A promise is a promise :) there you go!

Finally, you answer the Q. It wasn't so hard now was it?

I was never presented with an argument, before this, that the 3200+ could not be run at a higher multiplier, and thus I gave them the benefit of the doubt.

But if as you say AMD has put a ceramic topping over the L1 bridges, then we can rule out that option, and thus the information that the test system was using a multiplier of 11x becomes bogus.

Happy?

[EDIT]
Hey, it's not my fault you take 10 min to write the message I've quoted. Go check the message timestamps and you'll see what I mean.
[/EDIT]

 

why do you think I should do YOUR homework? if you make a comment and intend to debate something, you really should consider KNOWING the facts beforehand.....


I'm done, class dismissed

 

hmm.. this place is starting to read like the "old" Rage3d on a good day

...............

After ploughing through all the posts I've yet to find information that the tests with the NV40 are the same as Xbits. Rather, that the NV40 was tested with a low polygon small map called Deck 17 whereas Xbits ran a custom demo using all the maps in the demo - of which all are much bigger than Deck17.
But, taking that aside and assuming the tests were in fact on an even level saying the UT2004 is just CPU limited in botmatch would effectively mean that an ATI 9600XT would offer a similar fps as a 9800XT. which of course, it doesn't. Rather, it's closer to half the performance:

x4aa x 8af
9800XT 8 pipe - 37.2
9600XT 4 pipe - 18 (when o/ced closer to 9800XT spec)
x4 aa x4af
NV40 16 pipe - 71.9
(looks like a pattern to me)

There is also no information as yet as to the NV40's AA/AF efficiency either. I mention this as the Xbit tests would have been using MSAA where the Deck17 image is using SuperSampling or a at least a mix.
As the screenshot used SSAA or a mix it's likely that it was used on the complete test too.
..........................

While the initial early performance results could be seen as impressive, for todays new games there's still going to be the same old problems of either screen tearing or Vsync Stuttering.
For instance, whether you run Farcry at 30fps or 50fps you still encounter the same image problems until you can guarrantee a constant fps performance that is the same as your Monitor Hz.

Of course if the NV40 has a dual framebuffer...........

 

1) The Xbit labs results are not cpu bound, if they were cpu bound then the 9800Xt would have the same performance as the 9800 Pro, but it doesn't. Conversly increasing the 9800 Pro by overclocking would not increase fps at all, however it does.

Finally the XT9600 would not have half the frame rate at 1024x768 and 1/3 at 1600x1200. It would have the same frame for all resolutions.

2) The reason for the above is that the xbitlab benchmarks are not even botmatches, they are their own benches, see here

http://www.xbitlabs.com/articles/video/display/ut2004_3.html

"Since the Unreal Tournament 2004 Demo contains no pre-recorded demos and because of our policy to use our own demos and benchmarks when possible, we recorded a number of sequences in all levels available in UT2004 demo."


3) If the nv40 person had used botmatches then they will have been capped by the cpu, flyby would have been a better test.


Given the above I am surprised that so many people here do not know what a cpu capped test is ( all video cards have almost the same score ), I'm surprised people have shown graphs of UT2003 when the score was from UT2004 and I am surprised a botmatch score has been compared to a non botmatch Xbit bench ( I thought you guys liked " apples to apples " )

The amount of stick you guys give web site reviews and you prove you cannot even read what they are saying correctly

Some of you had spotted the mistakes that the critics had been saying, but as they were 10x as noisy , I'm afraid your voice got overwhelmed, so to compensate

THIS ARGUEMENT ABOUT CPU CAPPED BOTMATCHES IS BASED ON AN INCORRECT ASSUMPTION AND THEREFORE IS BOLLOCKS.

 

Exactly.

 

What about compression that nv3x and r3xx have? Isn't it likely it was improved over the nv3x algorithm, or does compression not matter for whatever way fillrate is tested? Even if it's the same as in nv3x, shouldn't it help at least some?

 

I'll tell you what, if some of you guys continue to act like you have in this thread in the coming months, you're not going to like me very well. Because I really don't have the patience for all this fan-based, back 'n forth bickering and sniping, and I don't feel like watching every upcoming R420/NV40 thread turned into this one.

 

Most rendering won't need the full memory bandwidth required by single textured rendering, which will be more and more the case moving forward with more complex shaders and more textures.

