Page 2 of 6

Re: 2023-02-06 snapshot

Posted: Fri Feb 10, 2023 2:39 am
by Jakob Hubbard
jlv wrote: Fri Feb 10, 2023 2:14 am I think disabling the Intel graphics in device manager will guarantee it isn't used.
I don't even see my integrated graphics showing up in my display adapters menu in the device manager, it only shows my dedicated GPU. Could be disabled in the BIOs, haven't checked. But I don't recall changing any BIOs settings to disable it so it could've disabled it by default. Anyways that's besides the point, it should be force running my gpu anyways.

Re: 2023-02-06 snapshot

Posted: Fri Feb 10, 2023 3:44 am
by jlv
Jakob Hubbard wrote: Fri Feb 10, 2023 2:39 am I don't even see my integrated graphics showing up in my display adapters menu in the device manager, it only shows my dedicated GPU. Could be disabled in the BIOs, haven't checked. But I don't recall changing any BIOs settings to disable it so it could've disabled it by default. Anyways that's besides the point, it should be force running my gpu anyways.
If you can disable it in the BIOS give that a try. Since turning down the resolution sped it up it's fill rate limited. It's really suspicious that a fast GPU like a 2070 is fill rate limited before a 2400g. It should be 5-10 times faster but it's 4 times slower instead.

Re: 2023-02-06 snapshot

Posted: Sat Feb 11, 2023 1:00 am
by Jakob Hubbard
jlv wrote: Fri Feb 10, 2023 3:44 am If you can disable it in the BIOS give that a try. Since turning down the resolution sped it up it's fill rate limited. It's really suspicious that a fast GPU like a 2070 is fill rate limited before a 2400g. It should be 5-10 times faster but it's 4 times slower instead.
I went into my BIOS and went into the integrated graphics config and the initiate graphic adapter setting was set to PCIE Graphics and then the setting below it was allowing for integrated graphics to be turned on and that setting was also disabled. So integrated graphics do not run on boot up of my system at all and are completely disabled. So it is my rtx 2070 that's being handled by the game.

Re: 2023-02-06 snapshot

Posted: Sat Feb 11, 2023 2:13 am
by jlv
Jakob Hubbard wrote: Sat Feb 11, 2023 1:00 am I went into my BIOS and went into the integrated graphics config and the initiate graphic adapter setting was set to PCIE Graphics and then the setting below it was allowing for integrated graphics to be turned on and that setting was also disabled. So integrated graphics do not run on boot up of my system at all and are completely disabled. So it is my rtx 2070 that's being handled by the game.
I wonder if it's not boosting correctly. Maybe have a look with GPU-Z to see if it's stuck at 200 MHz or something.

Now I'm thinking about building a GPU benchmark into the game. Do something demanding like render a Mandelbrot set and show a score so the user can tell if the GPU is working as expected.

Re: 2023-02-06 snapshot

Posted: Sat Feb 11, 2023 2:21 am
by Jakob Hubbard
jlv wrote: Sat Feb 11, 2023 2:13 am I wonder if it's not boosting correctly. Maybe have a look with GPU-Z to see if it's stuck at 200 MHz or something.
Looking at my evga precision program it's running at an expected 1605MHz, it can boost up to 1800MHz if needed.
jlv wrote: Sat Feb 11, 2023 2:13 am Now I'm thinking about building a GPU benchmark into the game. Do something demanding like render a Mandelbrot set and show a score so the user can tell if the GPU is working as expected.
Would be helpful for some gpu debugging.

Re: 2023-02-06 snapshot

Posted: Sat Feb 11, 2023 3:03 am
by jlv
Jakob Hubbard wrote: Sat Feb 11, 2023 2:21 am Looking at my evga precision program it's running at an expected 1605MHz, it can boost up to 1800MHz if needed.
It shouldn't idle that high. The fan would be running full speed all day. Should just be a few hundred MHz at idle.

Re: 2023-02-06 snapshot

Posted: Sat Feb 11, 2023 3:08 am
by Jakob Hubbard
jlv wrote: Sat Feb 11, 2023 3:03 am It shouldn't idle that high. The fan would be running full speed all day. Should just be a few hundred MHz at idle.
Sorry I didn't clarify, this isn't my idle speed. It's my speed running programs.

