18 replies to this topic

[NARC BAIT]

sorry if the resolution doesn't fit on your non 4K screen ... but clearly that's not my concern ... my concern is the myth that performance is CPU bound, which is why I bothered to graph those values, and to that end, it would be nice to see some *recent* results from people running intel's ... of course without an actual benchmark, the only thing we can *really* compare is the minimums ....

please note that the scale is not the same on every graph, both vertically and horizontally, and that I've only bothered to mark points *beneath* the 10th percentile ...

[GweNTLeR]

Thx for the graphs, was thinking about doing the same investigation(thinking about pc upgrade). Maybe it is mem bound? No clue how to check it however.
I can check it on my i5-4590, but I dont think it would be of much help since it is very close to your 1600X in terms of single thread and I don't have a 4K monitor.

Edited by GweNTLeR, 30 August 2017 - 02:19 AM.

[xWiredx]

Min, max, average, and frametimes should all be comparable. Unfortunately, I have not had time to play MWO in many many months now, and it doesn't look like I will for weeks still (or maybe ever).

The game is indeed CPU-bound, though. People that don't understand how it works always say "it can't be CPU-bound because my CPU doesn't reach 100%" but every single time they improve their CPU situation their performance gets better. Adding cores doesn't help beyond a point because MWO is very poorly multi-threaded.

We do also know that VRAM is a major thing for the bigger faction warfare scenarios (if that is still a thing). At 1080p, during the initial few Community Warfare patches I noticed over 3GB of VRAM being used at 'very high' settings.

What MWO doesn't really seem to require is a massive amount of GPU power. My GTX 980 very happily plays the game at 'very high' just fine at 1080p (which is almost assuredly the 'standard' resolution for MWO players as many play on potatoes).

[NARC BAIT]

xWiredx, on 30 August 2017 - 04:48 AM, said:

The game is indeed CPU-bound, though. People that don't understand how it works always say "it can't be CPU-bound because my CPU doesn't reach 100%" but every single time they improve their CPU situation their performance gets better. Adding cores doesn't help beyond a point because MWO is very poorly multi-threaded.

honestly, that's the echo of the poorest excuse that goes around, and its the justification that they have used to not make it any better, to quote apple, its not a crappy design, you've just never learned how to hold a phone ... the problem with coders, is you can tell them something is inefficient, but if it works most of them wont care, and the excuse has shifted the blame from where it has belonged ...

xWiredx, on 30 August 2017 - 04:48 AM, said:

I noticed over 3GB of VRAM being used at 'very high' settings.

I get higher than that ... whats the point of having 6 gb of graphics memory to only a third of it ... I can actually get a baseline on how many 'unpainted green mechs' there are based on how much graphics memory has been allocated ...

xWiredx, on 30 August 2017 - 04:48 AM, said:

What MWO doesn't really seem to require is a massive amount of GPU power. My GTX 980 very happily plays the game at 'very high' just fine at 1080p (which is almost assuredly the 'standard' resolution for MWO players as many play on potatoes).

whats the 980 'usage' get upto at 1080p ? 40 ? ... you've honestly probably got more 'power' than me, .... you should be able to pull off at least 1440p at a decent frame rate ....

might profile a few games with the HUD off, ie, no map ... if happen to TK anyone, ill just blame you ...

[Alcom Isst]

What software did you use to sample and generate these charts? I would love to make the same thing for my odd i5-4690k@4.4GHz + GTX 1080ti build.

[Thorqemada]

Well, you can say that the CPU is the most performance defining Piece of Hardware for MWO when you upgrade your GPU twice but not your CPU (ok i upgraded once but very early in the beta) and the min and average fps pretty much stay the same while every other 3D-Game doubled, trippled and quadruppled in performance.

You can also say a game is CPU bound when changing GPU-Bound graphic options to lower settings make nothing but changing CPU-Bound options giving you FPS boosts and that totally is the case.

Ofc even for MWO there are GPUs that are to weak and make it look like MWO could be GPU-Bound but these are entry-level GPUs you cant take for benchmarking.

[NARC BAIT]

Alcom Isst, on 30 August 2017 - 10:10 AM, said:

What software did you use to sample and generate these charts? I would love to make the same thing for my odd i5-4690k@4.4GHz + GTX 1080ti build.

last time around, a few years ago, we used afterburner/excel ... this time I used hwinfo64 for logging data points (far better), and mapped it into a google spreadsheet, which was pretty painful the bigger the data set got ...

