Low FPS // way over the top specs // server slow down?

spamilator

Hi guys! I'm wondering why I'm getting so low FPS on an oversized rig for NS2. Specs:

i5-3570k @4.4 Ghz
2x970 GTX Overclocked SLI
8 GB RAM
Samsung EVO 840 SSD

On a server with 42 players I drop down to ~50 FPS lowest while my hardware is bored (see screenshots). Can servers slow down the client FPS? r_stats is saying "7 ms waiting for world update thread"...

Anyone can explain what's going on?

In-game screenshot:


Hardware monitors:

AurOn2

How late into the game is this? i see you have an AA, maybe turning off Fancyinfestation (the bubbly nonsense) could help?
Could be the alien commander went nuts on the cysts etc.

Also, servers don't slow down client FPS, Last I checked.
If you're confused further, go play on a server with less than 30 ticks, it's noticable, but not frame lag.
(net_stats to console to see)

spamilator

Wasn't that far into the game, a few minutes I guess. Do you think the server could slow down the client FPS artificially? Anti aliasing isn't the problem for my cards...

Asraniel

42 player servers are not supported by ns2. The game has not been optimized to work with that playercount. If you want good performance play on standard ns2 servers, which is anything up to 24 players

GhoulofGSG9

spamilator wrote: »

Wasn't that far into the game, a few minutes I guess. Do you think the server could slow down the client FPS artificially? Anti aliasing isn't the problem for my cards...

Servers themselves shouldn't slow your fps down, but the amount of entities (which is basically a lot higher on a 40 slot server than a 18 slot ones), as the amount of physic checks increases exponential to the number of entities which can cost a lot frame calculation time.

Overall tips:

1. Don't play at 40 slot servers if you want to have a appropriated ns2 experience. There are real technically reasons why the player count is official locked to 24 and you have to hex-edit the server.exe to unlock more slots.

2. Enable multi-threaded physics (that should help ns2 a bit to use your cpu properly)

3. Turn off particles, shadows, AA and also turn down infestation quality. Those are known to use in-effective algorithms and therefor slow you down a lot.

If you want to help us (the CDT) optimizing the game you should run p_logall in the console before starting to play and upload the created .plog file (you'll find it in %appdata%\Natural Selection 2\) afterwards here

devel

CPU can't handle it (according to screenshot).

spamilator

devel wrote: »

CPU can't handle it (according to screenshot).

GhoulofGSG9 wrote: »

spamilator wrote: »

Wasn't that far into the game, a few minutes I guess. Do you think the server could slow down the client FPS artificially? Anti aliasing isn't the problem for my cards...

Servers themselves shouldn't slow your fps down, but the amount of entities (which is basically a lot higher on a 40 slot server than a 18 slot ones), as the amount of physic checks increases exponential to the number of entities which can cost a lot frame calculation time.

Overall tips:

1. Don't play at 40 slot servers if you want to have a appropriated ns2 experience. There are real technically reasons why the player count is official locked to 24 and you have to hex-edit the server.exe to unlock more slots.

2. Enable multi-threaded physics (that should help ns2 a bit to use your cpu properly)

3. Turn off particles, shadows, AA and also turn down infestation quality. Those are known to use in-effective algorithms and therefor slow you down a lot.

If you want to help us (the CDT) optimizing the game you should run p_logall in the console before starting to play and upload the created .plog file (you'll find it in %appdata%\Natural Selection 2\) afterwards here

That was helpful, thanks! Gonna try getting a plog file for you!

IronHorse

Agree with everything except turning down infestation. It results in a 2 fps loss at most, and is horrendously ugly when turned down.

Did you not get a warning pop up when joining the 40 slot server? 24 is the maximum supported and anything higher results in a warning to the client.

spamilator

Hi guys! Sorry for the delay, couldn't really find the time... On the bright side I found out some new stuff:

• NS runs fine most of the time and utilizes all GPU / CPU cores
• There are random FPS drops in certain scenarios (not so much dependent on visual demanding effects, see below)
• In certain view angles the game stops utilizing GPU cores and the frame rate drops down to a critical number
• This issue could be related to NVIDIAs new GPU generation (9xx series)

I found a way to demonstrate the issue using the spectator view on the map Eclipse: Go in first person fly mode and zoom out of the map in small steps. At some point when enough objects are in the view, GPU utilization decreases and the FPS drop. See this this album for a demonstration.

