Loading...
  OR  Zero-K Name:    Password:   

Performance of Zero-K

21 posts, 1714 views
Post comment
Filter:    Player:  
Page of 2 (21 records)
sort
Lynx
Last night http://zero-k.info/Battles/Detail/892119 was horrible. It was laggy, stuttery and I found it almost unplayable.

I realise it is a supreme cheek that I ask this. But I am going to ask anyway. Please forgive me.

Any chance that development work could focus for a while on improving in-game performance? Recently on large cluster games I am seeing very low FPS and very distracting stutters at various points throughout the game.

I do not know if this could be an issue to do with my software setup having recently ditched Linux for Win10. This morning I tried installing the latest Nvidia drivers rather than using whatever Win10 installed by default. Perhaps Win10 performance is actually worse than Linux (although Linux had a memory leak issue IIRC - Win10 sits at ~4GB without steadily growing after each game).

I have an i5-3570K CPU, Nvidia GTX 660 [2GB video ram], and 8GB ram. My system is obviously a bit old now, but based on the other games it can play very well it feels to me that it ought to be able to run Zero-K well.

+1 / -0


3 years ago
Impart infolog. Take a screenshot with /debug enabled. Do and link a benchmark.
+0 / -0
Lynx
3 years ago
I will try when I get more time AUrankAdminGoogleFrog. Is there any information on the wiki about benchmarking or performance optimisation etc.? What are the gfx settings set to by default on Steam install? I selected medium gfx settings out of frustration last night but no change. Perhaps restart required for change.
+0 / -0


3 years ago
A restart is needed for some changes. Some outdated benchmarks are under Help. Updating them is not a particularly difficult task, but it takes time, so someone could step up and do it.
+0 / -0


3 years ago
Is this the game that SErankZnack spammed 9,000,000 winds?
+0 / -0


3 years ago
No.
+0 / -0
Lynx
3 years ago
AUrankAdminGoogleFrog:

InfoLog from low fps battle 892119 here: https://pastebin.com/B6dE5b8C

Benchmark here: https://github.com/ZeroK-RTS/CrashReports/issues/30675

+0 / -0


3 years ago
First thing to try: Chili Deluxe Player List has terrible performance. Remove it and use the default one.
+0 / -0
Lynx
Thanks. Otherwise can you tell anything else from the benchmark? Do you see average FPS and stuff like that? Is my system good enough to run Zero-K well? On Linux I forced performance setting in respect of CPU to prevent throttling. Would that help in Windows? I have a feeling the Linux performance may actually have been better. But I'm not certain. Would 64bit vs 32bit impact performance?


Dumb question: how can a 2D player list with text, numbers and simple graphics like bars significantly impact performance? Has there been any attempt to speed up?
+0 / -0


3 years ago
Your average FPS was around 10, which seems ok. The benchmark doesn't have players in it so it won't measure the playerlist.

quote:
Dumb question: how can a 2D player list with text, numbers and simple graphics like bars significantly impact performance? Has there been any attempt to speed up?

There is no bound on how poorly you can make a piece of code perform. Deluxe playerlist was never (or at least not for a very very long time) enabled by default because it has terrible performance. It was based on the default playerlist, which was also terrible. I completely replaced the default playerlist with a cheap implementation. If anyone wants to write a version of deluxe with the replacement as a base, then they are free to do so. In the meantime people with deluxe have to suffer the performance cost.
+0 / -0
Lynx
3 years ago
Thanks. To be clear, which is the list I should enable? Is that the so-called crude player list?

Is an average FPS of 10 not terrible? I mean I thought at least 30 and more preferably 60 is needed for smooth gameplay?
+0 / -0


3 years ago
The benchmark is a stress test. What is your FPS when spectating a game with about 600 units?
+0 / -0
Lynx
I don't know. It might be good if there was a sort of benchmark that you could pass or fail to identify sufficiently decent performance to play the game.

I suppose if game never goes below 30FPS in cluster then all is well? Is that achievable? I see some posts about even extremely fast machines struggling, so I wonder if networking or something else is an issue.

I will try that replay on the troublesome battle above with the deluxe player list disabled and see if I can see a difference.

Oh, and what is the player list that I should enable?

Thanks for your help here by the way.
+0 / -0

3 years ago
quote:
if game never goes below 30FPS in cluster


I never had 30 fps in cluster after midgame when there is mass units. Game eat single core so you need CPU with good single core speed.



I didn't even know that Chili delux player list could impact performance in such levels. However i found that my ATI GPU preformed very bad on win7 64 bit. When i changed it with 1050 TI then my performance was greatly improved.
+0 / -0
Lynx
AUrankAdminGoogleFrog looking at the widget profiler I see deluxe player list sit at <1%, but I see Chili FrameWork Draw (or something like this) run at just over 15% - I presume this is the culprit? Surely presenting such simple graphics ought not to eat up so many processing cycles?
+0 / -0


3 years ago
A lot of the cost of deluxe playerlist shows up as activity in the Chili Framework. The good playerlist is "Chili Crude Player List".

quote:
Surely presenting such simple graphics ought not to eat up so many processing cycles?

This isn't how computers work. Literally anything can be done arbitrarily poorly.
+1 / -0
Lynx
3 years ago
Am I correct in supposing that the deluxe player list hurts performance in proportion to the number of players? Would the drag scale linearly with number of players or exponentially? Who is responsible for this madness?
+0 / -0


3 years ago
I don't know. The answers to those questions are unimportant. Where does knowing exactly how bad the performance is get us? Have you tried to read and improve other people's code before, or am I talking across a large gulf of understanding?

The author list says "CarRepairer, KingRaptor, CrazyEddie", which fits with my memory of who was generally writing the UI, but I mostly kept out of the original chili UI widgets because of all the crazy code.

At some point (I think during a November) I completely rewrote the bottom three parts of the UI (not the minimap) because they had terrible performance similar to the terrible performance of the playerlist. This was far easier than trying to read and significantly modify other people's code. I did not rewrite the playerlist because it looked like a big boring thankless task and there were other things to do.

Over the subsequent years I tried to get other people to take on the fairly well-defined task of writing a playerlist, so I could do other boring important that needed doing, but nobody managed to write a complete replacement. DeinFreund got the closest with the Tab playerlist (which also has terrible performance), which a least eased up on the amount of information that a rewritten playerlist would "need" to display. Eventually I got fed up with the one remaining widget of terrible performance, and the new Crude Playerlist is the result.
+1 / -0
Lynx
OK I have done some more experimenting.

The Deluxe Player List increases Chili Framework (DrawScreen) by around 10%

The Chili Crude Player List increases Chili Framework (DrawScreen) by around 3%.

Looks like the unit shadow widgets may also eat up quite a few cycles?

What are these percentages I wonder? At the bottom is says 'total FPS cost'.

By the way is the only way to reset to default widgets to manually delete the LuaUI Config files?

+0 / -0
3 years ago
Unfortunately, zerok's engine aka springrts is not using vulkan so there is significant cpu overhead and this game is unforutnately cpu intensive on top of that api overhead. You need a more recent cpu to do good for this game.
+0 / -0
Page of 2 (21 records)