Can anyone come to a conclusion about these questions?

[K.I.L.E.R]

If it isn't a problem, I would love a straight to the point short answer to these questions.

1. Has AF been broken for every nV card above the NV20? Is yes, is it broken in hardware or software?

2. Is the NV3x's hardware broken?

So far I keep getting mixed responses.

Thank You

 

[RoOoBo]

2. Is the NV3x's hardware broken?

Most complex hardware designs are 'broken'. That includes CPUs (P4 or Athlon) or GPUs. Just check Intel or AMD reference manuals (for GPUs is harder to know because the problems can be somewhat hidden by the API/driver). The important is how much and which part of the hardware is not working.

I don't know about NV3x and AF but it seems fog was not working. And ATI R300 seems to have a problem with hardware TruForm.

 

[Dave Baumann]

Well, clearly NVIDIA's 'Application' AF has had a speed up in comparison to GF4.

http://www.beyond3d.com/previews/nvidia/nv31/index.php?p=15

 

[K.I.L.E.R]

So from what I gather, NV30 is broken. (even if it is just 1 little thing)

Dave, it looks like the NV30 is running out of fillrate easily with AF and high res or it's just broken.

nVidia themselves have said that they don't apply 100% AF on the entire scene.
Neither does Ati.

R300 takes a jack all performance hit compared to the NV30.

The question still comes down to: Is the NV30's AF borked or just a sfotware thing?

If it isn't borked then it is just the inefficient algorithm.

That is what I am after.
Borked or just Inefficient?
Thanks

 

[BoardBonobo]

If you are using Aniso on the NV25 I think you lose the equivalent of 1 TMU per pipe. So you effectively half your fill rate in mult-texturing scenarios, I think. Plus the fact tha, IMO, their algorithm is too greedy and not very elegant.

Oh, and I'm fairly sure the Nv30\31 is borked in hardware. Something they thought could be worked around with software, but ultimately is either going to be too time consuming or make as mess of their UDA.

I think we'll see some very large driver files wen the NV35 arrives, as they separate the NV2x from the NV3x. Also I don't think NV30\31 performance + IQ will improve hand in hand, even if the NV35\36 does.

 

[K.I.L.E.R]

Thanks.

Borked+inefficient hardware.

 

[Grall]

nVidia themselves have said that they don't apply 100% AF on the entire scene.
Neither does Ati.

That's the whole point of anisotropic filtering buddy. It's SUPPOSED to be adaptive.

You only use more samples where you need it, that is what makes aniso generally better than trilinear filtering (which is an isotropic filtering method). The actual method employed to accomplish this isn't standardized like with bi/trilinear filtering, so actual results may vary. Some algorithms will look better and/or perform faster than others.


*G*

 

[BoardBonobo]

Thanks.

Borked+inefficient hardware.

I didn't say the hardware was inefficient, just borked

 

[K.I.L.E.R]

Thanks.

Borked+inefficient hardware.

I didn't say the hardware was inefficient, just borked

IMO, their algorithm is too greedy and not very elegant.

Seems to me you were saying both.

A 5-10 word answer would have been enough.

Thanks

 

[Ante P]

And ATI R300 seems to have a problem with hardware TruForm.

They simply don't have full dedicated hardware acceleration anymore.

 

[K.I.L.E.R]

And ATI R300 seems to have a problem with hardware TruForm.

They simply don't have full dedicated hardware acceleration anymore.

I beg you guys not to turn this into an Ati thread. You want me to start another thread about Truform? Last I heard was that it was a vertex shader thing anyway.

 

[MuFu]

2. Is the NV3x's hardware broken?

Most complex hardware designs are 'broken'.

I think that's the key point. Most incorporate redundant elements and/or have higher-level workarounds for fubar functionality.

You get the impression that NV30 was released with more than an "acceptable" level of known issues/escapes.

MuFu.

 

[Xmas]

1. Has AF been broken for every nV card above the NV20? Is yes, is it broken in hardware or software?

It works, doesn't it?
So it's not broken...

NV25 has the issue that with AF enabled, both TMUs per pipe work together as one trilinear TMU. So bilinear AF is only half as fast as you would expect. No difference for trilinear AF, though. That is fixed with NV30 AFAIK.

I wouldn't say the hardware is inefficient, but the algorithm is inefficient.
It does a good job quality-wise and has few "bad cases". It's pretty straight-forward and obvious. But there are other algorithms that do better with less effort (No, I don't mean ATI's implementation).

 

[tobbe]

(Truform) Last I heard was that it was a vertex shader thing anyway.

Not wanting to go OT, but todays vertex shaders are still 1 vertex in, 1 vertex out no?

 

[Ostsol]

(Truform) Last I heard was that it was a vertex shader thing anyway.

Not wanting to go OT, but todays vertex shaders are still 1 vertex in, 1 vertex out no?

Yep. There's no creation of vertices in the vertex shader.

 

[K.I.L.E.R]

I just said it was a vertex shader thing. I didn't mean what you guys are implying I ment. I don't know exactly what's wrong but I did hear that there was something about the VS unit that was causing crappy Truform performance.

Of course, newer drivers there are games that take NO performance hit on the R300 using Truform. IE: NWN.
RTCW which takes a slight hit. Only Truform games I tried. So the problem is either fixed or they were able to walk around it.