65 replies to this topic

[DeathlyEyes]

I'm starting to wonder if the game has an arbitrary limit added to it for the amount of video memory it can use. My performance hasn't been the same since after the "optimizations" for 12 man queue. When 12 man queue was first implemented I was able to run the game fine with an older GPU. After people complained they couldn't run the game anymore on integrated gpus and the like and PGI did an optimization pass I started to have issues. CW just was the tipping point. I have 4GB of available video memory and I never go over 800 MB of usage.

[FuzzyLog1c]

DeathlyEyes, on 30 December 2014 - 09:50 AM, said:

I'm starting to wonder if the game has an arbitrary limit added to it for the amount of video memory it can use. My performance hasn't been the same since after the "optimizations" for 12 man queue. When 12 man queue was first implemented I was able to run the game fine with an older GPU. After people complained they couldn't run the game anymore on integrated gpus and the like and PGI did an optimization pass I started to have issues. CW just was the tipping point. I have 4GB of available video memory and I never go over 800 MB of usage.

Arbitrary? No. Configurable? Yes. For both RAM and VRAM.

If you browse the CryEngine support forums, you'll find a lot of posts from the 2007-2008 timeframe where customers ran into memory allocation errors (exactly like the ones we're having right now) because their map assets size exceeded the default CryEngine memory allocation.

IIRC, MWO still uses the default settings:

sys_budget_sysmem = 2048

It boggles the mind. Unfortunately, I believe that setting is locked by PGI.

Edited by FuzzyLog1c, 30 December 2014 - 10:30 AM.

[DeathlyEyes]

FuzzyLog1c, on 30 December 2014 - 10:23 AM, said:

Arbitrary? No. Configurable? Yes. For both RAM and VRAM.

If you browse the CryEngine support forums, you'll find a lot of posts from the 2007-2008 timeframe where customers ran into memory allocation errors (exactly like the ones we're having right now) because their map assets size exceeded the default CryEngine memory allocation.

IIRC, MWO still uses the default settings:

sys_budget_sysmem = 2048

It boggles the mind. Unfortunately, I believe that setting is locked by PGI.

I managed to run sys_budget_sysmem = 8192 and it worked, at least according to the omicron file. Which when i tried disabling hwobjectocclusion told me it was forced off. So I think the omicron file is at least reliable in telling us what settings PGI is using. This has been a serious issue ever since 12 man optimizations for low end hardware after 12v12 was implemented.
CPU usage seems pretty even. It just isn't getting to 100% ever. It's like something is waiting for data somewhere and the bottleneck isn't the CPU or the GPU.

Edited by DeathlyEyes, 30 December 2014 - 10:38 AM.

[FuzzyLog1c]

DeathlyEyes, on 30 December 2014 - 10:36 AM, said:

I managed to run sys_budget_sysmem = 8192 and it worked, at least according to the omicron file. Which when i tried disabling hwobjectocclusion told me it was forced off. So I think the omicron file is at least reliable in telling us what settings PGI is using. This has been a serious issue ever since 12 man optimizations for low end hardware after 12v12 was implemented.
CPU usage seems pretty even. It just isn't getting to 100% ever. It's like something is waiting for data somewhere and the bottleneck isn't the CPU or the GPU.

In my testing, I played multiple matches with sys_budget_sysmem = 8192 and sys_budget_videomem = 3072 and did not notice any difference in the RAM or VRAM numbers reported by MSI Afterburner (typically 4500 MB and 1200 MB @ 2560x1600, respectively).

It's interesting that the RAM number reported by MSI Afterburner is nowhere near the number reported by Windows Task Manager, which is invariably at about 1200 MB (in-game) regardless of what's going on.

[FuzzyLog1c]

Can I get someone with a high-end CPU to double-check me on this? The following needs to go in your user.cfg:

r_MultiThreaded=1
gpu_particle_physics = 1
sys_physics_CPU = 2

I'm presently playing--gathering framerate and GPU utilization data. Also watching performance while getting shot in the face by LRMs and autocannons. The residual smoke after taking a pair of LRM20s to the face typically drops even my quad-Crossfire machine to less than 5 fps. This may help.

Edit: after three runs, performance (particularly when getting hit) appears to be remarkably improved. Was able to hold a consistent 62, 65, and 72 fps (limited by MSI Afterburner each time) with one overclocked R9 290X. Still testing, though--definitely needs peer review to reduce the potential that I've fallen prey to the placebo effect.

Edited by FuzzyLog1c, 30 December 2014 - 11:40 AM.

[AC]

I know my FPS takes a dump when the cockpit effects are going on (muzzle flashes and explosions). Turning off cockpit glass gave me a nice 8fps boost all alone. I think the cockpit glass was absolutely unnecessary and was a huge waste of resources as most people turn this off to be able to run the game at better fps. Not to mention, in certain scenarios, it was nearly impossible to see out of the cockpit with the glass effect turned on.

[xWiredx]

Your name being FuzzyLog1c makes sense. What we have tested to death and presented in mounds of data in the hardware subforum is soundly dismissed by you.

