I'm wondering if anyone has seen this one. I've seen it on just about every 3D card I've owned, and the new 5900 is no exception. I'm trying to figure out if it's a driver bug, a common game bug, or what. (I'm up to date on all my drivers by the way.)

Let's say you've got a scene with a lot of "mist" or "fog" effect to it. In other words, distant objects recede into the fog until there's nothing visible in the distance. The more distant the object, the more it is overlaid by the color of the mist.

Sometimes there will be one object in the scene that doesn't "obey the fog". If it is a distant object, it can be seen clearly through the fog even at a great distance. If the graphics engine was using the fog to reduce polygon counts for distant objects, you will often see this "rogue object" pop into existence as you approach it, instead of fade into place with the other objects.

Examples where I've seen this:

- The later levels in Unreal 2 where you're onboard a V'Ger-style cyborg planet. Some of the wall panels have this problem.

- In the 3DMark2003 "Nature" demo. There is a distant set of polygons representing mountains or something, and instead of being fogged, they are sharp and clear. Also, when the scene fades in and fades out (to black) the distant mountains stay fully lit during the fade, allowing them to be silhouetted against the black trees.

Doesn't happen very often, it's pretty rare. But when you see it, you'll recognize it. Anyone else seen this?

You're a real stickler for fog and mist, aren't you?

Or right. By the way, the new Nvidia card doesn't seem to exhibit those slowdown symptoms that I was noticing with those sprite effects on the old cards.

The problem I'm talking about is related to a different effect. I'm talking about whole-scene fog as opposed to a single translucent smoke/dust sprite as I was in the other thread.

For reference, here is a picture of what I'm talking about.

Heh... your example looks like it's part of the background image... hence it won't be affected by vertex fog.

However... this does sometimes occur with world geometry too. Especially if it's translucent in any way. Translucent polygons have never sorted properly, and a lot of times if you get two translucent polys drawn on top of another, the back poly will appear to draw in front of the closest one.

I've bitched and moaned to many a programmer about this... It appears to be a glitch that isn't going to be addressed by hardware anytime in the near future.

EDIT: By the way... did you ever figure out your last problem referenced above? Last time I tried to play Unreal2 I was reminded of your problem, but could not reproduce it...

Heh... your example looks like it's part of the background image

It's not, it's part of the world geometry. If I'd shown an example in Unreal 2, you would have been able to see it better.

However... this does sometimes occur with world geometry too. Especially if it's translucent in any way.

I'm reasonably certain the polys in question were not translucent. In fact, there are many translucent polys in Unreal and I think in that nature scene, too (the refractive water) which sort just fine.

It appears to be a glitch that isn't going to be addressed by hardware anytime in the near future.

I think it turned out to be a glitch in software. I posted on the Futuremark forums and they suggested backing off a version of the Nvidia drivers. They were right. 44.03 does not have the bug, 45.23 does.

Of course, the older version of the drivers is a lot slower, resulting in lower frame rates and a much lower 3DMark score. *SIGH*...

EDIT: By the way... did you ever figure out your last problem referenced above? Last time I tried to play Unreal2 I was reminded of your problem, but could not reproduce it...

Nope. All I know is that the new card and the new drivers don't have that problem at all. So as far as I'm concerned... problem solved.

I think it turned out to be a glitch in software. I posted on the Futuremark forums and they suggested backing off a version of the Nvidia drivers. They were right. 44.03 does not have the bug, 45.23 does.

Of course, the older version of the drivers is a lot slower, resulting in lower frame rates and a much lower 3DMark score. *SIGH*...

Again, dosen't really suprise me. Think of all the processor time the GPU was saving in the new drivers by rendering things wrong. NVidia claimed it was a bug that their drivers were rendering nothing beyond the camera in 3DMark, so, I'm guessing this is a similar "bug". Of course now Futuremark won't call out these cheats anymore, since NVidia is back in the beta program. This also means it will be harder for anyone else to detect these cheets.

3DMark is just a big marketing number, and should be treated as such.

Yeah, but it's cool to watch nonetheless.

What kind of scores do you get in GL Excess? I'm thinking I'm going to need to get another video card around the time HL2 comes out. My FX 5600 is already showing its age.

Oh, cool, another 3D demo program. I'll download it and try it tonight, thanks!

What kind of scores do you get in GL Excess?

The better score is without antialiasing, the worse score is with quincunx antialiasing accidentally turned on. 1024x768 pixels, 32 bit color.

Just for comparison:



Yellow = FILL RATE Tests
Green = POLYGON COUNT Tests
Blue = VRAM Tests
Red = CPU/FPU Tests

Well hell, my computer sucks.

Cool!