 

We are going to have to wait for months to pass by until we hear the actual specs damn you must know something I don't :/

 

edit 2 : I realize now that my points have been made earlier in this thread (pages 9, 10). Sorry for going further OT.


Don't be a bitch Tagrineth. You are being unfair to Veggetto-ex for NO reason. He tried his best to provide us with info that we would have to wait several weeks for. His friend who did the benchmarks was most certainly under NDA. The numbers he got us seem to be fairly accurate. The CPU-limited comment you make is complete bullshit. I have several points to make.

edit: source of Tagrineth and my benchmarks:http://www.xbitlabs.com/articles/video/display/ati-nvidia-roundup_13.html



1. Before you go calling anyone an idiot double check the benchmarks from the source that you got them. Unreal Tournament 2004 at 1600*1200 4x aa 8x af was NOT cpu limited. It was GPU limited. Why? DOUBLE CHECK THE DAMN NUMBERS. At 1024*768 all high-end gpu's scored in the 90's range. At 1600*1200 that number dropped to 65. However, physics and AI calculations are not increased as res goes up, proving that at 1600*1200 4x AA 8x AF the gpu, NOT the cpu, was the cause of the 37fps figure for 9800xt.

2. The 9800 pro OC figure is BULLSHIT. The 9800pro is overclocked to 420/360. This is 1.9% higher for gpu and 1.38% LOWER for memory. At such high resolution and AA/AF figures, there is NO reason for 9800pro overclocked to be faster than the 9800xt stock.

Please do not troll these threads posting useless crap. You are flinging mud, without ANY supporting evidence, and mucking up this thread. While I have a low post count I've been reading B3D actively for several years, and I absolutely cannot have you stand in the way of fruitful discussion. Also, why did you go and call him an IDIOT? He did nothing to prove or disprove his intelligence. He was merely relaying facts. As far as we know (Baron, Uttar) Veggetto-ex is truthful. God you pissed me off.[/url]

 

Although the 3dmark score was not posted did the person running it save the project ?

Can you take a screen print of the scores by game so we can look at the game by game scores ?

Thanks

 

This is completely irrelevant to the discussion here. I am disputing the numbers provided, not how useful something is or is not.

-FUDie

 

The only thing I can really say at this point:

1) I don't see why Veg's benchmarks numbers can't be possible. Botmatches aren't CPU limited at 1600x1200 with 4X AA and aniso, AFAIK. (See benchmark graphs that range from 640x480 though 1600x1200 with AA)

2) Having said that, because we don't know exactly what test, nor do we have any reference to compare too, the numbers are pretty useless.

There...I summed up the past 9 pages with 2 bullet points.

 

Fool me once shame on you. Fool me twice shame on me. I still haven't forgotten all of the bogus posts and news stories about nv30 before it's release. At this point I'll believe it when I see it. I won't believe any "a friend of mine told me" garbage this time around.

 

On a side note, I have Ut2K4, and a 9800 non-pro, running on a 3.2 Ghz P4. Maybe I'll try and run a few benchmarks to test the "CPU limitedness of 1600x1200 with AA and Aniso....

 

It would be refreshing to see _actual_ data used to prove/disprove whether or not its cpu limited.

It got real tiring seeing name calling instead of reasonable argument/debate from some of the more senior b3d members.

later,
epic

 

Being the lazy git that I am....

Can someone
1) point me to a page that shows how to benchmark UT2K4,
2) Ask me what maps they want to see test. (And does UT2K4 have fly-by, as well as bot-match benchmarks?) IIRC, the demo only officially had some botchmatches, but you could hack in a fly-by or two...)

 

This thread over at nvnews.net should help you out joe.

http://www.nvnews.net/vbulletin/showthread.php?s=&threadid=26167

later,
epic

 

With a 3Dmark01 score of >36K I would think it is safe to presume that he has considerable CPU power.

 

Well upon reaing test results I was suprised that noone commented on 36k 3DMarks01 as we know that it is also CPU bound test.

I could be wrong but I dont think that any configuration at stock frequencies can break 20k so >36k seems as wishfull thinking.

Remember when Radeon 9700 Pro arrived? It was only marginally faster in 3DMark01 from Ti4600 without AA and AF.

Zvekan