I can tell you right now that my 5820K @ 4.3GhZ and GTX 980 will do -minimum- fps in the high 50s to low 60s depending on the map with averages in the mid 80s to mid 90s and maxes in the 160-180 range. This is all settings at very high, 1080p, in-game AA turned off, 8x AA and 8X transparency AA forced in the Nvidia control panel, color grading and cockpit glass turned off in the user.cfg with a fov of 70. I have presented actual numbers, and in my last post a chart, showing this.

CPU utilization is never high, but is spanned across anywhere between 4 and 8 threads. Just because CPU utilization isn't peaking doesn't mean you don't need more CPU power. MWO requires executing as many instructions as possible as quickly as possible, not executing a few monolithic ones as quickly as possible. If you think you don't need more CPU power and that a higher clock speed won't help, you don't actually know how CPUs work.

We have already tested several things way beyond these simple user.cfg tweaks being discussed here. Read the hardware subforum posts about performance before telling somebody who has spent months helping people and testing various things that they don't know what they're talking about.

[FuzzyLog1c]

xWiredx, on 30 December 2014 - 11:37 AM, said:

Your name being FuzzyLog1c makes sense. What we have tested to death and presented in mounds of data in the hardware subforum is soundly dismissed by you.

http://en.wikipedia....iki/Fuzzy_logic

I ignore the MWO hardware subforums because they're filled with threads recommending non-functioning / locked CFG parameter tweaks and people (like you) that draw inaccurate conclusions from insufficient data.

[xWiredx]

The data isn't insufficient at all. If you avoid the mounds of data collected on both the Intel and AMD sides so far, you shouldn't be making -any- recommendations whatsoever. I've run at least 25 separate test runs on my Haswell-E system alone, and at least another 50 on my old Sandy Bridge system, which is in addition to the dozens of other threads full of data. We have in-depth data that is recorded. You offer none. We have credibility for solving almost every issue that gets presented. You have none. At this point, if anybody wants to help you stumble through trying to figure things out, that is their loss. Until PGI can make further optimizations, the hardware subforum is a far better place to look for help than threads like this with people like you that dismiss actual data and results.

[McHoshi]

... hmm ... why not ship this whole project MWO into a complete different engine like for example Unreal Engine 4.5 ? cryengine doesn´t make sense at all because it´s not a good engine at all?


PGI sattled the wrong horse! Now we all see the results !!!

Edited by McHoshi, 30 December 2014 - 03:56 PM.

[Maggiman]

Pgi knows about this problem actually. Somewhere in the past they stated, that they have problems with draw calls being heavy on the cpu (A reason why there is little graphic evolution despite enabling the directx 11 code branch).
And this doesn't comes as a suprise really. How many directx9 games do you know that efficiently use more than 2 threads?

Look at Deathly Eyes graph. 4 cores are being utilised at about 50%. Thats actually 2 threads hopping around. Just like pretty much every other directx9 game.

And the switch to dx11 was mostly for saying "yeah we can start working on the dx11 path now". They flat out stated, that they didn't do much in that department yet.
And they won't, because i bet hammer and nails that every single data structure and what else is unsuited for multithreading.
And these are peoply that hardcode EVERYTHING, look at how problematic that one hud update with the incorrect labeling was!
And don't get me started on 2 years of modifications to the base engine.

Even if they want to, they probably can't.

[Exarch Levin]

Quote

I can tell you right now that my 5820K @ 4.3GhZ and GTX 980 will do -minimum- fps in the high 50s to low 60s

That's rather frightening as a 5820K @ 4.3 GHz is a lot more CPU than my dated 8350 (and more than double my memory bandwidth, potentially) and yet it still struggles with this game.

A $400 USD beast of a CPU overclocked by 1000MHz still struggles to pull in 60 FPS minimum in MWO. That is astonishing.

[DeathlyEyes]

I notice that when I am dead I get great frame rates looking at my dead mech. Can we get the option to disable the entire cockpit back?

[Exarch Levin]

When I'm dead, whether I look at my mech or spectate, my FPS is generally >60 and fluid. No matter what I've done, actively playing MWO has never been that smooth.

If only I could spectate myself...

[Oderint dum Metuant]

FuzzyLog1c, on 30 December 2014 - 11:42 AM, said:

http://en.wikipedia....iki/Fuzzy_logic

I ignore the MWO hardware subforums because they're filled with threads recommending non-functioning / locked CFG parameter tweaks and people (like you) that draw inaccurate conclusions from insufficient data.

Im afraid sir, Wired has you on this one, the game is highly CPU dependent to the point AMD machines suffer greatly due to poor single thread performance.
This has been tested to death across 100's of threads past and present in the Hardware forum and the effect overclocking has (which is always substantially positive)

The Hud is a good source of pain for FPS, and needs some major work.

MWO is poorly optimized/coded that needs resolving for everyone.
But it doesn't seem as high on the list as making money.

Edited by DV McKenna, 31 December 2014 - 03:27 AM.

