Horrible performance on GeForce 6800

#41 Unlikely.

#42

Quote:The framerate only slows down when I am viewing the orkbase structure. If I move the camera so it is out of view, the framerate is fine.

Just read that, the same happens to me. When this happens though the processor fan goes really loud so maybe it's too much for my processor to handle. The performance of a game is limited more by the processor than the graphics card (i read). I am running a 6200 but only have a celeron 2.66 ghz. I will be upgrading to the new pentium 4 extreme edition with dual core, ht and 64 soon though (as soon as I have the money) so I will soon find out if its the processor, tse or graphics card.

#43 Mmm, skinning? That would be the most CPU intensive thing I could think of that would only happen when looking at a character.

#44 In your case James, I'd guess the 6200 is the bigger problem with performance than your CPU. I've run the TSE demo just fine with a Radeon 9700 and a 1.4 ghz Athlon.

#45 I just bought a Geforce 6600 GT to replace my ATI 9800 which died on me. The 6600 drops into single digits in the TSE demo when the skinned "space orc" character is on screen.... once it's offscreen the framerate jumps back up to normal. On the 9800 which is a lower performance card having him onscreen or off was not noticeable and the framerate was always excellent. I have to say there is definitely some sort of issue here with GF6 series cards.

@Ben - Could you download the old TSE demo from the site and run that on your 6600?

#46 Now looking at the Car demo it experiences the "purple junk" others have reported. The dojo demo, which seems to have the same skinned orc, performs just fine.

So *maybe* it is a fixed bug between the release of the TSE and Dojo demos. I need to get that particular orc into our prototype and report the results.

#47 Actually I have been doing some testing and it the frame rate drops to 1 when looking at the entrance hall in tghe ork base structure, when in the hallway's the fps goes way back up.

#48 We'll have to do another pass on the demo with the 6600 when we get MS2 out.

#49 Runs great for me on a mobile 9600 and a 1.6 Ghz cpu.

#50 Runs UBER fast on BFG 7800GTX OC and AMD 64-bit 3.7Ghz, no issues

#51 I have noticed the same problem Running the 6800 here too.

The lag is when viewing the ORC character.

If you move the camera so that it is looking anywhere but at the orc character it runs smooth as can be.

But if you look toward the orc character (in the demo i'm talking about) it will lag crazy as the camera moves.

#52 Orc does cause a framerate drop. Skinning code isn't tweaked as well as it should be yet.

#53 You could try removing the glow from the weapon, that brought performance back up for me when i used a Geforce 3

#54 I'm pretty sure the Orc uses like 4 textures for differnt shader effects on each section. This is probably why performance drops while viewing him.

#55 Hmm, that's probably a contributing factor, but I think the main hit is the skinning.

#56 Actually Ben. Once i DTX'd some textures in my scene... including the textures on the Space Orc... the slowdowns stopped.

#57 Hmm. Strange. Maybe you were exceeding available texture memory?

#58 According to the Nvperf tool's memory usage meter i was not, but all other tests seemed to suggest it was a fillrate issue. What was really strange is that an ATI 9800 128MB (same memory) ran it at 90fps. So two possiblilities.... the nvperf tool was reporting the wrong memory usage and the 6600 has a much bigger performance penalty for AGP usage than the 9800..... or there is some other bug i've not nailed down yet.

#59 I dont think that ATI's cards are really that much faster than NVIDIA's.

My guess would be that its a drivers thing. Since each vendor writes their own implementation of the connection between DirectX and the GPU hardware it could be that TSE sends something NVIDIA drivers dont expect and thus arent optimized for such a case. I noticed that during those slowdowns that i had an increased amount of time that was spent inside the driver and most samples were made in a function called GetIAtomString (57%). Additionally i registered lots of cache misses.

#60 Even weirder. Maybe it's trying to report some error/warning condition?