Major GeForce 6600 GT performance issues
by Tom Spilman · in Torque Game Engine Advanced · 05/24/2005 (1:53 pm) · 45 replies
So we have a map that performs beautifully on an ATI 9800 Pro, but is dog slow on my GeForce 6600 GT. I'm getting 2-3 fps as reported by metrics(fps) and NVPerfHud while on the 9800 Pro we get 80-100fps.
So i started by trying to force down the pixel shader level. I tried forcing it to 2.0, 1.4. 1.1... all of them seem to give me the same performance. I did this by changing the prefs to:
Which seems to work as the console reports it's forcing shaders down... but i can't be sure it's not lying to me. =)
Next i fired off NVPerfHud. It shows a tremendous amount of time is being spent in the driver. When i use the T key to force 2x2 textures performance comes up to a respectable frame rate. So according to NVidia's docs that means i'm limited by texture bandwidth. Now i do have alot of large textures in the scene... but the 9800 seems not to mind.
So the 6600 which is supposed to be a better performing card than the 9800 Pro has a texture bandwidth issue while the 9800 does not. That seems illogical to me.
I suspect one of the shaders is the culprit here. Anyone have any tips to help diagnose this problem?
So i started by trying to force down the pixel shader level. I tried forcing it to 2.0, 1.4. 1.1... all of them seem to give me the same performance. I did this by changing the prefs to:
$pref::Video::forcedPixVersion = 1.1; $pref::Video::forcePixVersion = 1;
Which seems to work as the console reports it's forcing shaders down... but i can't be sure it's not lying to me. =)
Next i fired off NVPerfHud. It shows a tremendous amount of time is being spent in the driver. When i use the T key to force 2x2 textures performance comes up to a respectable frame rate. So according to NVidia's docs that means i'm limited by texture bandwidth. Now i do have alot of large textures in the scene... but the 9800 seems not to mind.
So the 6600 which is supposed to be a better performing card than the 9800 Pro has a texture bandwidth issue while the 9800 does not. That seems illogical to me.
I suspect one of the shaders is the culprit here. Anyone have any tips to help diagnose this problem?
About the author
Tom is a programmer and co-owner of Sickhead Games, LLC.
#22
Here's what I discovered using my CVS build of TSE and the 66.93 drivers:
800x600x32: 22-30FPS, 8FPS while the portal passes through LOS
1280x1024x32: 27-30FPS(!), Demo crashes to desktop with no error while portal passes through LOS (yik!)
Using the demo on the TSE store page and the 66.93 drivers:
800x600x32: 127-139FPS, Demo crashes to desktop with no error while portal passes through LOS (yik!)
1280x1024x32: 65-86FPS, Demo crashes to desktop with no error while portal passes through LOS (yik!)
I think that we should just chalk this up to life on the cutting edge, but yes, there is something in the latest build that murders framerates (gee, Atlas?). What is impressive are the framerates at high resolutions. For the CVS build it was consistant, if not slightly better. For the precompiled demo it is downright dazzling.
Watching TSE evolve is like watching an extended pregnancy. We're at about the second trimester. You're used to the aches and pains by now, but you still want that baby to come out SOOOOOOOOOO bad so it will all be over and we can get on with our lives but its just not done yet.
Pickles and icecream anyone?
EDIT: I ran the terrain_water_demo using the older (66.93) drivers and saw an increase of 15FPS at both resolutions. Since the precompiled demo does not have Atlas terrain code there's no comparision, but the older drivers do seem to perform better than the newer drivers, although overall performance is somewhat decent (50FPS for large landscapes in LOS, 100+FPS for items solo, 60-70FPS for items and terrain in LOS).
06/16/2005 (8:59 am)
This is a worthy task so I've downgraded my drivers to the earliest version I could find online. That ended up being the OmegaDrivers version of 66.93 installed to optimize performance.Here's what I discovered using my CVS build of TSE and the 66.93 drivers:
800x600x32: 22-30FPS, 8FPS while the portal passes through LOS
1280x1024x32: 27-30FPS(!), Demo crashes to desktop with no error while portal passes through LOS (yik!)
Using the demo on the TSE store page and the 66.93 drivers:
800x600x32: 127-139FPS, Demo crashes to desktop with no error while portal passes through LOS (yik!)
1280x1024x32: 65-86FPS, Demo crashes to desktop with no error while portal passes through LOS (yik!)
I think that we should just chalk this up to life on the cutting edge, but yes, there is something in the latest build that murders framerates (gee, Atlas?). What is impressive are the framerates at high resolutions. For the CVS build it was consistant, if not slightly better. For the precompiled demo it is downright dazzling.
Watching TSE evolve is like watching an extended pregnancy. We're at about the second trimester. You're used to the aches and pains by now, but you still want that baby to come out SOOOOOOOOOO bad so it will all be over and we can get on with our lives but its just not done yet.
