23 replies to this topic

[Gas Guzzler]

Hey guys, I know everybody indicates that MWO is very CPU intensive, so I was curious if there was some wisdom on here regarding my system.


I play at 4K with all settings on very high, and in some of the newer/larger maps, my FPS drops to ~40-50, with occasional dips if I'm getting shot a lot or running hot (the hot steam that comes out of your mech).

I'm basically wondering if its my i7-4790k or the Asus ROG GTX1080Ti that is limiting my system. With manually setting my fan speed to 75% my GPU tops out at 53C which seems to be plenty cool. I think they are both stout pieces of hardware but the i7-4790k is getting a little aged.

I'd appreciate any insight!

[Fox the Apprentice]

I would recommend using a tool like hwinfo64 to profile your system while playing. Most tools will let you see load, temperature, clock speeds, etc. and view them on a graph.

[Tarl Cabot]

Is that i7-4790k OC'd and to what speed? Below is the frequency ranges if not OC and simply allowing Turboboost to do the layman's temporary overclocking. With proper cooling 4.6-4.8 ghz is possible, and yes, a few hundred ghz can make a difference in avg FPS, especially since MWO has a high overhead when comparing dropping in combat vs the Proving ground.

Frequency - 4000 MHz base frequency
Maximum turbo frequency 4200 MHz (4 cores)
4300 MHz (3 cores)
4400 MHz (1 or 2 cores)

The other will deal with game settings themselves since the high overhead, many believe the game is not as optimize as it could be, but then it is also on an beta version of Cryengine 3.

As for the CPU getting old.. if ya looking for an excuse to update... The above still holds, purchasing a K version (Intel) allows for OC beyond the baseline, and then have it run like that continuously without the frequency jumping up and down.

[MrVaad]

I have a GTX 1080 and it's not limiting the game.

You can use geforce experience, windows 10 game bar or msi afterburner to display CPU/GPU usage while gaming.
You should monitor all your cpu cores, to see if one of them is near 90%.

MWO uses many cores, but will use more intensively a few of them (due to some particle effects and some 3D objects/map design). PGI modified some parts of the cryengine (mainly physics and network it seems) but i don't think it was a beta when they migrated to 3.4.4 ?

The main problems are some maps objects and some particle effects. Those need to be fixed by PGI.
You can improve the graphics and performance with some settings in a user.cfg file,
but it won't double your fps

Edited by MrVaad, 26 January 2021 - 04:22 AM.

[Gagis]

One thing worth testing is to increase the delay for how often the HUD refreshes itself. If I remember correctly, by default it tries to refresh HUD data for every frame. You GPU might try to draw a lot of frames, and the single-threaded HUD will just throttle everything trying to keep up.

Goes something like this:

-- FLASH UI/HUD
sys_flash_newstencilclear = 1
sys_flash_curve_tess_error = 4
 
-----------------------------------------------------
-- INGAME HUD
gp_hud_ams_update = 36
gp_hud_compass_update = 4
gp_hud_ecm_update = 24
gp_hud_engine_update = 2
gp_hud_heading_update = 4
gp_hud_heat_update = 2
gp_hud_targetinfo_update = 12
gp_hud_textwarning_update = 36
gp_hud_throttle_update = 2
gp_hud_weapon_update = 4

in user.cfg

That translates to once every 2, 4, 12, 24 and 36 frames. Once every 3 might be indistinguishable from 1 with decent FPS still.

Edited by Gagis, 26 January 2021 - 03:43 AM.

[MrVaad]

Gagis, on 26 January 2021 - 03:43 AM, said:

One thing worth testing is to increase the delay for how often the HUD refreshes itself. If I remember correctly, by default it tries to refresh HUD data for every frame. You GPU might try to draw a lot of frames, and the single-threaded HUD will just throttle everything trying to keep up.

I'm curious if anyone saw an improvement with the gpu_hud settings. I've been using them all at [30], and all at [1]. I can't measure/observe any difference in CPU/GPU usage, drawcalls, etc.

Default is 3 or 2, depending on the setting.

I'm suspecting it only modifies when MWO gets the data. The HUD is displayed on each frame anyway, and this is probably what takes 90% of the usage

Edited by MrVaad, 26 January 2021 - 12:16 PM.

[NARC BAIT]

I found your issue ... its here ...