Re: 2023-02-06 snapshot

Posted: Sat Feb 11, 2023 6:49 am
by Jakob Hubbard
Did some more testing on this snapshot vs the previous snapshot on times, and the times seem pretty similar. All tests were done on 2023 Tampa at medium res. I also had my friend test on the same snapshot and track with the same resolution and game settings and his are posted at the bottom.

Snapshot 2-6-23 at 4096x4096:

Code: Select all

2347 0.008
2 0.009
5 0.014
3 0.015
2 0.016
1 0.017
2 0.018
1 0.019
3 0.020
1 0.021
1 0.025
2 0.027
1 0.075
2 0.076
1 0.080
1 0.081
3 0.089
1 0.093
1 0.091
Snapshot 12-23-22 at 4096x4096:

Code: Select all

1245 0.008
5 0.009
1 0.011
1 0.013
2 0.014
5 0.015
2 0.016
2 0.017
2 0.018
4 0.019
1 0.020
1 0.026
1 0.028
1 0.054
1 0.085
1 0.086
1 0.091
1 0.094
1 0.095
2 0.096
1 0.104
1 0.109
Snapshot 2-6-23 at 2048x2048:

Code: Select all

3002 0.008
12 0.009
2 0.015
2 0.016
2 0.017
3 0.018
2 0.019
5 0.020
3 0.021
4 0.022
1 0.023
1 0.025
1 0.042
1 0.055
Snapshot 12-23-22 at 2048x2048:

Code: Select all

3631 0.008
7 0.009
1 0.010
1 0.014
1 0.015
4 0.019
1 0.021
1 0.022
1 0.023
1 0.024
1 0.025
1 0.027
2 0.028
1 0.038
2 0.039
1 0.042
1 0.052
Friend's test on same track with Snapshot 2-6-23 at 4096x4096 medium res on an rtx 3070ti and i7-12700kf

Code: Select all

3862 0.008
94 0.009
1 0.039
1 0.063
1 0.064
1 0.065
1 0.066
2 0.067
1 0.069
1 0.071
1 0.072

Re: 2023-02-06 snapshot

Posted: Sat Feb 11, 2023 8:48 am
by Freddy694
Jakob Hubbard wrote: Sat Feb 11, 2023 6:49 am Did some more testing on this snapshot vs the previous snapshot on times, and the times seem pretty similar. All tests were done on 2023 Tampa at medium res. I also had my friend test on the same snapshot and track with the same resolution and game settings and his are posted at the bottom.
Just a quick question, I don’t know how much of a difference this makes but were you guys on similar skins or stock skins for this testing? Would the change between 2K/4K skins and different poly count for boots/bikes/wheels/helmets etc make a big impact for this too?

Re: 2023-02-06 snapshot

Posted: Sat Feb 11, 2023 10:15 am
by Jakob Hubbard
Freddy694 wrote: Sat Feb 11, 2023 8:48 am Just a quick question, I don’t know how much of a difference this makes but were you guys on similar skins or stock skins for this testing? Would the change between 2K/4K skins and different poly count for boots/bikes/wheels/helmets etc make a big impact for this too?
Similar skins, but this shouldn't affect the test anyways since the lag spikes are coming from the terrain engine drawing the terrain textures. But just to double check I ran another test on stock skins and yielded identical results.

Re: 2023-02-06 snapshot

Posted: Sun Feb 12, 2023 1:37 am
by jlv
Jakob Hubbard wrote: Sat Feb 11, 2023 6:49 am Did some more testing on this snapshot vs the previous snapshot on times, and the times seem pretty similar. All tests were done on 2023 Tampa at medium res. I also had my friend test on the same snapshot and track with the same resolution and game settings and his are posted at the bottom.
One thing that's weird about it is there aren't enough slow frames to account for the rendering. The incremental version does the decal rendering in 16 steps, so you'd think that there'd be 16 roughly equal slow frames per texture rendered if that was it. This version still clears the texture and generates the mipmaps in a single step. (One step for the clear and one for the mipmap.) I doubt the clear is the problem since that should be easy. That'd leave the possibility of slow mipmap generation. That would be a huge pain to make incremental since I'd have to do it with a shader that reads and writes to the same texture (albeit at different lod levels). Not even sure if that's kosher according to the OpenGL specs. I guess it'd be possible to keep a separate texture just for making mipmaps and then copy it back but man that'd be a mess.