Pickles and icecream anyone?
EDIT: I ran the terrain_water_demo using the older (66.93) drivers and saw an increase of 15FPS at both resolutions. Since the precompiled demo does not have Atlas terrain code there's no comparision, but the older drivers do seem to perform better than the newer drivers, although overall performance is somewhat decent (50FPS for large landscapes in LOS, 100+FPS for items solo, 60-70FPS for items and terrain in LOS).
#23
Do you have a particularly fragmented or slow hard drive? How much RAM do you have? What output do you get from the profiler?
06/16/2005 (10:28 am)
I'm pretty sure that it's something strange and broken in your setup. The terrain code is _fast_ - 100fps+ on a Radeon 9600 mobile, and the original version, on a GeForce 2, still got a better FPS than that. And there's not much else in the scene, either.Do you have a particularly fragmented or slow hard drive? How much RAM do you have? What output do you get from the profiler?
#24
06/16/2005 (10:51 am)
@Bryce: not trying to downgrade your specific issue or anything, but as Ben mentioned, Atlas is FAST--in fact, it is faster than even the old TGE terrain, much less the last set of terrain in TSE milestone 1. As Ben mentioned, there has to be something unusual/unique about your environment--hopefully we can help you sort it out!
#25
Not quite sure what happened, but wow... This is what I was expecting to see. :)
06/16/2005 (10:46 pm)
I kept playing with things and once I started to add my own Atlas terrain.... the FPS jumped up to 150-200FPS in terrain_water_demo, and never dipped below 85.Not quite sure what happened, but wow... This is what I was expecting to see. :)
#26
I'm getting back into TSE development with our game and my 6600 GT is still a dog. I've merged in the latest head release and did some more tests...
P4 2.8 GHz w/ 1 GB Ram and a 6600 GT ..... 7fps
P4 2.6 GHz w/ 768 MB Ram and a ATI 9800 Pro ..... 90fps
This has me seriously bummed as the 9800 runs freaking awesome with a crap load of texture detail just fine while the 6600 hovels along. There is nothing but a large interior, 3-4 simple DTS shapes, and the Orc. No terrain of any kind in the mission.
I tried DDS'ing my biggest textures and that seemed to make little improvement, but i was still in single digits. So i have to think that the Brian's batching theory is the culprit. I stepped thru the code and yes there is a good amount of material changes going on, but it seemed less than a dozen or so when rendering the interior. I'm going to look into this later tonight as a last ditch effort before going to buy an ATI card for development.
It's hard to believe that the 6600 GT would have such a higher penalty for texture state changes.
08/17/2005 (4:18 pm)
Bumping my old thread... blasphemy!I'm getting back into TSE development with our game and my 6600 GT is still a dog. I've merged in the latest head release and did some more tests...
P4 2.8 GHz w/ 1 GB Ram and a 6600 GT ..... 7fps
P4 2.6 GHz w/ 768 MB Ram and a ATI 9800 Pro ..... 90fps
This has me seriously bummed as the 9800 runs freaking awesome with a crap load of texture detail just fine while the 6600 hovels along. There is nothing but a large interior, 3-4 simple DTS shapes, and the Orc. No terrain of any kind in the mission.
I tried DDS'ing my biggest textures and that seemed to make little improvement, but i was still in single digits. So i have to think that the Brian's batching theory is the culprit. I stepped thru the code and yes there is a good amount of material changes going on, but it seemed less than a dozen or so when rendering the interior. I'm going to look into this later tonight as a last ditch effort before going to buy an ATI card for development.
It's hard to believe that the 6600 GT would have such a higher penalty for texture state changes.
#27
I had these issues with several NVIDIA cards (GF 3, GF FX 5600, GF 6800)
I reduced it to special buildings in our level. When i removed them framerate went up. So its not all interiors that will produce that.
You can verify that in the TSE Demo by removing the Large Building.
There is still a Tower that from the TGE demo and this tower doesnt have this issues. I added a few more of these towers to the scene and none of them seemed to impact framerate at all.
08/18/2005 (1:30 am)
Guess i will join in blasphemy ;)I had these issues with several NVIDIA cards (GF 3, GF FX 5600, GF 6800)
I reduced it to special buildings in our level. When i removed them framerate went up. So its not all interiors that will produce that.
You can verify that in the TSE Demo by removing the Large Building.
There is still a Tower that from the TGE demo and this tower doesnt have this issues. I added a few more of these towers to the scene and none of them seemed to impact framerate at all.
#28
08/18/2005 (1:57 am)
I was thinking of getting a 6800 very soon, but now I'm not so sure... 
