[digitalwanderer]

Originally Posted by The Baron

Four hours. Clean the carpets or something.

I am getting SO tired of cleaning up the carpets and floors...

[Joe DeFuria]

Originally Posted by tombman

Ok, here is proof that ut2004 is gpu limited at 1600 4aa/16af

1600x1200 4xAA/16xAF : 58 fps avg
640x480 4xAA/16xAF : 143 fps avg

If you were "GPU limited", you should be getting more like 360 FPS at 640x480.

I can give the .bat and .ini files i benched with so you can do some benches yourselves.

I can't, because I'm using the retail UT, which apparently does not work the same as the demo wrt botmatches. Every botmatch benchmark I've tried to run doesn't start in spectate mode, so the camera just sits there in first person view of an actual spawned player.

Again, the bottom line is it's pretty pointless at this juncture to say if the "NV40 numbers" can be legitimate or not. We simply don't know any of the deatils WRT to the timedemo used, nor do we know game settings.

If you're going to tell me that it's possible that 'some botmatch demo' could be benchmarked to get the NV40 scores, I'd say that I agree. But I'd also say it means absolutely nothing about NV40 performance until we can get a legitimate reference point....which could only be supplied by the same person who supplied the NV40 benchmarks in the first place.

[T2k]

Originally Posted by radar1200gs

I pretty much agree with most of that, however, on the subject of IQ, remember GF-FX is perfectly capable of proper Trilinear and Anisotropic filtering - just load up early drivers for evidence of that, it's simply that in later drivers nVidia decided to use brilinear filtering instead.

Is this the same guy?

Originally Posted by radar1200gs

Well, the fanATIcs are out in force, desperately attacking anyone who comes along claiming nVidia is outperforming ATi.

It would be amusing if it wasn't so pathetic.

Hardly unexpected though...

Why I'm not surprised when you agreed 'pretty much most of that' w/ Chalnoth...?

[PaulS]

The problem to me is that the 16x1/32x0 configuration rumours have come directly from more than one source who are in a position to absolutely and technically know what's going on (i.e. not just say something because they're confused about how it's actually working). Of course, they could have been deliberately lying (perhaps because of a company wide policy already mentioned), but they'd have to have fooled a number of knowledgeable folks (far more knowledgeable than the likes of Fuad). There's also the issue of this very thread containing someone who apparently has access to someone with a card who has stated it's 16x1. Again, this could be false too.

But ultimately, Fuad is on his own at the moment. If we take all of these things as "evidence" (which they really aren't), then there's more evidence to support 16x1 than not. I don't know, but The Inq have been wrong just as many times as the forum rumours have been.

[Natoma]

Originally Posted by tombman

Ok, here is proof that ut2004 is gpu limited at 1600 4aa/16af

1600x1200 4xAA/16xAF : 58 fps avg
640x480 4xAA/16xAF : 143 fps avg

---> GPU LIMITED! (and gpu ram bandwith of course)

(cpu @ 2750 FX51 on both runs)

I can give the .bat and .ini files i benched with so you can do some benches yourselves.

Yes, please provide a link. I'll bench on my system when I get home and confirm to get the wolves off your back, or make sure you get eaten alive. :P

[tombman]

lol @ 360fps :rolleyes:

You just multiplied the 58fps score with factor 6 cause thats 2million pixel for 16x12 / 300k pixel for 648x480 ---> that´s pure crap. That´s not how games work. It´s always a combination of gpu and cpu and there are no linear fps ramps.

The drop clearly is huge enough to show me that the graphics card is the bottleneck there and not the cpu. If it had been only 5 to 10 fps more in 640x480 it would clearly be the cpu being the limiting factor but i was NOT.

You can draw you own conclusions there and i will draw mine.... and i know i´m right cause i do this benchmarking thing long enough.

[L233]

Originally Posted by Joe DeFuria

If you were "GPU limited", you should be getting more like 360 FPS at 640x480.

He said "gpu limited at 1600 4aa/16af". Doesn't mean it can't be CPU limited at 640x480.

[T2k]

Copied from an old UT2k4 thread of another forum:

1280x768, BOTMATCH (min/avg/max):
dm-rankin
35.411644 / 70.706512 / 167.554550 fps
Score = 70.826759

as-convoy
18.810818 / 41.666744 / 81.418251 fps
Score = 41.727188

ons-torlan
16.080410 / 48.993755 / 98.724342 fps
Score = 49.063923

br-colossus
30.182770 / 85.150612 / 207.068161 fps
Score = 85.251747

ctf-bridgeoffate
31.970549 / 88.812943 / 201.651260 fps
Score = 88.937256

Bottom config (CPU at ~2500MHz) and Catalyst v4.1, UT2k4 demo.

