Doom3 benches revisited.

[Himself]

Wasn't there something about the 3.2 cat drivers not supporting 256MB or something?

 

[Ilfirin]

Ok, THAT would explain it

 

[Pete]

They might have used medium as a concession to the lower-end cards they tested (9200 and 5200). I wouldn't think of using TC with a 256MB card, though.

Russ, the 5900U is only 256MB. It was also the only 256MB card tested in Doom 3, as the Cat3.2's used limited the 256MB 9800P to only 128MB (this might have impacted memory bandwidth, too, depending on how it accesses the memory?).

 

[Snyder]

But wouldn't the 128 meg GFFX have that same problem?

(Or was the GFFX 256MB and the 9800 128MB?)

Oh, it was a 256MB GFFX. Still doesn't explain why the 256MB 9800Pro meets almost identical speed decreases as the 128MB version.

Didn't they bench D3 with the Cat 3.2 which only utilizes 128MB of the 256MB 9800Pro? That would explain it.

Edit: Damn.

 

[demalion]

Heh, Ilfirin, you might break the record for the amount of redundancy of information reduncancy repetition record that a given single post has spawned. I don't keep track, so maybe not, though.

 

[RussSchultz]

Also, texture compression may be more beneficial to the GFFX architecture a la the Kyro. ("Free" trilinear when enabled)

Or the whole 256/128 thing.

 

[OpenGL guy]

Also, texture compression may be more beneficial to the GFFX architecture a la the Kyro. ("Free" trilinear when enabled)

If that were the case, then there would be no need for the trilinear hacks introduced with the GFFX: Just force texture compression instead.

 

[YeuEmMaiMai]

I'll pass on that comment........

 

[John Reynolds]

If that were the case, then there would be no need for the trilinear hacks introduced with the GFFX: Just force texture compression instead.

What hacks? Are you referring to the new AF modes?

 

[demonic]

1334 (translated: elite)

Sorry to burst your bubble..

But elite = 1337.

 

[Nagorak]

I can't believe I didn't notice this before. Changing to high quality settings really gives a different picture of the 9800 vs. nv35 performance in doom3. Who here is actually going to play doom3 in medium quality on their 9800 or nv35?

1334 (translated: elite) multiplayer frag-fest tournament monkeys. Actually, I'm wrong, they'll use minimum quality settings in order to max out their framerates.

I can't believe that some people haven't heard yet, but Doom3 isn't a multiplayer game.

 

[YeuEmMaiMai]

It better be network capable nutting sux more than not being able to frag y our friend...

 

[OpenGL guy]

If that were the case, then there would be no need for the trilinear hacks introduced with the GFFX: Just force texture compression instead.

What hacks? Are you referring to the new AF modes?

The trilinear hacks have nothing to do with AF modes. Run a texture quality testing program with the performance slider set to different modes and watch the difference in trilinear results. This has been discussed here before.

 

[DegustatoR]

The trilinear hacks have nothing to do with AF modes.

Well, they are there to speed up the AF modes...

 

[Borsti]

(If it hasn't been yet) The ATI R9800-256 in the D3 benches used Cat 3.2s which do not recognize 256MB of ram. Any conclusions about effect of ram for 128vs256 ATI cards is invalid.

That´s not true. The Cat 3.2 detects the 256MB! We confirmed that in the ATI driver option/details panel. As you can see with the NV30 scores, 256/128 Mb does not matter much in Doom III. We also tested on a AGP4x machine so if a card runs out of memory we would have seen a huge performance drop.

One interresting thing is that JC is very unpleased that a driver can override the Aniso levels a game wants to use:

"There may be different responses to program overrides of control panel
options. Doom always sets the anisotropy level, so a well-behaved (in my
opinion) driver would respect that, and not override any settings."

....
"This is probably going to be a significant benchmarking issue. I am making
a note that I should implement a "testFiltering" and "testAntiAliasing"
option in the code for benchmarking."

About the whole driver situation: The reason for using cat3.2 instead of cat3.4 was simply because 3.4 had some troubles in Doom III.

Lars (Tom's Hardware Guide)

 

[Joe DeFuria]

That´s not true. The Cat 3.2 detects the 256MB! We confirmed that in the ATI driver option/details panel.

Just because the memory is reported, doesn't mean it's properly utilized by the hardware.

"This is probably going to be a significant benchmarking issue. I am making a note that I should implement a "testFiltering" and "testAntiAliasing" option in the code for benchmarking."

Actually, part of that job is yours, Lars. To look and compare visual quality so YOU can determine if things look (relatively speaking) the way they should. The entire burden can't be put on developers, because thy can make all the tests they want...and driver writers can find ways to work around them. The "buck" stops with you or any other reviewer.

About the whole driver situation: The reason for using cat3.2 instead of cat3.4 was simply because 3.4 had some troubles in Doom III.

Lars (Tom's Hardware Guide)

What about the anomoly that shows NV35 being FASTER with AA on, than without? How is that explained?

 

[Nagorak]

It better be network capable nutting sux more than not being able to frag y our friend...

I believe the latest word is "not at release".

One interresting thing is that JC is very unpleased that a driver can override the Aniso levels a game wants to use:

"There may be different responses to program overrides of control panel
options. Doom always sets the anisotropy level, so a well-behaved (in my
opinion) driver would respect that, and not override any settings."

Too bad for him.

 

[Ollo]

What about the anomoly that shows NV35 being FASTER with AA on, than without? How is that explained?

Yes Lars, now that we have your attention, inquiring minds want to know.

 

[Borsti]

What about the anomoly that shows NV35 being FASTER with AA on, than without? How is that explained?

Yes Lars, now that we have your attention, inquiring minds want to know.

What do you mean in detail? In quality mode or where?

In my results NV35 was not faster with FSAA. Here some results in 10x7 Medium quality

No AA: 83,0
2xAA: 70,5
4xAA: 57,1
8xA: 80,8
4xAA+8xA: 55,9

In HighQ:
No AA: 55,0
2xAA: 47,2
4xAA: 40,2
8xA: 54,5
4xAA+8xA: 39,6

Looks as expected to me.

About the ATI driver again. I wonder why the driver would detect the larger framebuffer but won´t use it...

But there´s also no big performance difference between 128/256

MedQual 16x12:
128MB: 31,1
256MB: 31,4

HighQual 16x12:
128MB: 29,5
256MB: 29,8

Lars

 

[Joe DeFuria]

In my results NV35 was not faster with FSAA. Here some results in 10x7 Medium quality....Looks as expected to me.

Did you read your whole article?

http://www6.tomshardware.com/graphic/20030512/geforce_fx_5900-12.html

http://www6.tomshardware.com/graphic/20030512/geforce_fx_5900-13.html

1024x768

HIGH QUALITY:
Radeon: 61
GeForceFX 5900: 55

HIGH QUALITY + 4X FSAA:
Radeon: 43.3
GeForceFX 5900: 57.1

Radeon acts as one would expect, certainly not the GeForceFX.

About the ATI driver again. I wonder why the driver would detect the larger framebuffer but won´t use it...

Who knows...there can be any number of reasons. Detecting more memory is a far cry from having drivers that utilize it efficiently, or at all. Cat 3.2 was never an official driver for the DDR-II based Radeon. The facts are, ATI told [H] that the 3.2's don't utilize 256 MB. Do you only believe nVidia when they tell you there are driver issues?

But there´s also no big performance difference between 128/256

You simply can't say that, unless we know that the Radeon9800 256 MB version is actually using 256 MB. All you know is that the 9800 with 128 MB, acts the same as a 9800 with 256 MB, that is only using 128 MB. No surprises there.