Go Back   nV News Forums > Graphics Card Forums > Other Desktop Graphics Cards

Newegg Daily Deals

Reply
 
Thread Tools
Old 10-15-02, 10:27 PM   #37
Chalnoth
Registered User
 
Join Date: Jul 2002
Posts: 1,293
Default

Quote:
Originally posted by Joe DeFuria
You're just fogging the issue. Sure...with high enough transistor budget (additional logic) and bandidth, etc, you may 'negate' a performance penalty.
Here's the way I think of it (as it applies to the GeForce4 vs. Radeon 9700). The GeForce4 can compute one degree of anisotropic per pixel pipeline per clock. The Radeon 9700 can do the same. Therefore, there is no reason that the Radeon 9700 cannot perform a similar calculation.

And, as I attempted to explain previously, the Radeon 9700's aniso degree selection appears to require a very similar transistor count to what would be required for doing the accurate calculation. I'm really beginning to think that the decision actually had very little to do with transistor count budgets, but was more an engineering time constraint, or something along those lines.

As another way to state it, since the original GeForce, nVidia's hardware has been able to do what looks like the accurate aniso degree calculation for one texture per pixel pipeline per clock, I see absolutely no reason why the Radeon 9700 cannot do the same for its one texture per pixel pipeline per clock.

And, stated again, the NV30 should therefore, with its eight pipelines, be able to put out very similar performance figures with anisotropic enabled (Actually, it should be able to beat the 9700 without too much trouble...given the almost certainly higher core speed).

Quote:
Doesn't the GeForce3 have that ability though?
I don't think so. I would assume that the GeForce3's anisotropic is identical to the GeForce4's. I don't currently have a GeForce3 to test.
Chalnoth is offline   Reply With Quote
Old 10-15-02, 10:40 PM   #38
Chalnoth
Registered User
 
Join Date: Jul 2002
Posts: 1,293
Default

Oh, and one other thing.

Might I remind you of the procedure of going through scientific experiements? The idea is to vary just one variable in order to test that single variable's influence on the outcome.

When you compare the GeForce4's anisotropic to the Radeon 9700's, you are changing far more than one variable, and therefore it is impossible to accurately tell which of the numerous changes is the one that is the cause of the performance discrepancy.

In other words, just because the GeForce4 has more of a performance hit than the Radeon 9700 with anisotropic enabled doesn't mean that you can't have an anisotropic implementation that is invariant upon rotation about the z-axis with a performance hit similar to the Radeon 9700's. You don't have enough information to make that statement, and looking at the technical aspects of the situation contradict that statement.
Chalnoth is offline   Reply With Quote
Old 10-15-02, 11:10 PM   #39
Moose
Cheating is for losers!!!
 
Join Date: Jul 2002
Posts: 241
Default

Well. I must say that I was a big skeptic of the R9700. I have the Ti4600 and I now have the R9700.

There is simply no comparison at ALL IQ wise. The R9700 wins hands down.

You can argue all you want about what looks good at what angle but the fact is that at 6xFSAA and 16x aniso the R9700 IQ is vastly superior than anything the TI4600 can do. Period.

Add to that that most games are actually playable at those settings on the R9700 and its a done deal.

Believe me I'd love to say it isn't true, I'm as big of a fan of Nvidia as the next guy but the R9700 is just so much better it isn't even a contest.

We may have a different story when Nv30 arrives but for now the R9700 is the best.
Moose is offline   Reply With Quote
Old 10-15-02, 11:14 PM   #40
Joe DeFuria
Registered User
 
Join Date: Oct 2002
Posts: 236
Default

Quote:
In other words, just because the GeForce4 has more of a performance hit than the Radeon 9700 with anisotropic enabled doesn't mean that you can't have an anisotropic implementation that is invariant upon rotation about the z-axis with a performance hit similar to the Radeon 9700's.
Right.

Which also implies : it doesn't mean you can't have an anisotropic implementation that maintains the 9700 performance hit that is invariant upon rotation about the z-axis.

Sound simliar?

It should.

The point is, you speak about "fundamentals" such that nVidia isn't "fundamentally" slower. Yet using your logic, I can turn it around and say that ATI isn't "fundamentally non-uniform in quality."

In other words...I think your point concerning "fundamentals" is, well, pointless.
Joe DeFuria is offline   Reply With Quote
Old 10-16-02, 12:04 AM   #41
Chalnoth
Registered User
 
Join Date: Jul 2002
Posts: 1,293
Default

Quote:
Originally posted by Joe DeFuria
The point is, you speak about "fundamentals" such that nVidia isn't "fundamentally" slower. Yet using your logic, I can turn it around and say that ATI isn't "fundamentally non-uniform in quality."

In other words...I think your point concerning "fundamentals" is, well, pointless.
I hope I didn't explicitly say that nVidia's technique isn't fundamentally slower. What I'm trying to say is that a technique that is invariant in rotations about the z-axis doesn't need to be noticeably slower than a technique like that seen in the Radeon 8500 or Radeon 9700.