Gas Guzzler, on 25 January 2021 - 12:46 PM, said:

I play at 4K

I've had a 4K screen for years .... but I think maybe I've always been too blind to truly appreciate it .... essentially, @ 4K your cpu is doing far more 'pre preparation' than was ever really expected to be done ... when the version of cryengine came out, very few machines were pushing 1920x1080, thats pretty much the maximum considered resolution for the time, and 4K is 4x1920x1080 blocks ..... there are a couple of stages that have to be done BEFORE a frame gets sent to the GPU, and its here things all start to fall apart ....

so, obviously, you want to shift the load from the CPU to the GPU, and the only way you can viably achieve that, is to drop the resolution, while increasing the visual detail .... increasing the quality beyond just putting your settings to 'very high' needs to be done through a user.cfg file, with the caveats that go with that ... but when you say the 1080ti is at 53c, that tells me its not being worked hard enough, even at 4K ...

firstly, you'll need to work out the minimum amount of pixels you 'NEED' ... if I stick my 4K monitor down to 1080p, its quite blocky, pixels can easily be made out ... for me personally, my ideal resolution, for my vision quality / distance to screen is just a little bit over 2K .... I believe the perceivable quality difference between 2K and 4K during 60+ fps is pretty negligible, if you can see the difference, at all, your looking too hard at one specific detail and probably about to be shot from the side ...

in relation to quality, a few tweaks in the user.cfg file REALLY add 'load' to the GPU ....

this disables the 'level of detail' system, meaning that we just always use the highest quality, without 'thinking' about it ... for most cards lower than yours, this is going to be a signifigant load increase itself, but yknow, go check out the dropship ring on tourmaline desert, at full quality, from a kilometre away ...

e_Lods = 0

but that only does 'objects' .... mechs, buildings, dropships, wrecks, stuff like that ... chances are .... the terrain is still trash, and constantly applying weird 'LOD' changes, and when the software drops the level of detail like that, its reducing the quality that your 1080ti can do at its maximum .... terrain in MWO is 'created' from voxels, and 'objects' intersect them ... anyway

e_TerrainLodRatio = 0

pretty much turns off the system ...

MrVaad, on 26 January 2021 - 03:50 AM, said:

I'm suspecting it only modifies when MWO gets the data. The HUD is displayed on each frame anyway, and this is probably what takes 90% of the usage

yes, most of the 'flash updates' only actually occur when fresh data comes in, so the testing grounds is not a viable place to look for impact .... the only ones i really change are

--;	 Time in ms to wait between each mouse move hardware event |
gfx_inputevents_mouse_move_threshold = 0.016
 
gp_hud_compass_update = 10
 
--;	  How many frames till we update the heading. ( this is the map AND heading )
gp_hud_heading_update = 2

tune gfx_inputevents_mouse_move_threshold to be your ideal render latency, that value is for 60 fps, 1 sec / 60 fps = 0.016, 1 / 144 = 0.006 etc ...

some of the older CPU's from both vendors benefitted from setting

--;	 Clears stencil buffer alternate mode | 0) normal - clear command / 1) draw rect
sys_flash_newstencilclear = 1

[Ignatius Audene]

Never touch 4k in mwo! Modified cfg might help a bit. But i still got dips sometimes on my ryzen 3600. Mwo just loves Intel. And some "lags" are more in relation to bad servers.

Your gpu is sitting idle no matter what. Back in the day i played mwo on an gtx 770

Edited by Ignatius Audene, 26 January 2021 - 08:53 PM.

[Gas Guzzler]

So I did some tuning, and ended up at 1.265V and 4.68 GHz on my CPU. I also made some of the suggested changes to the user config file.

FPS definitely improved and seems to be more consistent.


I also tried turning on TXAA since I had assumed that would be all GPU. That produced a slight dip in frames, which wouldn't be so bad but I would also get 1-2 second lockups. I thought that was interesting.

Looking at OCCT, the usage between all 4 cores ranged from 70%-95%, while my GPU core load went up to 100%, even just reverting to FXAA. Core temp reached 80C which may be why it didn't quite use 100%.

I should add that even though my GPU hit 100%, temperatures didn't go above 48. Does that mean I can get more out of it?? I have my fans fixed at 75% capacity.

Edited by Gas Guzzler, 27 January 2021 - 01:43 PM.

