Tracks Sveg, Sweden

mantasisg · Nov 12, 2017

I think there still could be less trees in the left. Does the trees really are that dense in that location. Left is more realistic, as in the right they are just too dense. An ultimate battle must be going on between the roots

LeonS · Nov 12, 2017

I don't think it's too dense. It is pine forest with some birches mixed in, they tend to stand rather close to each other.

Here's my new fps counts:

track: 137 (partial dds+)
track + landscape: 127 (partial dds)
track + landscape + grass: 114 (LOD done, dds)
track + landscape + grass + tree walls: 90 (one shadow wall, dds)
track + landscape + grass + tree walls + trees: 76 (no shadows, dds)

I somehow lost 8 fps on the base track and I don't really know why (I don't think I changed anything?!). Also my new tree walls are extremely colstly and I don't quite know why yet. That's certainly what I'll have to investigate.

LeonS · Nov 12, 2017

this is how it looks at a min of 76fps:

LeonS · Nov 12, 2017

Ok, so I figured out where I lost my fps for pure track: time of day has quite a strong impact. For the pure track I get 7-8 fps (~5%) more at 4pm than at 9:30am.

These are the current values at 4pm:

track: 144 (partial dds+)
track + landscape: 133 (partial dds)
track + landscape + grass: 121 (LOD done, dds)
track + landscape + grass + tree walls: 97 (one shadow wall, dds)
track + landscape + grass + tree walls + trees: 82 (no shadows, dds)

I now have 1-2 more ideas on how to improve the fps (might not help much though), but yey, we've reached the 80s!

mantasisg · Nov 12, 2017

Thats a great discovery.

RasmusP · Nov 12, 2017

Little question about that: is it actually your gpu limiting here? Is the 1070 at 99-100% load while these fps happen?
I ask because mine is never, as long as I don't use supersampling etc.
It's always my CPU (2600k @4.4 GHz) that's limiting, without actually being at high loads too.
I found out that you can read out the actual threads with process explorer and if any program goes over 12%, for me the CPU is at its limit.
So I tell you this because when I activate the app substanding I take a hit about 20 fps. If I deactivate all apps, I get doubled the fps.
CPU stays below 60% as well as every core. My 1070 is at 30-50% load.
But one thread in process explorer is always around 12% or a little bit higher so there lies the hidden CPU bottleneck.
Maybe worth checking that for your work

Not that it would change anything, but I guess it would be nice to know if it's CPU or gpu limited here!

LeonS · Nov 12, 2017

RasmusP said:
Little question about that: is it actually your gpu limiting here? Is the 1070 at 99-100% load while these fps happen?
I ask because mine is never, as long as I don't use supersampling etc.
It's always my CPU (2600k @4.4 GHz) that's limiting, without actually being at high loads too.
I found out that you can read out the actual threads with process explorer and if any program goes over 12%, for me the CPU is at its limit.
So I tell you this because when I activate the app substanding I take a hit about 20 fps. If I deactivate all apps, I get doubled the fps.
CPU stays below 60% as well as every core. My 1070 is at 30-50% load.
But one thread in process explorer is always around 12% or a little bit higher so there lies the hidden CPU bottleneck.
Maybe worth checking that for your work

Not that it would change anything, but I guess it would be nice to know if it's CPU or gpu limited here!

Sorry, how exactly do I find this out?

edit: I found another fps. now at 83 without the environment grass, and 99 at minimum settings ( = no tree shadows at all) - this may not be a great track for AC minimum specs, but it could be well playable for slightly medium spec systems after all.

RasmusP · Nov 12, 2017

LeonS said:
Sorry, how exactly do I find this out?

Already in bed, I can give a better tutorial tomorrow!
You need msi afterburner and then go into the settings and activate the CPU loads and gpu load in the on screen display options. At the bottom of afterburner you'll see all the graphs. Click on the little "expand". Go on track for a minute or so and then have a look at the graphs.
If the gpu load isn't over 95%, it's the CPU that is limiting.
May I ask what CPU do you got?

The "better" version of it would be to install riva tuner too so you get all the data you want to see instantly while being ingame. But that can be quite complicated. But wanted to mention it...
Would look like this, depending on the settings

LeonS · Nov 12, 2017

RasmusP said:
Already in bed, I can give a better tutorial tomorrow!
You need msi afterburner and then go into the settings and activate the CPU loads and gpu load in the on screen display options. At the bottom of afterburner you'll see all the graphs. Click on the little "expand". Go on track for a minute or so and then have a look at the graphs.
If the gpu load isn't over 95%, it's the CPU that is limiting.
May I ask what CPU do you got?