Could this be related to this nvidia issue with the current 9xx cards? Anyone could hop into the sandbox and verifiy if the utilization of the GPU drops when more objects are in the viewport?

Kouji_San

You are going outside the map though so that FPS drop is too be expected. You are not within the OcclusionGeometry, meaning everything is being rendered and no occlusion is done what so ever, or is indeed very buggy when doing this.

Also Spectator mode is not loading the lua files required for player movement and collisions, so that bit is also not present in the recorded data

spamilator

Kouji_San wrote: »

You are going outside the map though so that FPS drop is too be expected. You are not within the OcclusionGeometry, meaning everything is being rendered and no occlusion is done what so ever, or is indeed very buggy when doing this.

Also Spectator mode is not loading the lua files required for player movement and collisions, so that bit is also not present in the recorded data

How does that explain the low FPS despite low hardware utilization?

Kouji_San

Engine choke probably

wooza

AurOn2 wrote: »

If you're confused further, go play on a server with less than 30 ticks, it's noticable, but not frame lag.
(net_stats to console to see)

FYI, the 42p server has tweaked network variables with 20tick as default (and there is no noticable lag)

(how many disagrees i will earn? :-? )

Kouji_San

wooza wrote: »

AurOn2 wrote: »

If you're confused further, go play on a server with less than 30 ticks, it's noticable, but not frame lag.
(net_stats to console to see)

FYI, the 42p server has tweaked network variables with 20tick as default (and there is no noticable lag)

(how many disagrees i will earn? :-? )

You like living on the edge don't ya



When I was back on my old C2D E6420 (build180-ish?), the lower FPS made the netlag less noticeable. Quite an odd but logical side effect However with all the added stuff and visuals, I'm sure that old timer can't handle NS2 right now even if clocked to 3.4Ghz

IronHorse

wooza wrote: »

AurOn2 wrote: »

If you're confused further, go play on a server with less than 30 ticks, it's noticable, but not frame lag.
(net_stats to console to see)

FYI, the 42p server has tweaked network variables with 20tick as default (and there is no noticable lag)

(how many disagrees i will earn? :-? )

That's because tickrate alone is not a good measure.
This is why matso has been working on better methods in which to gauge server performance.
Your "sitting idle" 20 tickrate is going to be smoother than a "under load / stressed" 20 tickrate of a regular server, which has a lot of other factors involved.

Not to mention, that's a pretty subjective claim to make, dont ya think? Disregarding the stats which show that server hitting ~11-13 tickrate often enough, every time I play there, it is most definitely noticeable to me... I guess I am sensitive to such a thing though, and therefore it is always noticeable to me.

wooza

There are better tools available to measure server performance. For instance you could post a screenshot of the perfbrowser, showing how terrible it performs.

ATF

IronHorse wrote: »

wooza wrote: »

AurOn2 wrote: »

If you're confused further, go play on a server with less than 30 ticks, it's noticable, but not frame lag.
(net_stats to console to see)

FYI, the 42p server has tweaked network variables with 20tick as default (and there is no noticable lag)

(how many disagrees i will earn? :-? )

That's because tickrate alone is not a good measure.
This is why matso has been working on better methods in which to gauge server performance.
Your "sitting idle" 20 tickrate is going to be smoother than a "under load / stressed" 20 tickrate of a regular server, which has a lot of other factors involved.

Not to mention, that's a pretty subjective claim to make, dont ya think? Disregarding the stats which show that server hitting ~11-13 tickrate often enough, every time I play there, it is most definitely noticeable to me... I guess I am sensitive to such a thing though, and therefore it is always noticeable to me.

Oh, gosh. You looked really hard to find data from a previous experimental engine build which had a bug in it, causing those issues.

When - exactly - have you last played there? Since your NS2 ID does not exist in the logs of the last 13 months of server operation.

Kouji_San wrote: »

You like living on the edge don't ya

You have no idea! Haha.

IronHorse

Only a few weeks ago. Only one server with Wooza in the name, right?