WTH? GFFXU beats out ATI?

[Shamrock]

No flaming here, just reporting news.

According to Xbit labs, the GeForce FX beat ATI in about 10-13 tests (ALL are 3dmark2k3) They have a 13 card line up.

XBit Labs

How trustworthy is XBit Labs?

Shamrock


P.S. Xabre 600 won one test!

 

[Rand]

All of the tests were derived from 3DMark 2003, and I think most would agree that 3DM isnt exactly the best indicator of real world gaming performance.
In any case, most benchmarks have shown that the FX Ultra is indeed faster then the 9700Pro in 3DM03 in most instances, so this isnt particularly surprising.

They made no real effort at all to equalize image quality between the different adapters in any case, so I can't say I'd put much faith in the results.

The quality offered by the anisotropic filtering and FSAA implementation between the varying graphics cards differs dramatically even at identical settings.
4X FSAA/8X anisotrophy was used for each card.... the quality of said setting varies dramatically between the GF4/R200/GF FX/Xabre.
Beyond FSAA, and anisotrophy they've left all driver setting to default... and with each graphics card having different default settings it efectively ensures their never comparing each board using the same settings with similar image quality.

IMHO all XBit has shown is a bunch of numbers, that individual graphics cards may achieve depending on their settings.
Without attempting to equalize image quality however they are most definitely not comparable.

 

[chizow]

Interesting comments Rand, weren't you defending 3DMark2K3 in a previous thread when it was released?

Chiz

 

[nemesismk2]

Originally posted by: Rand
All of the tests were derived from 3DMark 2003, and I think most would agree that 3DM isnt exactly the best indicator of real world gaming performance.
In any case, most benchmarks have shown that the FX Ultra is indeed faster then the 9700Pro in 3DM03 in most instances, so this isnt particularly surprising.

They made no real effort at all to equalize image quality between the different adapters in any case, so I can't say I'd put much faith in the results.

The quality offered by the anisotropic filtering and FSAA implementation between the varying graphics cards differs dramatically even at identical settings.
4X FSAA/8X anisotrophy was used for each card.... the quality of said setting varies dramatically between the GF4/R200/GF FX/Xabre.
Beyond FSAA, and anisotrophy they've left all driver setting to default... and with each graphics card having different default settings it efectively ensures their never comparing each board using the same settings with similar image quality.

IMHO all XBit has shown is a bunch of numbers, that individual graphics cards may achieve depending on their settings.
Without attempting to equalize image quality however they are most definitely not comparable.

I'd like to see you try to get ati/nvidia comparable image quality on a Xabre!

I do understand what your saying though, the problem is that video card companies can put alot of pressure on review sites to get the review they are happy with. It would explain the huge divide with the Geforce FX reviews for example, in one review it could be losing badly to the Radeon 9700 Pro and in another review with a similar spec system it's a totally different story.

 

[Lonyo]

Of course, you could say comparing the 6 month old 9700PRO to 5800Ultra is infair, and that they should have included the 9800PRO which woul dhave shown different results.

 

[YBS1]

This whole test is invalid before it even started. Look at the driver revision they used for the FX.

When they decide to test the FX with a WHQL certified driver using comparable image quality settings give me a call.

 

[Schadenfroh]

on paper the GFFXU is superior to the 9700pro

Geforce fx verus raddy 9700
core clock 500mhz verus 325mhz
memory clock 500mhz versus 310mhz
process .13 versus .15

 

[YBS1]

From the conclusion on Xbit:

First of all, we should congratulate NVIDIA and the company?s software developers. Their 42.68 drivers helped GeForceFX 5800 Ultra based card to outrun its main rival, RADEON 9700 PRO nearly in all the tests and resolutions.

ROFL...I wonder if they would "congratulate" them if they had tested using the pending WHQL 43.xx series driver, or for that matter the latest still pending 41.09 driver listed on nVidia's site dated December 3rd. LOL...The "nVidia makes 1337 drivers d00d!!!" myth still remains amazingly.

