A promise is a promise :) there you go!

My problem with these modes is text. I can't stand blurry text. While it's not always easy to notice the blurred textures in a "normal" 3D scenario, when text appears, it will be blurred. I play far too many RPG's to use a mode like this, not to mention I actually like to talk to people on occasion in games like UT2k4.

 

I wasn't thinking only of storage problems, but the detection of intersections, which should be easier in my mind if you have all or most data supplied prior to rendering. Additionally aren't usually edge AA algorithms using back to front routines or am I shooting myself in the foot here?

As for as two pass mechanisms goes, I thought that coverage mask techniques are in essence two pass mechanisms (sort of anyway). A clever combination/modification of fragment and coverage mask AA should do the job.

 

On a MR9700 I'd expect that to be the case, given the bandwidth available - given whats being discussed here, how many multiples of fill-rate / bandwidth are we talking about?

 

Yes I agree with you there, in Everquest I refuse to use it, Where I do alot of reading (Obviously Ingame Chat)

But I am usually not bothered by it in alot of games. Need for Speed Underground is a great game for 4x9tap Quincunx Anti Aliasing

 

Here's something I was wondering.

Does resolution make a difference to the cpu's workload? In otherwords, does the cpu have the same workload when a game is running at 640x480 as when it's running at 1600x1200?

 

The workload per-frame remains consistent for the CPU over resolution if you are not bound by the fill-rate / bandwidth of the 3D board.

 

Sure, but
- the PC was powered by an athlon that gives you much more raw power under UT
- 1600*1200 is 45% more demanding than 1280*1024
So I wouldn't say it's not the case at 1600*1200 4AA + 8AF. On the bench @X-bit, even Convoy, which seems to be the more demanding, is GPU limited at those settings.


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

 

A.) We're still not seeing any UT2004 Botmatches above 60FPS here.
B.) Are you suggesting that you don't think NV40 will have twice the performance of 9800 PRO?

 

I agree. I just said that UT 2k4 is GPU limited at 1600*1200 and AF/AA enable. Do you agree?

Where did i say that? But yes, i don't think i will see the NV40 being 2 times faster in almost every bench. Perhaps in some really stressing games at 1600*1200 + AA/AF enable.
Edit: What do you think?

 

What kind of CPU are you using ?

 

Demalion asked some quite pertinent points in this post which were probably simply overlooked.
The main point IMO is:
If UT2004 is using the CPU for skinning, does a higher resolution result in a higher CPU workload because of a higher LoD for characters/objects?

 

I was under the impression that increasing the frame resolution did not increase the number of vertices in a character/model; if so, then a higher res should not increase the amount of vertex blending needed.

 

Increasing res can increase vertex count if the application bases its LOD calculations on a measure of 'pixels per polygon' - i.e. in screen space - rather than some game-world space measure. High resolutions will then result in more vertices.

I haven't seen an application that does this myself, but it's a reasonably limited set I've looked at.

 

So if there are features and characters just outside your field of view at 640x480 that would be in your view at 1600x1200, that doesn't increase the load on the cpu at 1600x1200? So the whole level is there whirling away madly at the same burn rate of cpu cycles whether I can see it or not at any given moment? Yikes, what a load of wasted cpu cycles!

 

No, that's not what he's saying. FOV is independent of resolution (given the same aspect ratio, anyway, as is the case here.)

He's saying that if you have a game that does some dynamic level of detail calculations to "calculate" how many polgons should be drawn for a given object, it's possible that said calculation can be based on resolution. Higher resolution can equal more and smaller pollies....but again, only if the LOD system is designed this way.

 

Someone that has a dual opteron and a 9800xt might wnat to do this . After all the ut2k4 and 3dmark scores for the nv40 were done on dual opterons

 

I'll be taking donations...

 

This is something that UT2003 definitely can do (as was shown by tech demos early on), and you'd think that it would be smart to scale the LOD by resolution, but I'm not really sure what the algorithm used is.

 

Why do you have this habit of ignoring part of my post in your replies? Look at Convoy, where the 5700 scores are similar to the 5900 without FSAA, and double up on the 5600 Ultra. Sure, the 9800's are limited by the CPU, but the others aren't. Look at the framerate-resolution relationship. It's flatter than the pixel count change. Look at this Stalker benchmark, and you see similar patterns.

Finally, frames of nearly every game have some regions with a high vertex to pixel ratio, making the framerate-pixelcount relationship non-constant with any CPU.

Don't push T&L in the least? That's quite an exaggeration. In fact, it seems like developers are pushing that more than pixel shaders because it's easy. That's why the 5700 is a good choice for today's games and some of tomorrow's.