EDIT: fixed 4xAA + 8x Q AF, of course.

EDIT2: I forgot the used settings...

In subdirectory called Benchmark


[botmatch-all]

..\System\ut2004 dm-rankin?spectatoronly=true?numbots=12?quickstart=true?attractcam=true -benchmark -seconds=77 -exec=..\Benchmark\Stuff\botmatchexec.txt 
..\System\ut2004 as-convoy?spectatoronly=true?numbots=12?quickstart=true?attractcam=true -benchmark -seconds=77 -exec=..\Benchmark\Stuff\botmatchexec.txt 
..\System\ut2004 ons-torlan?spectatoronly=true?numbots=12?quickstart=true?attractcam=true -benchmark -seconds=77 -exec=..\Benchmark\Stuff\botmatchexec.txt 
..\System\ut2004 br-colossus?spectatoronly=true?numbots=12?quickstart=true?attractcam=true -benchmark -seconds=77 -exec=..\Benchmark\Stuff\botmatchexec.txt 
..\System\ut2004 ctf-bridgeoffate?spectatoronly=true?numbots=12?quickstart=true?attractcam=true -benchmark -seconds=77 -exec=..\Benchmark\Stuff\botmatchexec.txt

../Benchmark/Stuff dir

[botmatchexec.txt]

showhud
ship

edit4: shitty copy&paste

[Ailuros]

Originally Posted by PaulS

The problem to me is that the 16x1/32x0 configuration rumours have come directly from more than one source who are in a position to absolutely and technically know what's going on (i.e. not just say something because they're confused about how it's actually working). Of course, they could have been deliberately lying (perhaps because of a company wide policy already mentioned), but they'd have to have fooled a number of knowledgeable folks (far more knowledgeable than the likes of Fuad). There's also the issue of this very thread containing someone who apparently has access to someone with a card who has stated it's 16x1. Again, this could be false too.

But ultimately, Fuad is on his own at the moment. If we take all of these things as "evidence" (which they really aren't), then there's more evidence to support 16x1 than not. I don't know, but The Inq have been wrong just as many times as the forum rumours have been.

Hmmm I've been saying for quite some time that it's important what comes out on the other end. I'm anything but a professional when it comes to 3D, I do not run a tabloid news site, nor have I signed a NDA.

I honestly would love to see how it's possible without any conditionals involved to get more than 8 pixels/clock out of a 256bit bus (no I'm not thinking of framebuffer compression, TBDR, on chip framebuffer or anything else anyone could come up with...).

The idiot in me oversimplyfies things down to:

8*32=256
16*32=512

Wow and I thought I was bad with math.....

[Joe DeFuria]

Originally Posted by PaulS

The problem to me is that the 16x1/32x0 configuration rumours have come directly from more than one source who are in a position to absolutely and technically know what's going on (i.e. not just say something because they're confused about how it's actually working).

In all honesty, being able to "technically know" what's going on doesn't mean you aren't in a position to be manipulated, by being given only select pieces of information, while leaving out others.

All you have to do is provide a "benchmark" or number that one would traditionally think can't be done unless you have a 16x1 architecture. And one of the best ways to do this is to use a benchmark that in the past was used to "prove" that NV3x was not 8x1. Say...3DMark01 fill rate tests.

This doesn't mean that said benchmark can't be manipulated in new and previously unthought-of ways.

The only way we're ever going to know, is for some of the knowledgable people to actually get a card and run it through a variety of their own tests.

[tombman]

Originally Posted by L233

He said "gpu limited at 1600 4aa/16af". Doesn't mean it can't be CPU limited at 640x480.

exactly

With a 3.5 ghz FX53 it WILL go much faster @ 640x480

[Ailuros]

1280x768

[Joe DeFuria]

Originally Posted by L233

He said "gpu limited at 1600 4aa/16af". Doesn't mean it can't be CPU limited at 640x480.

Correct. It also doesn't mean it can't be significantly CPU limited at 1600 4AA / 16 af either. (Both CPU and GPU limitations.)

I so much just want to be able to run my own tests.....sigh. I can't believe it's been out about a week now, and no one knows how to benchmark it?

[tombman]

Originally Posted by Ailuros

and? I could do this res too ..> it´s a widescreen resolution and my 24" Sony 16:10 CRT has no problem with it

[PaulS]

Originally Posted by Joe DeFuria

In all honesty, being able to "technically know" what's going on doesn't mean you aren't in a position to be manipulated, by being given only select pieces of information, while leaving out others.

Absolutely, but it tends to give the whole thing more credibility that just a PR guy pissing up the wall. I'm as skeptical as anyone about this and I've had enough private discussions for that not to be the case. It would be a mammoth step for nVidia, and I can't see it happening without a shed load of conditionals.