The "better" version of it would be to install riva tuner too so you get all the data you want to see instantly while being ingame. But that can be quite complicated. But wanted to mention it...
Would look like this, depending on the settings

Ah. I've got a 4790k @ 4.0 GHz. I think "CAM" can also show that stuff. I'll see what I can find out.

edit: yea so I got 30-40% CPU and 98-99% GPU load in game.

LeonS · Nov 13, 2017

Just FYI, these are the tree walls that allow me to reduce the number of individual trees to 1/3 of what it was

LeonS · Nov 13, 2017

I've almost optimized everything now (including conversion to dds). Only idea I have left is LODing the tire walls.

track: 155 (dds)
track + landscape: 144 (dds)
track + landscape + grass: 137 (LOD, dds)
track + landscape + grass + tree walls: 109 (shadows off, one LOD'ed shadow wall, dds)
track + landscape + grass + tree walls + trees: 89 (no shadows, dds)

89 is pretty good I think! Leaves me some room to put the pretty 3d-grass behind the wall back in. Just the pretty forest ground I'm afraid will have to stay out, as it lives in the same place as the costly trees.

LeonS · Nov 13, 2017

hmm. dds vs. png water NRM. png looking much better.

PNG:

DDS:

Stereo · Nov 13, 2017

Yeah, sometimes it's better to go half-resolution uncompressed dds for textures. dxt5 compression really does a lot of artifacts on certain patterns. If you use PNG it just converts to uncompressed dds in memory.

RasmusP · Nov 13, 2017

LeonS said:
Ah. I've got a 4790k @ 4.0 GHz. I think "CAM" can also show that stuff. I'll see what I can find out.

edit: yea so I got 30-40% CPU and 98-99% GPU load in game.

That's some heavy stuff then. But better than hitting the cpu limit with trees and shadows. That would be really difficult to tweak with lower settings or similar things!
Are you at 1080p or at a higher resolution?
Anyway, if you want a second system with a 1070 to check the performance just tell me. I'm honest: I won't have time to really "beta-test" your track as it was mentioned earlier here, but a short performance check isn't a problem. Always interested in performance testing

Mitja Bonca · Nov 13, 2017

LeonS said:
hmm. dds vs. png water NRM. png looking much better.

PNG:
View attachment 221966

DDS:
View attachment 221965

Can you have png files? And which shader do you use, and which settings? Thx.

LeonS · Nov 13, 2017

Mitja Bonca said:
Can you have png files? And which shader do you use, and which settings? Thx.

Yea, it works with png. Performance takes a hit, but for one texture file I don't think it matters much.

And this is finally a request for a setting, so I'll put it in my next video

AccAkut · Nov 13, 2017

Do you have a driveable version working yet? Especially regarding performance there can be differences between driving and flying in F7 mode

LilSki · Nov 13, 2017

AccAkut said:
Do you have a driveable version working yet? Especially regarding performance there can be differences between driving and flying in F7 mode

Yes this is why this all started with optimization. I convinced him to load it in sim and he found the fps was not good at all.

LeonS · Nov 13, 2017

LilSki said:
Yes this is why this all started with optimization. I convinced him to load it in sim and he found the fps was not good at all.

Correct, all the fps figures are from hotlap mode (i.e., single car). I started out at 39fps with maximum details and got it to 89 now, so I'm quite happy about that. It might decrease a little with the grass in the environment, but I'm quite positive we can keep it above 80 at all times on maximum details. And of course it will have the option to get more performance by turning stuff off (won't really look too good then though). DIP stays below 2200 everywhere so far, but may increase a bit with more details and forest closer to the pit lane, but I believe it will still be on a fair level overall. Further optimizations in that regard will have to wait until after 1.0 -- performance is alright now and almost better than I was hoping for, so I'll soon be moving on to completing the remaining 60% of forests and trees, and do the physical mesh, cameras and AI.

edit: speaking of performance, I feel like the performance hit I take from the tree walls is a bit on the hefty side. They are quite low poly and there's only one wall that has shadows turned on, so I'm not sure where the big hit is coming from, really. Any ideas?

LeonS · Nov 13, 2017

RasmusP said:
That's some heavy stuff then. But better than hitting the cpu limit with trees and shadows. That would be really difficult to tweak with lower settings or similar things!
Are you at 1080p or at a higher resolution?
Anyway, if you want a second system with a 1070 to check the performance just tell me. I'm honest: I won't have time to really "beta-test" your track as it was mentioned earlier here, but a short performance check isn't a problem. Always interested in performance testing

mhh thanks, but I think it would be more interesting to get some lower spec'd systems to report performance. After all, our 1070s won't differ that much.