Official HardOCP HL2 benchmarks

Well what about those people who would prefer to understand the underlying factors that determine each offering's performance? It's one thing to sit and grin because X is faster than Y but that's not enough for some people. As a developer and computer enthusiast I am very used to knowing the cause behind the effect.

 

Someone has begun to answer my question!!! Thank you sir. How many alu's does R420 have at it's disposal?

 

I guess it's obvious with the XT vs Ultra but how come the Pro is so fast at shading with only 12 pipelines. Is pixel shading power somehow detached from the pipeline count and more reliant on core speed?

 

1 full dx9 PS alu, 1 'mini' alu(still not sure about which operations it can do exactly), and a texture op.

 

Per pipeline, not the entire chip.

 

thats kind of obvious

 

Just saw this little tidbit from a digit-life article.. Is that what you're referring to?

Seems pretty detailed at first glance. Gonna read it through and see if it answers some of my questions without pestering you guys

 

somewhat. the nv40 has i believe, 2 fully operation dx9 alus per pipe, which can, under optimal conditions, compute up to 5 shader ops. but 1 of the alus is shared for texture ops.

 

I'll just add Deano C's comments about the NV40 vs X800.

And further on:

And Valve is pretty much in bed with Ati (upcoming "Ati" levels" f.e) so i'm guessing that the time Valve spent on optimizing any NV40 path is close to much non existing.

 

And with the xbox2 going to ati and ms's xna stuff, you are probably going to see a whole lot more like this in the future.

 

Hmmmm that sucks. Anything Nvidia can do at the driver level?

 

Just for the sake of experimenting. (I dont have HL 2 yet) with 3danalyze and how Forcing Fp16/Fp32 precision? To see the effects it has on performance?

Also the smoke issue should also be investigated.. I guess I need a copy of HL2.

 

did some tests:

1280x1024x 16AF (without af mipmap opt. for the 6800U!), max IQ , "waterreflect "all", with a FX53

anand's demos,

6800U, driver 70.41:

canals 08 90 fps
coasts 05 113
coasts 12 99
prison 05 111
C17-12 73

X800 XT PE :

canals 08 88 fps
coasts 05 115
coasts 12 105
prison 05 119
C17-12 73

ocp-benchmarks;

6800U

canals01 112
coast 03 100
Canals09 60

X800XT PE

canals01 106
coast 03 103
Canals09 75



As you can see only waterspray canal09 takes a hit on 6800 hardware

 

What worries me is a trend that is developing. I dont want IHVs to work so closely with developers that their next game runs worse on the other IHVs hardware. We are starting to see signs of that (NV/Doom3 and ATI/Valvle). Would be nice if we could all just get along.....

 

jb, I absolutely agree. And while I will be flamed for saying this(again ), it's been nVidia that has been pushing this. The failed NV3x and CG was an attempt to take over the the gaming production market with nVidia specific API extentions. ATI is now following suit... The idea of building a better mousetrap and the world will follow has been replaced with a strongarm approch. And the ones caught in the middle are both the software developers and the consumer.

 

your benchmarks are kind of odd. can you elaborate.

 

That second quote sounds a little Politically Correct to me. IIRC there were numerous comments from people that little work was needed in order to have efficient code on the R3xx compared to the nV3x. I would assume that this would carry forward with the R4xx.

 

Well, what can I say, my experience is the 6800u is dooing fairly well against the X800XT. Maybe the 70.41 have something to do about it.

And I watched the IQ arefully and see no difference in shader IQ.

 

i meant could u include the results from the websites u d/led the demos from so we can compare your results to theirs. also which driver for the xt pe. are those ur own results with an xt pe or are they from 1 of the aforementioned sites.

 

Maybe the reason it's "non existing" is that the nv40 cards didn't really need as much optimizing as the nv3x line. So they chose to spend all of their time trying to optimize for the nv3x cards.

http://www.hardocp.com/article.html?art=NTE5LDM=