4.1 GPU Spike

[amerigoV]

I am seeing a notably heaver GPU load in 4.1 vs 4.0. Just opening the exact same map, none of the new toys my GPU will from from the 60s to the 80s. Once I turn on lighting and toss out a few candles, it jumps further to the 90s (example a couple of candles, a light on a token, then turn on Lighting/LOS/Player View).

In playing with it, I have very little change with LOS on or off in 4.0. But turning on either LOS or Lighting 4.1 is causing that initial bit jump.

I am not sure what you need that is helpful. But here are some specs.

OS Name Microsoft Windows 10 Home
Processor Intel(R) Core(TM) i7-8550U CPU @ 1.80GHz, 2001 Mhz, 4 Core(s), 8 Logical Processor(s)
Name Intel(R) UHD Graphics 620
Adapter RAM 1.00 GB (1,073,741,824 bytes)

[amerigoV]

An update - if I turn of Grid Snap and Token lock, it drops to the 80s. Still high, but better.

[Zarestia]

Fresh Campaign, no extensions, no player's connected.
Imported a .jpg map (2 mb, ~1500px*1000px), created some LoS, added ambient lighting, a torch and a candle. My GPU goes up to 65-70% until I close the map (deactivating LoS / Lighting does not help) going down to about 3%.
This is on a Ryzen 7 3700x and Radeon 5700 XT.

Same Map in 4.0 with more LoS and my GPU goes up to 7-8%, ambient lighting and two light sources probably shouldn't be 10 times more load heavy, right?

Can providge further testing tomorrow if needed.

[Moon Wizard]

This may be expected with everything going on with LoS/lighting. I'll run this thread past @cpinder.

Regards,
JPG

[Weissrolf]

GPU percentage readings can be misleading, because they measure a percentage of the current GPU clock-rate. Lower clock-rate = higher percentage for less load. Here are measurements that put GPU load into more context.

No LoS, no Lighting:
https://i.imgur.com/WTyMHzH.png

LoS, no Lighting:
https://i.imgur.com/bebZphO.png

LoS, Lighting:
https://i.imgur.com/8IcaJb7.png

Edit: corrected first image

[Zarestia]

Just some further testing from my side. D&D 5E, no extensions, no player is connected. Interestingly the big load/clock increase happens as soon as I click on the new image options (enable LoS / enable lighting). How many LoS points and torches/visions etc. I draw after that is nearly irrelevant.
Strange, that the clock jumps up to 800 MHz more than in v4.0, both only LoS enabled, that seems fishy.

https://i.imgur.com/YR4OE4S.png

[Weissrolf]

If LoS was (optionally= more GPU than CPU calculated then the clock-rate increase would be okay, but it does not seem to be that way? At least CPU load is still the main bottleneck for approved token movement.

[Zarestia]

Aprospros tokens.

One token, no token locking. Darkvision / Blindsight / Truesight, nothing really impacted the numbers in v4.1. In v4.0 I could see some slight increases with LoS enabled when moving the token very fast or through many LoS sources.
I would suspect that v4.1 is preloading most things and maybe because of that the numbers skyrocket as soon as I enable LoS / Lighting? Just some food for thought :)

[Kelrugem]

Aprospros tokens.

One token, no token locking. Darkvision / Blindsight / Truesight, nothing really impacted the numbers in v4.1. In v4.0 I could see some slight increases with LoS enabled when moving the token very fast or through many LoS sources.
I would suspect that v4.1 is preloading most things and maybe because of that the numbers skyrocket as soon as I enable LoS / Lighting? Just some food for thought :)

I believe Moon Wizard wrote somewhere that the new system indeed precalculates a lot which is why there are essentially three different vision types in the code to avoid overloading that (but I may misremember that) :)

[pindercarl]

Just some further testing from my side. D&D 5E, no extensions, no player is connected. Interestingly the big load/clock increase happens as soon as I click on the new image options (enable LoS / enable lighting). How many LoS points and torches/visions etc. I draw after that is nearly irrelevant.
Strange, that the clock jumps up to 800 MHz more than in v4.0, both only LoS enabled, that seems fishy.

https://i.imgur.com/YR4OE4S.png

Can you provide a copy of the campaign so that we are looking at the same data? Thanks.

[Zarestia]

Sure, see the attached 4.1 campaign. If you need any further info just ask :)

Steps to reproduce:
- Open the image/map in the campaign (should work with any other too)
- Enable LoS
> See drastic increase in GPU usage (even without a single point of LoS) --> LoS in 4.1 seems to hog way more resources than in 4.0.
- Enable Lighting
> See an even more drastic increase in GPU usage (even without a single point of Lighting)

[Weissrolf]

Does LoS CPU load decrease in return for increased GPU load (4.0 vs. 4.1)?

[pindercarl]

Sure, see the attached 4.1 campaign. If you need any further info just ask :)

Steps to reproduce:
- Open the image/map in the campaign (should work with any other too)
- Enable LoS
> See drastic increase in GPU usage (even without a single point of LoS) --> LoS in 4.1 seems to hog way more resources than in 4.0.
- Enable Lighting
> See an even more drastic increase in GPU usage (even without a single point of Lighting)

Thanks.

[Zarestia]

Does LoS CPU load decrease in return for increased GPU load (4.0 vs. 4.1)?

Just tested on a Curse of Strahd Module Map. Token Movement with only LoS seems to be a bit more smooth but still going 100% CPU Core/Thread if you hold down an arrow key for fast movement.
I think this was mentioned several times in other places and I want don't want to start another discussions in this thread but having HWMonitor the whole time open I can say most performance hiccups ('slow' initial campaign loading, several seconds loading when opening things in the asset menu, etc.) correlate with a 99%-100% utilazation of one CPU Thread. I know multi-threading is not the easiest part of programming, but I think there could be huge gains in responsiveness and performance. Just my 2 cents.

If someone wants anything further tested, just tell me :)

[Weissrolf]

4.0 - No LoS, no lighting: https://i.imgur.com/M0R6VxE.png
4.1 - No LoS, no lighting: https://i.imgur.com/ik00OOM.png

4.0 - LoS, no lighting: https://i.imgur.com/iDcYdwh.png
4.1 - LoS, no lighting: https://i.imgur.com/5902gzy.png

4.1 LoS calculation seems to use considerably more GPU processing compared to 4.0, while CPU load seems about the same.

This is a comparison of the same map (imported campaign from 4.0 to 4.1), just one PC token, no NPC tokens, no movement on the map (just display). Nvidia RTX 2070 Super, power-saving disabled (fixed clock-rate).

[Moon Wizard]

Please note that for the issue reported by Zarestia in post #3 along with the data provided in post #11; @cpinder made some adjustments to the lighting blur calculations to improve performance in the most recent build pushed today.

Regards,
JPG

[Weissrolf]

That improved GPU load considerably. Maybe slightly higher CPU load, though?

No LoS, no lighting:
https://i.imgur.com/QJsLOa3.png

LoS, no lighting:
https://i.imgur.com/NuHLfJm.png

LoS, lighting:
https://i.imgur.com/BgQFbVl.png