Not as badly as mine:

Which is why I think you should get a decent Shuttle with a Radeon:





pca

Here's my office machine. Not bad for something that will never play a game.

Oops. I ran the test in 16-bit the first time. Here's the 32-bit results.

Based on Tom's scores and yours, I see that the ATI card beats the Nvidia card pretty handily on fill rate.

I'm still a little fuzzy on exactly what "fill rate" is, and how important it is for game performance. Anyone want to discuss that topic and clarify it?

This is dated, but it still gives a good idea of what fill-rate is. See here. It's PowerVR biased, but if you ignore all of that it still has good technical data.

Oh, and just for the hell of it, this chart gives a nice comparison. It reflects what we're seeing here, although the FX 5900 is missing from the picture as it wasn't out yet.

Hey Tony...

I came back to the office for a few minutes (after watching SWAT ) before heading out to dinner... I just happened to ask one of our gurus and he had this to say: "It's because his card sucks ass" Oh well...

Tony, sell it and I'll get you my employee discount on a RADEON 9800.

Bruno

Why should Tony get your discount? Here I am, a loyal ATI fan, and I have a lowly Radeon 9600! It's not fair I tells ya!

"It's because his card sucks ass"

Heh. Yeah, competition is a lovely thing, isn't it?

Nope. All I know is that the new card and the new drivers don't have that problem at all. So as far as I'm concerned... problem solved.

Turns out this problem is NOT solved, I'll go back to that other thread and comment on my observations in more detail.

Ah. That's better:

Isn't that interesting. Your fill rate is somewhat higher than on my system, yet your polygon rate is considerably lower. I wonder why?

pca

Just loaded the latest drivers

Not a bad score for a system that hasn't been optimized for gaming in any way.

I guess it would be better if I ran the test at 32bit and 1024x768 like everyone else.

I guess my system needs some updating to be able to run the latest and greatest games...?



Stig

haha, intel based what??

Might be that GL Excess didn't like my dual PIII 1GHz setup....

Stig

You just need to run the Detonator 50 drivers. The quality is a bit lower, but the frame rates much increased. Demonstrated in the attached screen capture of the Dawn demo.


Bruno

Juvenile, but funny.

Yeah, that made me laugh out loud.

Interestingly, the real Dawn demo refuses to run on my system. I think perhaps because it's not Windows XP. Even though their little SDK (cg.dll) doesn't say anything about XP being a requirement, I get a page fault in cg.dll when trying to run any of their recent demos.

How's this for bad timing. Damn you Valve. HL2 is 3 days late.

My new PC at work:

Heh, I hadn't thought about this test for my work machine:


(sorry about the colors. I only have Paint on this PC. it's not very good with GIFs)

New system. This one has the advantage over the Shuttle of just fitting into a backpack and including a screen. It's a Dell plastic brick, err, Inspiron 9100.

This one is from my brand new home system, Falcon Northwest Fragbox 2.

Is my new system running too slow ?




I didn't expect the X300SE to be the fastest 3D card in the world, but I am surprised that my system seems to be very slow in the Doom 3 demo and the benchmarks that I have tried.

The x300/x600 both use a newer variant of the chips used on the Radeon 9600 line. With the x300, it is basicially a PCI-e version of the Radeon 9550, an oddly slower card released by ATI recently.

It seems in the video card arena, $200 is still around the price point you need to pay to be able to play the more modern games. The PCIe cards that fit that range would be the NVidia 6600 GT or the Radeon x700. Benchmarks ont hese cards against the other PCIe boards is here. It seems very few sites even looked at the x300 much.

Still, there must be something not right. I've got a 9700 AIW but a much slower system than he does, and my score is nearly three times as high. I would have thought the X300 would at the very least be better than my card, and that coupled with a better system it would give him far better scores than I got.

Nope. The 9700(R300) is a very good chip for what it cost and wasn't designed to do much beyond push pixels very quickly. The heat and power pull was the highest ATi shipped at the time, but was still a one slot solution. The x300 (RV370) is on the other hand a chip built more for low power/heat mobile use (most RV products turn into mobile solutions as well).

The low end of a new product line is rarley better then what the near high end was in the last generation. For example, it took on average a Radeon 9500 or above to beat the Radeon 8500. In most cases, the 9000/9100/9200 cards were noticibly slower then the 8500.

Hey, good to know! I don't like the AIW part of my card, so if I upgrade sometime soon, I may get the non-AIW 9700.

Ah, that's even better:

Showoff.

Close, but not quite... (Dual 6800GT's in SLI)

OK, how does one get these results? When I run the demo, it just runs the demo. I don't see how to get the timings.

Don't run the demo. Run the benchmark.