I'll dust off my Nvidia system and see if I can reproduce it.

If you run a lap do the slow frames keep piling up?

Re: 2023-02-06 snapshot

Posted: Sun Feb 12, 2023 3:06 am
by jlv
Jakob Hubbard wrote: Sat Feb 11, 2023 6:49 am Did some more testing on this snapshot vs the previous snapshot on times, and the times seem pretty similar. All tests were done on 2023 Tampa at medium res. I also had my friend test on the same snapshot and track with the same resolution and game settings and his are posted at the bottom.
One more question - does lowering the resolution even more get rid of the slow frames?

Re: 2023-02-06 snapshot

Posted: Sun Feb 12, 2023 4:05 am
by Jakob Hubbard
jlv wrote: Sun Feb 12, 2023 1:37 am One thing that's weird about it is there aren't enough slow frames to account for the rendering. The incremental version does the decal rendering in 16 steps, so you'd think that there'd be 16 roughly equal slow frames per texture rendered if that was it. This version still clears the texture and generates the mipmaps in a single step. (One step for the clear and one for the mipmap.) I doubt the clear is the problem since that should be easy. That'd leave the possibility of slow mipmap generation.
Yeah I definitely think that it's something to do with the new tilemap shader, because the lag spikes are almost identical despite the incremental rendering and also like you said it should be spread out over 16 frames if it was something in the decal rendering.
jlv wrote: Sun Feb 12, 2023 1:37 am If you run a lap do the slow frames keep piling up?
It doesn't appear that they do, I still have frame times from 0.08-0.1s for around 9-10 frames.
jlv wrote: Sun Feb 12, 2023 3:06 am One more question - does lowering the resolution even more get rid of the slow frames?
Lowering to 1024x1024 I still had one slow frame at 0.05s, then about 17 frames ranging from 0.01-0.027s. Lowering to 512x512 also produced identical results.

Re: 2023-02-06 snapshot

Posted: Mon Feb 13, 2023 2:28 am
by jlv
Jakob Hubbard wrote: Sun Feb 12, 2023 4:05 am Yeah I definitely think that it's something to do with the new tilemap shader, because the lag spikes are almost identical despite the incremental rendering and also like you said it should be spread out over 16 frames if it was something in the decal rendering.
Tried it out on a GT 1030 and it duplicated your results. Right at the beginning I get slow frame times around .07 spaced around 12 frames apart. Once it gets past 150 frames it's smooth sailing at around .008 the rest of the way.

Could be the driver being clever and recompiling the shaders in the first couple of seconds to do things like convert uniform variables that always have the same value into constants. I'll run some more tests to see if I can isolate it to something specific.

This isn't a show stopper for anyone right? This was a really slow system and it was running perfectly after the first 2 seconds.

Re: 2023-02-06 snapshot

Posted: Mon Feb 13, 2023 2:46 am
by Jakob Hubbard
jlv wrote: Mon Feb 13, 2023 2:28 am This isn't a show stopper for anyone right? This was a really slow system and it was running perfectly after the first 2 seconds.
It is for racing, lags a couple seconds out of the gate for both me and my friend on high gpu's and this is pretty essential for racing scenarios. This post I made on the old snapshot highlights what's happening, it still happens on this snapshot.
Jakob Hubbard wrote: Sat Dec 24, 2022 3:38 am Having frame dips at load-in and couple seconds after gate drop. Usually didn't have this problem on the 5-19 snapshot. Other than that it runs perfect at max settings for me with no frame dips!

Also the after gate drop frame dip only happens once, going backwards in a demo and replaying it doesn't cause the issue but it happens upon first playthrough.