[NARC BAIT]

personally, I'd skip out on post-processing, its done after the frame is complete, adding to the time it takes a frame to get spat out, and at 60 fps, the next frame comes along in 16.7 ms, so if a jagged line does occur, your only looking at it for a fraction of a second anyway, its probably really most beneficial when people are using 720p resolution or lower .... technically, post processing never actually adds detail, its an extra process, that removes it .... I think MWO's mech perspective makes it even less useful that normal, because no matter what, your always further away from anything than what a 2m tall char would be, on the ground, the situation where it tends to clean up an image the most, is when your looking through a thin object, like a chain link fence, and having that chain link fence be 'cleaner' helps you see whats beyond it ...

if your CPU cores hit 80c, they might downclock to maintain thermal profile .... if the main thread happens to be on that core, its likely to create some sort of stall ... heat is an unfortunate side effect of throwing electricity through the tiny spaces occupied by the cpu cores ... I'm going to assume you've got a standard cooler on it, whereas most of the people who overclocked that CPU probably had more beefier coolers strapped onto them ....

if you happen to have a 'Z' series motherboard, to match your 'K' series processor then you probably also have XMP capable ram, and depending on the board in use, some have a variety of 'tweaks' that can yield a better results, alot of it might just make the whole thing more unstable for a few extra percentage points ....

post-processing will blow out the GPU processing, but probably wont add as much heat as getting the GPU load up to 100% .... depending on the scenario, it might hit 100% usage, but not actually be at its highest possible 'performance level' ... the 1080ti's were an overdesigned card, NVIDIA never replaced many for failures, and the performance level is still comparable to current gen, its pretty much equivalent in raw oomph to a 3060ti ... just no ray tracing, or other useless features ... your 1080ti is still overpowered for MWO ....

[MrVaad]

If you are talking about the post processing setting ingame, it does nothing literally

The only values changed are on low :
- change depth of field mode
- increase HDR eye adaptation cache.
- disables sun shaft

If you activate geometry instancing with those three lines, you will significantly lower drawcalls (and potentially your cpu usage) :
r_BatchType= 0
r_GeomInstancing = 1
r_GeomInstancingThreshold=2

For objects LOD, i prefer to activate the engine setting that will remove useless LODs (the ones that don't improve performance because they go too low in texture/geometry)
e_LodsForceUse=0

Having kept good enough LOD objects, raise the distance at which they will change (to 30 or more as you wish):
e_LodRatio = 30

That being said, the most impactfull thing at 4k is probably particle effects. Particle effects use a lot of pixel fillrate.
At that resolution, it's a loooooot of pixels per frame, and particles don't hide others like geometry. They are always drawn even when they are on top of each other (smoke, for example)

Edited by MrVaad, 27 January 2021 - 03:44 PM.

[Gas Guzzler]

NARC BAIT, on 27 January 2021 - 03:09 PM, said:

personally, I'd skip out on post-processing, its done after the frame is complete, adding to the time it takes a frame to get spat out, and at 60 fps, the next frame comes along in 16.7 ms, so if a jagged line does occur, your only looking at it for a fraction of a second anyway, its probably really most beneficial when people are using 720p resolution or lower .... technically, post processing never actually adds detail, its an extra process, that removes it .... I think MWO's mech perspective makes it even less useful that normal, because no matter what, your always further away from anything than what a 2m tall char would be, on the ground, the situation where it tends to clean up an image the most, is when your looking through a thin object, like a chain link fence, and having that chain link fence be 'cleaner' helps you see whats beyond it ...

if your CPU cores hit 80c, they might downclock to maintain thermal profile .... if the main thread happens to be on that core, its likely to create some sort of stall ... heat is an unfortunate side effect of throwing electricity through the tiny spaces occupied by the cpu cores ... I'm going to assume you've got a standard cooler on it, whereas most of the people who overclocked that CPU probably had more beefier coolers strapped onto them ....

if you happen to have a 'Z' series motherboard, to match your 'K' series processor then you probably also have XMP capable ram, and depending on the board in use, some have a variety of 'tweaks' that can yield a better results, alot of it might just make the whole thing more unstable for a few extra percentage points ....

post-processing will blow out the GPU processing, but probably wont add as much heat as getting the GPU load up to 100% .... depending on the scenario, it might hit 100% usage, but not actually be at its highest possible 'performance level' ... the 1080ti's were an overdesigned card, NVIDIA never replaced many for failures, and the performance level is still comparable to current gen, its pretty much equivalent in raw oomph to a 3060ti ... just no ray tracing, or other useless features ... your 1080ti is still overpowered for MWO ....

Thanks for all this info.

I do have an ASUS Z97 Sabertooth Mark 1 motherboard, and I got a Noctua NH-U12S air cooler. Not quite water but much larger than what came with the chip stock.

I did see the XMP in the bios utility... wasn't sure what it did though.

[General Solo]

Gas Guzzler, on 27 January 2021 - 01:41 PM, said:

I should add that even though my GPU hit 100%, temperatures didn't go above 48. Does that mean I can get more out of it?? I have my fans fixed at 75% capacity.

You have good cooling, water cooled GPU maybe imo
So you could overclock the GPU core some
Same doesn't apply for GPU RAM doh, thats its own thing and mostly temp unrelated

Edited by General Solo, 28 January 2021 - 02:01 AM.