 

[YBS1]

Yeah, paper is funny like that...real world however is very different.

In the real world:
Though it does indeed have a core clock advantage, it has to be clocked higher, it's fairly obvious it not as efficient on a per clock basis. Further, it's texturing pipeline is poor compared to the 9700's, the FX doesn't behave as a true 8x1 design.
Memory clocks - FX=500mhz(1000mhz effective) on a 128bit interface, 9700=310mhz(620mhz effective) on a 256bit interface. Memory bandwidth: FX=approx. 16gb, 9700=approx 20gb. Owned, move along nothing to see here...
.13 vs .15 - Manufacturing process has nothing to do with the performance of a chip. A 9700 built on a .13 process clocked the same would be no faster, likewise an FX built on a .15 process clocked the same would be no slower, attaining said clock speeds however would prove tricky.

 

[Lonyo]

Originally posted by: Schadenfroh
on paper the GFFXU is superior to the 9700pro

Geforce fx verus raddy 9700
core clock 500mhz verus 325mhz
memory clock 500mhz versus 310mhz
process .13 versus .15

That mem clock means nothing unless you look at the double width bus of the 9700PRO which gives it more mem bandwidth than the FX.

 

[Keysplayr]

Originally posted by: YBS1
This whole test is invalid before it even started. Look at the driver revision they used for the FX.

When they decide to test the FX with a WHQL certified driver using comparable image quality settings give me a call.

Why would we call you?

 

[Keysplayr]

Originally posted by: YBS1
From the conclusion on Xbit:

First of all, we should congratulate NVIDIA and the company?s software developers. Their 42.68 drivers helped GeForceFX 5800 Ultra based card to outrun its main rival, RADEON 9700 PRO nearly in all the tests and resolutions.

ROFL...I wonder if they would "congratulate" them if they had tested using the pending WHQL 43.xx series driver, or for that matter the latest still pending 41.09 driver listed on nVidia's site dated December 3rd. LOL...The "nVidia makes 1337 drivers d00d!!!" myth still remains amazingly.

Joke

 

[Keysplayr]

Originally posted by: YBS1
Yeah, paper is funny like that...real world however is very different.

In the real world:
Though it does indeed have a core clock advantage, it has to be clocked higher, it's fairly obvious it not as efficient on a per clock basis. Further, it's texturing pipeline is poor compared to the 9700's, the FX doesn't behave as a true 8x1 design.
Memory clocks - FX=500mhz(1000mhz effective) on a 128bit interface, 9700=310mhz(620mhz effective) on a 256bit interface. Memory bandwidth: FX=approx. 16gb, 9700=approx 20gb. Owned, move along nothing to see here...
.13 vs .15 - Manufacturing process has nothing to do with the performance of a chip. A 9700 built on a .13 process clocked the same would be no faster, likewise an FX built on a .15 process clocked the same would be no slower, attaining said clock speeds however would prove tricky.

More Jokes

 

[kylebisme]

the only way the fx beats the 9700pro in 3dmark03 is by running at a lower percision and hurting the quality of the image. here is a good thread on the issue:

link

 

[DaveBaumann]

They benched on the 'default' driver settings. This means that under normal rendering ATI is doing proper Trilinear filtering, but GFFX is doing a lower quality form of filtering thats somewhere between Bilinear and Trilinear.

 

[Bingo13]

I had an 1600 point increase in my 3DMark 2003 scores when moving from the 43.00 back to the 42.68 they tested. These scores are driver related as the images produced by the 43.00 are better in real games compared to the 42.68 drivers. It does not matter in the end as my GFX5800 is going back or will be sold. I found my Ti4600 card had better image quality and "real world" performance in games than the 5800. The 5800 certainly benched better but the high/low scores were all over the map. If you have to buy a new card then spend your money on the 9700/9800 cards.

 

[1ManArmY]

Cool 9800/9700 here I come

 

[sash1]

Originally posted by: TheSnowman
the only way the fx beats the 9700pro in 3dmark03 is by running at a lower percision and hurting the quality of the image. here is a good thread on the issue:

link

you do realise that everyone who believes that is completely idiodic. With FSAA and AF enabled, both cards will look different. ATi and nVidia dun use the same techniques to do AF + AA, thusly they will look different. As for ATi being brighter, whatever.

3DMark is crap, doesn't say anything about performance. IMO, it proves which company is better at optimising their drivers to make their card look better on 3dmark. My ol' Kyro II can;t run a single test in 3dMark2k3, thusly I have a score of ZERO. SO my card must not be able to play any games, is what this is suggesting?

~Aunix

 

[YBS1]

lol@keysplayr Your arguments are overwhelming...I stand corrected.

 

[kylebisme]

Originally posted by: AunixM3

Originally posted by: TheSnowman
the only way the fx beats the 9700pro in 3dmark03 is by running at a lower percision and hurting the quality of the image. here is a good thread on the issue:

link

you do realise that everyone who believes that is completely idiodic. With FSAA and AF enabled, both cards will look different. ATi and nVidia dun use the same techniques to do AF + AA, thusly they will look different. As for ATi being brighter, whatever.

3DMark is crap, doesn't say anything about performance. IMO, it proves which company is better at optimising their drivers to make their card look better on 3dmark. My ol' Kyro II can;t run a single test in 3dMark2k3, thusly I have a score of ZERO. SO my card must not be able to play any games, is what this is suggesting?

~Aunix

reality is waiting for you when you are ready for it.

 

[Rand]

Originally posted by: chizow
Interesting comments Rand, weren't you defending 3DMark2K3 in a previous thread when it was released?

Chiz

I was defencding the developers decisions in how the scenes were rendered. I do not believe 3DM03 is a good benchmark, but I do believe that it's often being unfairly maligned.
My personal reasoning for disliking it tends to be dramatically different then the common sentiment as to 3DM's faults.

It does have it's benefits however.... frame based rendering system, and anisotrophy texture filtering analysis are quite good all considered.

 

[gururu]

what do you mean WTH?

even if it is, and I'm sure this will become clear with time and drivers, my 9700pro will still be up in the ranks.

 

[BFG10K]

(ALL are 3dmark2k3)

There's your problem right there.

As for real games, the FX only wins when memory bandwidth isn't the primary limiting factor (ie fillrate limited settings such as 1024 x 768) and also at the cost of reduced image quality.

If you want a card offering the highest image quality possible at the fastest possible speed then the Radeon 9700 Pro is the one to get.

 

[Keysplayr]

Originally posted by: YBS1
lol@keysplayr Your arguments are overwhelming...I stand corrected.

Umm, those comments were in your favor dude...

Oh, and thanks for the registry edit for disabling v sync on my GFTi4600. Worked perfectly..

Keys.

 

[nemesismk2]

Originally posted by: AunixM3

Originally posted by: TheSnowman
the only way the fx beats the 9700pro in 3dmark03 is by running at a lower percision and hurting the quality of the image. here is a good thread on the issue:

link

you do realise that everyone who believes that is completely idiodic. With FSAA and AF enabled, both cards will look different. ATi and nVidia dun use the same techniques to do AF + AA, thusly they will look different. As for ATi being brighter, whatever.

3DMark is crap, doesn't say anything about performance. IMO, it proves which company is better at optimising their drivers to make their card look better on 3dmark. My ol' Kyro II can;t run a single test in 3dMark2k3, thusly I have a score of ZERO. SO my card must not be able to play any games, is what this is suggesting?

~Aunix

I agree 100% with your comments about 3DMark, I've heard 100's of comments in favour for 3dmark and
IMO none of them are valid. I still think people should be using games for benchmarking, what will help put the final nail in the 3DMark coffin is if more games had easy to use built in benchmarking software.

I'm sure the Kyro2 fans will find a way to run the tests, maybe the person behind 3danalyser?