This is my primary concern with people who say, "but it's faster," when I say that it doesn't look as good. With the Radeon 9700 vs. GF4, which one is faster is obvious (Heck, it does have double the pixel pipes and double the memory bandwidth afterall, it sure had better be faster...).

I'm also trying to say that nVidia doesn't need to change their anisotropic implementation at all in moving from the GF4 to the NV30 in order for it to outperform the Radeon 9700's anisotropic. As long as everything that affects performance besides anisotropic is on par (as it most certainly should be), and it runs at a higher clock speed (again, as it should be, given the .13 micron process), then the NV30 should most certainly outperform the Radeon 9700's anisotropic.
Chalnoth is offline   Reply With Quote
Old 10-16-02, 12:44 AM   #42
Bigus Dickus
GF7 FX Ti 12800 SE Ultra
 
Join Date: Jul 2002
Posts: 651
Default

Quote:
Originally posted by Chalnoth
What I'm trying to say is that a technique that is invariant in rotations about the z-axis doesn't need to be noticeably slower than a technique like that seen in the Radeon 8500 or Radeon 9700.
I think I should point out that, as you stated previously about some similar statement, you probably don't have all the necessary information to make such a definitive statement. That might be the case, but just because you "can't see any reason" doesn't mean that a reason doesn't exist.

NV has had this capability since the GeForce. Do you really think that time constraints were the primary issue? For how many years? Sure, it looks obvious to you that after so many years they should have been able to "figure it out" or "do it right," but in all likelihood there was a reason years ago not to do it that way, and that fundamental reason hasn't changed, hence their implementation hasn't fundamentally changed.

In other words, IMO, it's much more likely that there is some substantial trade-off between the two methods such that they made an informed decision to go the route they did. I doubt this trade-off was due to the time required to do it (hundreds of hardware designers over several years, and they just couldn't squeeze it in?), or due to their lack of understanding of 'how' to do it. That leaves (1) transistor count, (2) performance, (3) some other unknown variable.

I can't imagine what that other variable might have been, but that doesn't negate the possibility that I (and you in your infinite wisdom as well) simply are not aware of it. Or is could be (1) or (2)...
__________________
IMO, Mr. Derek Smart is a hypocrite: Only someone who is either (a) lying (b) ashamed of their products (c) just plain ashamed, would hestitate to give out some simple and straight forward information. - Derek Smart, Ph.D.
Bigus Dickus is offline   Reply With Quote
Old 10-16-02, 01:20 AM   #43
Chalnoth
Registered User
 
Join Date: Jul 2002
Posts: 1,293
Default

Quote:
Originally posted by Bigus Dickus
NV has had this capability since the GeForce. Do you really think that time constraints were the primary issue? For how many years? Sure, it looks obvious to you that after so many years they should have been able to "figure it out" or "do it right," but in all likelihood there was a reason years ago not to do it that way, and that fundamental reason hasn't changed, hence their implementation hasn't fundamentally changed.
Figuring it out should not be a concern in the least. The math already exists, and is very, very simple. The issue that I see is with priorities. I feel that ATI just has not prioritized this particular issue. The Radeon 8500->9700 move appears to be a natural extension of the move from two texture units per pixel pipeline to one.

The way I see it, these many engineers have been very busy for a very long time on the Radeon 9700's hardware. With such exciting developments as the increased programmability and increased precision, it seems very easy to overlook such "little" things as MIP map and aniso degree selection algorithms.

Of the possible implementation issues, performance is certainly not a significant one. If it were, people with Radeons would notice performance changes in situations such as the rotating room in Serious Sam, or in flight sims.

With performance obviously thrown out, that just leaves engineering-side implementation issues, such as alotted development time, transistor budgets (which I no longer think was a major reason), transistor layout, and so on.

For example, on the transistor layout part, it might be much easier to copy as much of the pipeline from the Radeon 8500 as possible, so that the transistor layout needs as little redesign as possible.

Regardless of the reasons it wasn't done, I'm very certain that performance is not one of them, and nothing can change the fact that ATI should fix the problem once and for all with their next major architectural change (I don't expect any improvement in the Spring...that would be far too much to ask for).

Update: By "should" I don't mean I think they will, I think that it's something they should do to improve the visual quality of their products (and I don't mean to imply that I think they won't, either...).

Last edited by Chalnoth; 10-16-02 at 01:26 AM.
Chalnoth is offline   Reply With Quote
Old 10-16-02, 05:23 AM   #44
StealthHawk
Guest
 
Posts: n/a
Default

Quote:
Originally posted by Chalnoth
I don't think so. I would assume that the GeForce3's anisotropic is identical to the GeForce4's. I don't currently have a GeForce3 to test.
then why did the GF4 have a greater performance hit for enabling AF than the GF3? comparing the GF3/4 AF scores, you would have expected the GF4 to at least outperform the GF3 by the clockspeed increase %.

additionally, AFAIK AF in D3D is still slow as mud with a GF4. the performance hit with OGL is considerably less than it was with the card's introduction. also, last i checked the tweaks for AF in D3D for disabling texture stages only work with the GF4 and not the GF3. not exactly conclusive proof, but i find that fact interesting.
  Reply With Quote

Old 10-16-02, 10:23 AM   #45
jbirney
Registered User
 
jbirney's Avatar
 
Join Date: Jul 2002
Posts: 1,430
Default

Quote:
There is no evidence to support that ATI's aniso method is fundamentally of higher performance.
ATIs method is the same on the 8500 as it is on the R9700 just with Tri-linear and an angle tweak. Its been shown on a GF4 vrs 8500 comparison that once you hack down the GF4 to also only do Bi/AF (so you level the playing field so to speak) that the 8500's AF had a lower performance hit.