[Ignatius Audene]

The gpu usage is normal. The low temps on the gpu shows u, that it is not in full work (they adjust speeds etc.)

Your cooler is good. How old is the thermal paste? Maybe repaste the cpu.

[NARC BAIT]

Gas Guzzler, on 27 January 2021 - 03:20 PM, said:

I did see the XMP in the bios utility... wasn't sure what it did though.

XMP stands for something like extreme memory profile .... you'd think EMP would have made more sense, but intel were probably struggling to copyright that .... anyway, essentially they contain information for the motherboard to configure itself, mostly for high speed operations .... if your memory is capable of it, and your not using it, chances are you can get some free performance there .... could just open a rabbit hole of headaches though .... RAM with XMP profiles tends to alleviate alot of the initial guesswork though ....

without ranting on for too many hundreds of words, basically, you'll want to work out what your memory frequency currently is, you can find this in the windows 10 task manager, on the performance tab, select memory from the left hand panel, and its at the top right hand side ...

then work out what your ram is rated for, use a phone, take a photo, it will almost always say its rating on the sticker ... if the two values are mismatched, then you should probably try to get XMP to work, for instance, it might say its running at 1600mhz, but the sticks are rated capable of 2133mhz or 2400 mhz ... even if the values are matched, but XMP is not enabled, you might be able to get an extra few percentage points faster

[General Solo]

Added some of that to my user.cfg

It adds lag

For me

[MrVaad]

Which settings ? Can you post your user.cfg ?
Dont forget you need a file tuning.cfg (see my guide) else some settings are overwritten by MWO

Edited by MrVaad, 28 January 2021 - 04:49 AM.

[General Solo]

I had a user.cfg with stuff in it already

I added things in this post, in alphabetical order, as I wasn't sure if it made a difference, the order that is.

My damage went way down

Then I moved the user.cfg out of the game folder and it runs better damage wise, with some hitching

Here the lastest user.cfg

e_Lods = 0
e_TerrainLodRatio = 0
e_LodsForceUse=0
e_LodRatio = 30
--; Time in ms to wait between each mouse move hardware event
gfx_inputevents_mouse_move_threshold = 0.016
gp_default_cockpit_light = 2
gp_hud_ams_update = 3
gp_hud_compass_update = 10
gp_hud_ecm_update = 3
gp_hud_engine_update = 3
gp_hud_heading_update = 2
gp_hud_heat_update = 3
gp_hud_stealth_armour_update = 3
gp_hud_targetinfo_update = 3
gp_hud_textwarning_update = 3
gp_hud_throttle_update = 3
gp_hud_weapon_update = 3
sys_SleepIfInactive = 0
r_ShadowsDepthBoundNV = 1
r_BatchType= 0
r_GeomInstancing = 1
r_GeomInstancingThreshold=2
r_DepthOfField = 3
r_DepthOfFieldBokeh = 1

[Ignatius Audene]

Do u know anything specific about your ram? Clock speed (timings etc). U can try activating xmp profile. It's basically an small tested oc for the ram. Normal ram speed without xmp is only 2666 (ddr4). I am out of touch with the older gens. So u can take look at xmp or fine tune manually (further guides recommended, like cpu dont go up with voltage to much).

Edited by Ignatius Audene, 28 January 2021 - 12:13 PM.

[MrVaad]

General Solo, on 28 January 2021 - 05:07 AM, said:

...

1. Try that group of gp_hud settings alone (disable everything else) and see if you like the result. 2 and 3 are default values so you're not gaining much.

2. Set e_Lods = 1 and play with e_LodRatio = 30 (higher values will make object lower their geometry farther). As NARC_BAIT said, it's a matter of preference and if you have spare GPU power

3. e_TerrainLodRatio is locked to 2 on many maps (i have a list somewhere).
0 could be overkill, try 0.5 or 1.

4. Nobody really knows what r_ShadowsDepthBoundNV does. It could cause bugs. Anyway, I've never measured any gains (only it adds a few drawcalls).

5. Those depthOfField settings set another form of depthoffield, it's a matter of preference. It could need more GPU power, i'm not sure because i didn't spent much time on it.

6. gfx_inputevents_mouse_move_threshold is mainly for mouse detection when hovering over buttons (and tooltips). It does not seem to have any impact on mech gameplay.

7. gp_default_cockpit_light sets cockpit ligjts (from 1 to 5, 1 disable lights).

Quote

There are other user.cfg settings that help a lot more.
But again, don't forget to add a file tuning.cfg (all the guides mention it). Because many of those settings will be overwritten by MWO without it.

Edited by MrVaad, 28 January 2021 - 01:52 PM.