A promise is a promise :) there you go!

http://www.guru3d.com/article/article/125/3/

One more link from heraldia

And what would this part specially, mean to all of you in the know.

 

Yeah, it is. But every so often people need to be reminded of the hyperbole we've been fed in the past- it's always good to have a little skepticism whenever you here outlandish performance claims.

 

See above. Don't believe everything the Inquirer says.

 

On the encoding hint, some time ago I seem to remember there being a comment on the NV40 being capable of encoding both MPEG2 and MPEG4 streams.

That could be interesting, and if the architecture is flexible enough it certainly could dramatically speed up encoding of MPEG4, but I've been doing all of my encoding at double-precision floating point, so if true, I'm still not sure I'd use it.

 

akira888 suggested storing depth slopes per pixel would solve the intersecting objects problem and papers like Z3 have shown this to be the case. Of course this increases the memory requirements, but it's probably the most straight forward way to antialias intersecting objects. I suspect using multiple passes will have lots of corner cases, but I've never thought about it.

Also, I wasn't suggesting read efficiency is ideal any way a 256 bit bus is partitioned. I was just saying that Parhelia does indeed partition the bus by using multiple memory controllers.

 

Where I come from, essentially all means most. How does emphasizing the "essential" make the statement true, legal or not? I can't see it as anything but a bold-faced lie, but maybe I'm missing something. Are you saying nV is doing something so unbelievably sneaky as referring to CPU-limited situations? :?

 

I didn't really mean to emphasize essential, so much as include it with the rest of the emphasis that SsP45 put in place.

 

Slightly OT now:

Anyone remember Jedi Outcast? OpenGL game, based on the Q3 engine. This is (for me) the prime example of a CPU-bound game.

Take a look:

http://www.anandtech.com/video/showdoc.html?i=1735&p=8

3 fps difference between 10x7 to 16x12... that is cpu-bound.

http://www.anandtech.com/video/showdoc.html?i=1619&p=9

Look at the CPU scaling graph there... The MX400 is clearly maxed out, but everything else is CPU limited. The 7500 hits a wall at about 1533 mhz in which it goes no higher, becoming gpu limited.


In conclusion, UT2k4 is not cpu-bound. At high resolutions, the burden is mostly on the GPU and system bandwidth.

We now return to our scheduled thread.

 

That was also a significantly slower CPU than most benchmarks today use.

 

yeah but UT2K4 is far more CPU intensive than Jedi Outcast

 

No no no ... I wasn't trying to prove anything or start any more arguement... I just wanted to show everyone what truely cpu limited gameplay looks like...

 

About 20 pages ago people were arguing that Nature in 3dmark2001 was capped by the video card ( I can't be bothered to find it )

Here's some results on a P4 running at 3400Mhz on 875i with 5800 video card

640x480 122.8 fps
800x600 94.7 fps
1024x768 64.3 fps
1600x1200 30.8 fsp

It's not capped at all on a P4, at least above 640x480. It might be at 640x480 but I did not have a lower resolution to test it with.

I still can't decide whether 36k is valid or not for 3dmark2001, but you can't shoot it out of the sky with a cpu capped nature and dragothic theory.

Cheers

Andy

 

overclock your gpu leaving cpu at default clock, then overclock the cpu with gpu at default clocks and run both tests at 1600x1200

it would probably be easier on an amd since you have access to lower multipliers without having to factor in the effects of the fsb

 

While we are on CPU limitations. Why is Comanche 4 so CPU limited? I ran some benches at 640,1024, and 1280 and I got around 38-39 avg FPS for all three tests? Why? When I turned AA at 1280 I got around 35 fps.

 

Flight sims are frequently more CPU-limited than FPS's. Why? Well, I guess it's either the AI, the flight physics, or both.

Why AI? Well, I can imagine how it'd be more complex to make an AI that believably flies aircraft, and knows how to actually get from point A to point B at the same time.

Why physics? Well, the flight models can be very complex, for a good flight sim. The amount of lift and drag on the aircraft can change dramatically depending upon altitude, airspeed, angle of attack, etc. I'm sure helicopters are similarly complex.

 

Commache 4 is not really a sim as such and I doubt its physics are the same as something like Lock On or Falcon 4.

It is possible Commanche 4 was poorly programmed..

 

Hey Vegetto! Glad to see you came back for all the fun!

The next 48 hours are going to be a BLAST!!!!

 

I'm so excited I just peed my pants!

No, but really. I'm excited in the way only geeks can get excited over the launch of a new GPU architecture 8)