Quote:
Regardless of the reasons it wasn't done, I'm very certain that performance is not one of them, and nothing can change the fact that ATI should fix the problem once and for all with their next major architectural change (I don't expect any improvement in the Spring...that would be far too much to ask for).
Once again there has been public announcements made by ATI PR people, ATI Engineers and ATI Software (ie drivers) writters that they intended to design their AF this way as the speed difference was worth the "problems" in ATIs mind. I understand that you have a technically correct reason. But with all the other information your choice although valid is not the reason they went that way.
jbirney is offline   Reply With Quote
Old 10-16-02, 11:05 AM   #46
Chalnoth
Registered User
 
Join Date: Jul 2002
Posts: 1,293
Default

Quote:
Originally posted by jbirney
ATIs method is the same on the 8500 as it is on the R9700 just with Tri-linear and an angle tweak. Its been shown on a GF4 vrs 8500 comparison that once you hack down the GF4 to also only do Bi/AF (so you level the playing field so to speak) that the 8500's AF had a lower performance hit.
I've already explained the primary reason that the GeForce4 has so much of a performance hit. If the first texture has anisotropic enabled, the second texture does not go through. That is, disabling anisotropic for the first and third texture stages results in a huge performance increase. While using bilinear filtering does allow for some improvement in performance, it doesn't come close to the improvement from doing what I described above.

And I have yet to see any ATI representative give a satisfactory answer. Any links?
Chalnoth is offline   Reply With Quote
Old 10-16-02, 02:08 PM   #47
jbirney
Registered User
 
jbirney's Avatar
 
Join Date: Jul 2002
Posts: 1,430
Default

Links? Not off the top of my head but I remember reading these and will look to find them ASAP.

BTW Does the GF3 suffer from this same issue with AF? The 8500 and GF3 Ti500 are close in specs yet the 8500 is way ahead in AF perfromance.

Last edited by jbirney; 10-16-02 at 02:12 PM.
jbirney is offline   Reply With Quote
Old 10-16-02, 02:23 PM   #48
jbirney
Registered User
 
jbirney's Avatar
 
Join Date: Jul 2002
Posts: 1,430
Default

Ok here is one from a user named Dio who just happens to write drivers for the R9700:

http://www.beyond3d.com/forum/viewto...ighlight=aniso

Notice the part where he says A lot is due to the adaptive algorithm which sounds a lot like a speed optimization to me.


Next taken from the Extremtech AF article:

Quote:
ATI's adaptive implementation of AF is a trade-off between not applying AF where it isn't needed (in ATI's opinion) and using dynamically more filter taps on affected areas, with the most severely affected areas getting the most taps. It's a kind of budgeting of AF that ATI believes delivers the best net effect.

ATI readily admits that its technique selectively applies AF, and makes no secret about not applying AF to the entire 3D scene – unless every surface is sloped relative to the view camera, an almost unheard of case. ATI engineers knowingly implemented AF in a sort of "triage" fashion, where the most severely affected areas get the most filtering, and unaffected areas get none (other than bilinear or trilinear).
http://www.extremetech.com/article2/...,559037,00.asp

However I dont have time to find the ATI engineer nor the PR speak that also says the same thing that they wanted to create a "performance mode of AF". When I have more time tonight I can try to find those if you still need to see em. But I think I have proved my point that the R9700 AF was designed to run this way for speed. Not due strictly to a budget of transistors. And thus is not "broken" in ATI's eyes.
jbirney is offline   Reply With Quote
Reply


Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Forum Jump


Similar Threads
Thread Thread Starter Forum Replies Last Post
R9700 Oficially Announced druga runda Other Desktop Graphics Cards 20 04-25-08 11:28 PM
My UT2003 Tweak Guide DXnfiniteFX Gaming Central 48 10-31-02 12:59 AM
Understanding CineFX - MUCH more than the R300 Uttar Rumor Mill 68 10-02-02 02:02 AM
FalconNW and Voodoo under ATI 9700 spell!!! mizzer Other Desktop Graphics Cards 12 09-20-02 08:53 PM
GeForce4 image quality - need some HONEST opinions ErrorS NVIDIA GeForce 7, 8, And 9 Series 24 08-22-02 07:39 AM

All times are GMT -5. The time now is 04:00 PM.


Powered by vBulletin® Version 3.7.1
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Copyright ©1998 - 2014, nV News.