[FuzzyLog1c]

DV McKenna, on 31 December 2014 - 03:26 AM, said:

Im afraid sir, Wired has you on this one, the game is highly CPU dependent to the point AMD machines suffer greatly due to poor single thread performance.
This has been tested to death across 100's of threads past and present in the Hardware forum and the effect overclocking has (which is always substantially positive)

The Hud is a good source of pain for FPS, and needs some major work.

MWO is poorly optimized/coded that needs resolving for everyone.
But it doesn't seem as high on the list as making money.

Where did you get that idea? Your statement literally flies in the face of all AMD and Intel CPU benchmarks (including the data provided in this very thread) where the benchmarker isn't some moron trying to game on a laptop (which are throttled based on the limited, shared CPU/GPU cooling solution very quickly) or an archaic K8 (circa 2003-2007) or Conroe (circa 2006-2007) architecture.

Overclocking your CPU provides a very non-linear performance boost in applications that are GPU limited, since the higher clock frequency marginally improves data availability to the GPU. I can underclock my primary system to 2 GHz and show you that my framerate barely budges on MWO.

This quasi-technical debate aside, I agree that the game needs serious optimization work. However, as you correctly stated, PGI--despite their newfound freedom from IGP--has not meaningfully changed its ways since the launch of open beta back in October of 2012. To date, they've spent 95%+ of their very limited technical and artistic expertise on adding more mechs, and have not made meaningful efforts to provide the features expected of modern simulation games, such as:

- SLI/Crossfire support
- In-game voice communication
- TrackIR and HOTAS/pedal support
- Contextual quick-key chat (e.g. "Help needed in Grid X")
- Comprehensible UI design (i.e.: standard "upper left to lower right" flowpath)
- Robust networking features (e.g. socially-aware user profiles and achievements)

They're also leaving buckets of money on the table (money that I think the wider userbase would be happy to spend) because they're artistically and technically incapable of implementing multi-crewed mechs (a la the Atlas DDC) and hangar modules (a la Star Citizen) that your friends can visit, as well as a more robust, persistent world in which the clan invasion could occur. These features would make this game more attractive to a larger audience that may not be as interested in participating in the actual battles themselves.

Edited by FuzzyLog1c, 31 December 2014 - 11:49 AM.

[Oderint dum Metuant]

FuzzyLog1c, on 31 December 2014 - 11:31 AM, said:

Where did you get that idea? Your statement literally flies in the face of all AMD and Intel CPU benchmarks (including the data provided in this very thread) where the benchmarker isn't some moron trying to game on a laptop (which are throttled based on the limited, shared CPU/GPU cooling solution very quickly) or an archaic K8 (circa 2003-2007) or Conroe (circa 2006-2007) architecture.

Overclocking your CPU provides a very non-linear performance boost in applications that are GPU limited, since the higher clock frequency marginally improves data availability to the GPU. I can underclock my primary system to 2 GHz and show you that my framerate barely budges on MWO.

Overclocking from stock AMD machines when they have had FPS problems has improved FPS performance touching nothing else on the system. MWO is single threaded performance dependent something current gen AMD chips do not excel at, the overclocking helps to bridge or paper over that crack.

And as wired told you, there are alot of threads in the Hardware Forum that show this, infact there are at least 2 threads on the first page that are in depth performance logs, you don't have to look far.

Performance in MWO is horrific, because it is simply un-optimized or poorly coded there is no escaping that.
But to deny that overclocking chips on both sides of the divide doesn't help is woefully inaccurate and displayed fully in the HW forum threads.

Edited by DV McKenna, 31 December 2014 - 11:41 AM.

[Kain Demos]

I've come to the conclusion that their optimization is so poor that anyone that says they never see anything less than 60fps is mistaken or outright lying. I got sick of seeing the intense framerate drops in CW warfare as well and upgraded an already high end PC (overclocked, got faster RAM and a GTX 980) that really shouldn't have needed any upgrades to run this game well. The results were improvements but in CW in very intense fights with lots of mechs in close proximity I have still seen some drops as low as 26fps.

Honestly, I blame the servers. It doesn't matter how fast your computer is it can't render faster than it receives the instructions.

Edited by Kain Thul, 31 December 2014 - 01:13 PM.

[Katrina Steiner]

You can believe me or not but I actually get anywhere from 80-100 fps. I just upgraded from a dual core/8GB ram/1GB radion video card sff pc that got about 30-40 fps to a quad core I7-4790K that runs 4Ghz/8GB ram / SSD drive and a Nvidia GTX 460 1GB video. I also run the settings at ultra high.
To alleviate the 80-100fps jumps I run the game vsync so I always get 60fps so no jumping. I believe the processor makes all the difference.

[Bors Mistral]

What resolution are you playing at, Kat? Something like 900p?

Optimization in MWO is pretty abysmal, especially when it comet so simple effects like smoke that seem at times to choke any system. Wouldn't be surprised if there was an high-poly ocean getting rendered all the time under every map.

My frame-rates are all over the place, from 30-something to 70-ties. And the slowdowns are usually either from a bunch of mecs that are on the other side of a mountain and shouldn't be rendered at all in the first place, or a simple weapon heat vapour totally killing it.