But just because something is possible (like being manipulated), it doesn't mean it's the way it happened.

[tombman]

here is something funny i did:

3.3 Million Pixel 2304x1440 4xAA/16xAF: 22fps avg

[CyFactor]

Originally Posted by Ailuros

Looks like a widescreen mode. UT2004 supports widescreen modes for HDTV and widescreen panels (I know because I run it on a 65" HDTV in 1080i).

[T2k]

Originally Posted by tombman

and? I could do this res too ..> it´s a widescreen resolution and my 24" Sony 16:10 CRT has no problem with it

Exactly - dunno what's Ail's problem...

PS: BTW it's probably the same Sony here - just resold as HP...

[Pete]

Joe, just d/l the UT2K4 demo and UnrealMark(.com), and queue up some 16-bot matches in all four test maps. You can also use the 3DCenter botmatch demo (I believe someone linked previously in this thread), as it's far more stressful and runs longer than any UMark benches.

[digitalwanderer]

Originally Posted by Joe DeFuria

I so much just want to be able to run my own tests.....sigh. I can't believe it's been out about a week now, and no one knows how to benchmark it?

<puts on tinfoil hat>

nVidia is behind it, they're paying Epic/Atari to not release the benchmark until the nV40 is "released"

<removes tinfoil hat>

[Hyp-X]

Originally Posted by Ailuros

The idiot in me oversimplyfies things down to:

8*32=256
16*32=512

Wow and I thought I was bad with math.....

It's not that simple because it's 2*256 if you use DDR.
Might be more if the RAM has higher clocks than the GPU.

So it's possible to push 16 pixels trough a 256bit DDR bus.

It's just not possible to do that the way the 3DMark fillrate test work - with alpha blending enabled. That means you need to read, modify and write back the pixels requiring twice the bandwidth.

[Joe DeFuria]

Originally Posted by PaulS

Absolutely, but it tends to give the whole thing more credibility that just a PR guy pissing up the wall. I'm as skeptical as anyone about this and I've had enough private discussions for that not to be the case. It would be a mammoth step for nVidia, and I can't see it happening without a shed load of conditionals.

But just because something is possible (like being manipulated), it doesn't mean it's the way it happened.

Don't get me wrong...I'm certainly not assuming one way or the other. But I've seen people who are usually "in the know" go back and forth so many times I've lost count. That, to me, is an indicator of some type of "conditional" situation, with the "conditions" are not being spelled out by whoever is ultimately the source of the info.

Put me in the "skeptical of the true 16x1/32x0 spec" category myself, but I certainly wouldn't say anything remotely like "there's no way it can happen."

[MatiasZ]

somehow back on track.. after htting F5 for about half an hour around here i got bored and decided to take a look at the "facts" once again so i went to the first page and looked at the "benchmarks" posted by vegetto and realized this:

UT2004 ( 4xAA, 4xAF ) 1600x1200 71.9fps - 1280x960 82.9fps

there is LITTLE difference between the two resolutions, which would implie that the benchmark is somewhat CPU+GPU limited, and i think after all that was posted here, we can conclude that a fast CPU (and no fillrate limitation - aka running at 640x480 without AA & AF) make's it possible for ~130 fps to be achieved in UT2K4, and if you have a powerfull enough video card, you can keep up to it while running 1280x960 4xAA, 4xAF , at least to the point of achieving 80 fps... it just makes sense (not saying the benchmark is correct and coming from NV40, just saying is not neccesarily wrong)

edit: bolded some things

[T2k]

Originally Posted by MatiasZ

somehow back on track.. after htting F5 for about half an hour around here i got bored and decided to take a look at the "facts" once again so i went to the first page and looked at the "benchmarks" posted by vegetto and realized this:

UT2004 ( 4xAA, 4xAF ) 1600x1200 71.9fps - 1280x960 82.9fps

there is LITTLE difference between the two resolutions, which would implie that the benchmark is somewhat CPU+GPU limited, and i think after all that was posted here, we can conclude that a fast CPU (and no fillrate limitation - aka running at 640x480 without AA & AF) make's it possible for ~130 fps to be achieved in UT2K4, and if you have a powerfull enough video card, you can keep up to it while running 1280x960 4xAA, 4xAF , at least to the point of achieving 80 fps... it just makes sense (not saying the benchmark is correct and coming from NV40, just saying is not neccesarily wrong)

edit: bolded some things

That's why I posted my 1280x768 on the last page.

[Mariner]

I'm not sure if I'm getting confused or not here, but would the different aspect ratio of a widescreen resolution put a different strain on the amount of CPU power required by the game? :?

I've read so many posts in the speculative threads over the weekend, I'm losing touch with reality...