even faster compared to NV in SLI (FEAR)...OTHERWISE i think a single card comparison against SLI is dumb. ANd as said..you can ALWAYS get TWO 1900 and run them in crossfire.
ANd yes, again, X1900XT(X) is *still*the fastest card and ATI does not need an asterisk or sidenotes ort is cheating or whatsoever.
It doesn't depend on the game so much as it depends if you do multisampling or supersampling. I believe 8xS is actually 4xMSAA and 2xSSAA. I'm not sure about ATI's 6xAA. I'm not even sure how you can supersample something and end up with a multiple of something other than a power of two (6?) They must have to make some sacrifices in the blending algorithms.
For alpha textures, supersampling is the only option. Multisampling cannot be done because it's processing a 'raster' image, not a 'vertex' one, a set of vertices and points. Multisampling is blending the edges of a line so that it matches with the background better. Supersampling is rendering at double the resolution, and scaling it back down by blending pixels. This generates a smoother image. This also requires more memory. 2xSSAA requires double the area so in reality it needs 4x (2²) as much memory to do as no AA! I'm not certain about MSAA but I don't think it's a memory hog as much a GPU one. I'm not completely sure how it works but if you Google Beyond3D and antialiasing maybe you can dig up some really good articles on it.
In any case this looks like a good read: http://www.nvnews.net/vbulletin/showthread.php?s=&threadid=15821
Also, xtknight: your post indicates that the question about AA algorithms was asked by Cuular, but it was me (dredd2929) that asked it. Not that big a deal to me but Cuular may not appreciate being quoted on something he didn't say.
Oops. Fixed that.
I thought that was only for the Quadros (for default anyway)? I can go in nHancer and there is a specially labeled rotated grid AA mode while the others don't say rotated grid. Then again, they don't say ordered grid either so I don't know. The fact that just a couple modes were labeled rotated grid made me think it was some kind of exclusive feature.
What I wonder is how the driver determines that the texture has alpha components in it? Does it spend cycles determining that or is it flagged somehow?
flexy
Diamond Member
- Sep 28, 2001
- 8,464
- 155
- 106
i think you're REALLY just trolling.
Just another post where i have to say, "hey dude this is a HARDWARE forum and THIS is the video section..and YES i THINK that quite a bunch people here KNOW about the upcoming NV cards, especially G71...and YES, it WILL be a fast card"/ What is your point, AGAIN ?
The X1900XT does NOT get slower because NV will come out with G71 soon.
And, no, why on earth should someone think NV sat on their a$$ for 7 months..who said that ? I personally THINK that g71 and X1900XT will be pretty on PAR actually.
Actually yes, it is. It's also faster in real world benchmarks too.
No. I don't care either way. I don't own stock in either company (ATI or nVidia).
I did...that's the whole reason for my original post. I was surprised that nVidia's GTX 512 MB SLI rig still won in a few benchmarks.
Only in most cases though, which is what puzzled me. With the specs of the X1900 series, I was expecting a landslide victory, which isn't the case. Also the asterisk comment was directed toward whoever titled the AT review article, not ATI. And, if you had actually read the entire thread, you would have learned that I was corrected on this by a few of the responses, to which I readily admitted.
That's fine that you think that. You're wrong.
It's that, like many other people, I'm underwhelmed by ATI's new GPUs and I was hoping they'd be faster. But again, as others have stated, ATI's drivers for the X1900 are still in beta, so performance won't go anywhere but up.
I hope you're right.
you also have to take into consideration that nvidia had many months to optomize their drivers, where on the other hand, ati just released a new video card and the drivers are still pretty 'raw' i guess you could say.
You sure about that? I was pretty sure that the GF5/6 still used ordered-grid AA, at least. If they changed it in the GF7 they sure haven't talked much about it.
Well, okay, there is that (on R5XX), but you still can't do full-scene SSAA like the NVIDIA boards can. I don't think it's that big a deal (since it's not usable on most newer games), but it's a little odd.
ATI's 6xAA is MSAA with six samples per pixel, no supersampling (unless you have Transparency AA enabled, and then only on alpha textures). NVIDIA's "8xS" mode is, indeed, 4xMSAA with 2xSSAA applied on top of it.
The numbers don't match up at all once you start comparing the SLI-AA and SuperAA modes. NVIDIA and ATI use different "multipliers" to describe the same combinations of SSAA and MSAA.
I believe the application specifies whether it is RGB or RGBA. There are a number of texture formats you can use, and only some of them have alpha channels.
Indeed they dont. Looking further at the b3d reviews of x1800 and x1900, you can see that Crossfire has these SuperAA modes: 8x, 10x, 12x, 14x. Among those, only 10x and 14x have 2x SSAA, the others just use more samples for MSAA. So, while we have seen multiple reviews show SuperAA 8x perform much higher than SLIAA 8x, it can be partially attributed to SLIAA doing some form of SSAA, while Crossfire is not. In 14x and 16x modes, however both solutions are doing SSAA to an extent, but I'm not sure if SLI is doing 4x or 2x SSAA. If it was doing 4x SSAA, that again would be part of the reason why SLI in 16x falls so far behind Crossfire's 14x.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter