R680 To Score 20K In 3DMark06

[ViRGE]

Originally posted by: munky
All this bitching about 3dmock, I know it's meaningless, so why doesn't everyone use some standard benchmark that we all agree on? Oh, that's right, because there isn't one!

Quake 1. I dare anyone to complain that Quake 1 isn't suitable.

 

[Dribble]

tbh I think ati's drivers are about 6 months behind nvidia's and they've only just started to get the X2900's to work well in DX9 let alone DX10. Look at bioshock:
DX9, single card X2900XT=8800GTX.
DX10, single card X2900XT frames drop to about 2/3's of DX9 score, 8800GTX frames stay about the same.
DX10, crossfire with X2900XT is actually slower then just using a single card.

 

[Ackmed]

Originally posted by: n7
It kinda seems like they are.

Look @ the 2900s.

They all do ridiculously well in 3DMock, & yet terribly in games comparably

Dual GPU scares me...if it requires CF to be enabled, it'd better be dirt cheap, since we all know how well CF works...not.

Another reason 3dmark can be misleading, and flat out wrong. Yet some people think its great.

The 2900 doesnt do "terrible" though in games, but its not close to the 8800gtx in most cases.

 

[tshen83]

Dual RV680s score 20K(ie 4 GPUs), not one RV680, which is dual RV670s. So 4xRV670s score 20K.

ATI's marketing playing with the numbers.

 

[fleabag]

Originally posted by: SickBeast
I could actually see this type of thing happening and being of great benefit. AMD should be able to easily 'attach' multiple ATI GPUs together just like they would a multi-core CPU. It would probably be faster than crossfire, and would be much much cheaper. 1 PCB vs. 2, and you wouldn't waste the memory of one of the cards.

Actually you do, just check out the images of the card and you'll see how its configured..

 

[thilanliyan]

Originally posted by: tshen83
Dual RV680s score 20K(ie 4 GPUs), not one RV680, which is dual RV670s. So 4xRV670s score 20K.

ATI's marketing playing with the numbers.

I doubt that. A single 3870 can already score over 10000 in 3DM06. If it takes ATI 4 GPUs to get to 20000 they are really screwed. I'm pretty sure it's a single R680 that would do that.

 

[Rusin]

Well single HD3870 X2 can do it's overclocked. If it gets 17000 points in 3dmark06 with E6600

 

[RussianSensation]

....and the minute you enable 4AA the score drops to 2K

 

[Drayvn]

do remember guys, we can bicker and argue over amd using 3dmark as their goal. But really in the end, what do they care about us, they can use that for marketing, so if that card can reach 20000 points and the comparable nvidia card can only reach half that or 15000 points, in the minds of anyone reading an advert, who do you think won?

 

[AzN]

Originally posted by: votelibertarian35
wow, i sure hope that AMD isn't so stupid that they are engineering their card around a 3dmark benchmark! try actually achieving performance in games for a change! R600 wins no meaningful benchmarks against any competition (i.e. settings that actually matter, AA in a game where you would use it, etc.)

3dmark is dependent on shader. Having 2 3870 on a single board with 640SP might have something to do with 20,000 3dmark scores. This card however would surely beat 8800ultra and then some in real games most likely.

 

[AmdInside]

I really wish there was a better bench more akin to something like Viewperf where it would run a card through a bunch of benchmarks designed around different gaming engines. Like there could be a test for the Unreal3 engine, and a different test for the engine which HL2 uses, etc. and the final score would be based on how well the card does as a whole. I hate reading 3dmark 2006 scores when is shown to not directly have a direct translation to real world performance.

 

[Killrose]

Seeing pics of Intel's own latest and greatest product being shown with AMD(ATi) dual core product at CES has got me to beleive that ATi does have a very good product about to be released. If Intel is using its competitors product line to show off their high teir product its got to make you wonder. I think Intel still wishes it had gotten their hands on ATi.

 

[thilanliyan]

Originally posted by: AmdInside
I really wish there was a better bench more akin to something like Viewperf where it would run a card through a bunch of benchmarks designed around different gaming engines. Like there could be a test for the Unreal3 engine, and a different test for the engine which HL2 uses, etc. and the final score would be based on how well the card does as a whole. I hate reading 3dmark 2006 scores when is shown to not directly have a direct translation to real world performance.

That would be ideal. Is there any possibility of this happening?

 

[nullpointerus]

Originally posted by: thilan29

Originally posted by: AmdInside
I really wish there was a better bench more akin to something like Viewperf where it would run a card through a bunch of benchmarks designed around different gaming engines. Like there could be a test for the Unreal3 engine, and a different test for the engine which HL2 uses, etc. and the final score would be based on how well the card does as a whole. I hate reading 3dmark 2006 scores when is shown to not directly have a direct translation to real world performance.

That would be ideal. Is there any possibility of this happening?

The game engine is not the only factor: level data also affects fps.

From what I've read, different level designs can change engine performance characteristics. So one level for a given game engine might favor one card while a different level could favor another. It all depends on what effects etc. are used to what degree.

Getting an accurate benchmark for a particular game would require creating custom level(s) for each game...which means you are back on a per-game basis instead of a per-engine basis.

EDIT:

I think the ideal benchmarking tool would not be a game benchmark at all, but rather a programmable engine enabling reviewers to write scripts that can test various aspects of a card's performance directly. So if game benchmarks point to a possible issue with a memory controller, shaders, or AA in general, those things could be tested more directly instead of just speculating and contacting nVidia/ATI for public statements.

 

[thilanliyan]

Originally posted by: nullpointerus
I think the ideal benchmarking tool would not be a game benchmark at all, but rather a programmable engine enabling reviewers to write scripts that can test various aspects of a card's performance directly. So if game benchmarks point to a possible issue with a memory controller, shaders, or AA in general, those things could be tested more directly instead of just speculating and contacting nVidia/ATI for public statements.

Isn't that what 3DMark does? (ie. it tests shaders, etc. for the most part independently?)

 

[n7]

wtf did this get bumped from two months ago...

 

[Syntax Error]

Originally posted by: n7
wtf did this get bumped from two months ago...

Didn't notice that in the first read of this thread, but now it does make sense.

Always fun to see who was the bumper-in-question.