#29
TSE for ATI only ... I think not :)
08/18/2005 (4:00 am)
Well I can't believe this won't be resolved !TSE for ATI only ... I think not :)
#30
I do however use a 6800 now and there are no problems.
08/18/2005 (4:37 am)
I also had these issues with my GF4, 5200 and 6600.I do however use a 6800 now and there are no problems.
#31
inside and outside of the Interior i get about 110 and 230 fps and when the portal is in view it drops to 40-60 fps
08/18/2005 (10:16 am)
With my 6800 when i face the portal from the inside of an interior to terrain the engine stutters for a little and then performance gets up a little again. inside and outside of the Interior i get about 110 and 230 fps and when the portal is in view it drops to 40-60 fps
#32
I know there are also translucency issues and that the Orc is getting drawn multiple times because of it, plus the glow pass, so that's all quite bad. The skinning code is also bad related to the Orc.
Not sure about interiors, didn't think there were issues there.
08/18/2005 (1:38 pm)
This will get resolved, we're not quite at the performance point of tuning MS3. Ben and I have talked about it and experimented a tiny bit. The current theory is that something bad may be happening in the blending of the old terrain code on 6600 cards.I know there are also translucency issues and that the Orc is getting drawn multiple times because of it, plus the glow pass, so that's all quite bad. The skinning code is also bad related to the Orc.
Not sure about interiors, didn't think there were issues there.
#33
Yesterday i found that if i DDS a certain amout of my textures the 6600 GT goes instantly from 9fps to 120fps (no gradual improvement). I sort of gave up and decided that the 6600 GT just sucks on texture heavy scenes.
Today i'm out of town on my P4 2.3GHz laptop with a ATI Mobility 9000 64MB (shared system memory). For kicks i thought i'd try it on here. It had to scale back to 1.4 shaders (no per-pixel specular), but the Mobility 9000 smoked the 6600 GT at a little over 40fps. That seems highly unlikely to me.
I also found that the 9000 suffered from that weird input lag where your card buffers up too many frames. There is some framebuffer locking trick that fixes that. I'll look into it.
08/20/2005 (12:29 pm)
Everytime i convince myself that i'm just using too much texture memory, i come across another test that convinces me there is a bug.Yesterday i found that if i DDS a certain amout of my textures the 6600 GT goes instantly from 9fps to 120fps (no gradual improvement). I sort of gave up and decided that the 6600 GT just sucks on texture heavy scenes.
Today i'm out of town on my P4 2.3GHz laptop with a ATI Mobility 9000 64MB (shared system memory). For kicks i thought i'd try it on here. It had to scale back to 1.4 shaders (no per-pixel specular), but the Mobility 9000 smoked the 6600 GT at a little over 40fps. That seems highly unlikely to me.
I also found that the 9000 suffered from that weird input lag where your card buffers up too many frames. There is some framebuffer locking trick that fixes that. I'll look into it.
#34
Something got chopped off.... Sorry to hear it.. I blame "best buy" (humor)..
08/21/2005 (12:18 pm)
Yea it could just be something in the card's GPU. I think it's like 1/3 the size of the 6800.Something got chopped off.... Sorry to hear it.. I blame "best buy" (humor)..
#35
08/21/2005 (12:57 pm)
My GeForce 6800 run with 9 FPS with Terrain "legacy" Engine and 21 FPS With terrain "Atlas" Engine(on starter.fps)... Don't speak of my lapTop with a GeForce 5250 :(
#36
08/22/2005 (12:50 pm)
Interesting, because my GeForce 6800 was getting 200+ fps with the original demo. I haven't had a chance to do any manual messing around, but I was extremely happy with the 6800 performance...
#37
i get the slow framerates when i compile it here on my computer
08/22/2005 (2:05 pm)
The precompiled demo on the GG website runs fluid here too...i get the slow framerates when i compile it here on my computer
#38
08/23/2005 (2:38 pm)
I confirm fluid running with original GG Demo. Framrates I gave was with a DEBUG build. So Any issue? 
#40
Thanks.
08/27/2007 (9:17 am)
Welcome back to an old thread! Not to wake a sleeping dog here, but we're experiencing a problem today with the current release of TGEA. Our current environment runs fine on most cards in the office, including as far down as the GeForce Go 7800 in a laptop, but our frame rates are absurdly slow running the same environment specifically on the Quadro FX 550, which should be a better card. Is there just no support for the Quadro's specific GPU in TGEA? We can't seem to figure out any other reason for this.Thanks.
Torque 3D Owner Stephen Zepp
Now, please keep in mind that I am talking about the publically downloaded demo--I don't actually have the latest TSE on this machine, and won't for the next month or so, so it is possible that something recent isn't working well, and wouldn't work on my set as well. Try downloading the current demo from the web page here (under Torque Products), and see if it gives you the same effect as your compile of the latest CVS?