Thorqemada, on 30 August 2017 - 05:01 PM, said:

You can also say a game is CPU bound when changing GPU-Bound graphic options to lower settings make nothing but changing CPU-Bound options giving you FPS boosts and that totally is the case.

it 'bounding' is not from the engine, its the code ... and I'm very sure it could have been better, years ago, if the primary objective hadn't been mech packs, and the general attitude wasn't 'buy an intel, like we use' .... I have found configurations that can redline all cores, mostly within the streaming system, its possible to do, but doesn't bring any advantage, or last for that matter ... the engine is configured to take advantage of at least 6 cores, with the potential for the job system to take it further than that ...

a big part of the problem comes from abuse of flash elements, ie the map ... even when you cant see it (like 3pv), its spazzing out at points, like when your view area drastically changes and it starts calculating a bunch of things that you a) cant actually see and are too far from to have it show on the minimap .... a few days ago I was messing around with some aspect ratio stuff, which completely exposed how the flash HUD elements don't scale to aspect ratio, which amusingly, is about the only 'advantage' vector based flash is meant to have over, anything ...

the devs must have been drunk when they did the numbers for the particle system, and in desperate need of a calculator ... you can field 24 mechs in a game, each one can easily generate three or four particle effects, but every individual LRM makes multiple at high enough 'quality' if there are 200 tubes in total in a game, even the highest normal settings don't add up ... and I've seen plenty of games with more like 200-400 total tubes in quick play ...

Edited by NARC BAIT, 30 August 2017 - 10:19 PM.

[visionGT4]

Nice work here and thanks for putting in the effort. MWO loving high cpu clocks is a direct result of the 'poorly' optimsed code, its debatable if that technicaly means CPU bound. However theres no doubting the fact that MWO at 4ghz on a 'modern' Intel cpu's runs a truck load better than the same CPU would at 3ghz.

LRM spam is the only time i see framerate drops (6700k @ 4.7 + 980m SLI) with everything on low except textures. Disabling 1 card makes no differance in this situation - its all about the cee pee u's

[Thorqemada]

Yeah, i understand that theoretically CryEngine in a different configuration is capable of doing parallel processing better with a positive Performance Impact - but exactly that is the Thing - it does not in MWO!

So the game is setup/coded pretty CPU bound bcs PGI made it so and nothing but PGI can change that,,,

[xWiredx]

NARC BAIT, on 30 August 2017 - 06:46 AM, said:

honestly, that's the echo of the poorest excuse that goes around, and its the justification that they have used to not make it any better, to quote apple, its not a crappy design, you've just never learned how to hold a phone ... the problem with coders, is you can tell them something is inefficient, but if it works most of them wont care, and the excuse has shifted the blame from where it has belonged ...


I get higher than that ... whats the point of having 6 gb of graphics memory to only a third of it ... I can actually get a baseline on how many 'unpainted green mechs' there are based on how much graphics memory has been allocated ...


whats the 980 'usage' get upto at 1080p ? 40 ? ... you've honestly probably got more 'power' than me, .... you should be able to pull off at least 1440p at a decent frame rate ....

might profile a few games with the HUD off, ie, no map ... if happen to TK anyone, ill just blame you ...

Eh? Being 'bound' is a function of the end result. You don't say something is not CPU-bound just because the code is not optimized for multi-threading. If it isn't, being CPU-bound is still a product of that lack of optimization. The game is indeed CPU-bound, and there is nothing you can do about it except buy chips with higher IPC and overclock them higher. Is it ideal? No, why should the burden be on the gamers to compensate for developer inefficiencies? It shouldn't, but it doesn't change the end result. At this point, those optimizations are probably never coming, so it does fall to us to deal with it in whatever ways we choose.

My VRAM measurement was, again, at 1080p and on a very old version of community warfare. Things have changed a lot. If you change the resolution, of course more VRAM is going to be used. Not sure what your point is there. People playing MWO at 4K are a very very tiny minority.

Of course I could run MWO at 1440p, but I don't have a 1440p monitor and I abhor the lack of UI scaling. There is literally no point in increasing the resolution for this game beyond native res as long as AA works. You get a slightly sharper image with a 4k monitor, sure, but you have no advantage in battle. You only have a disadvantage when using the UI. Though I was planning when building my next system to get a nice 2K screen... more for productivity than anything else.

[G4LV4TR0N]

You might need to try another game on Forest Colony to compare results. I do wonder were those frame drops there caused by day/night shifts. I'd also like to see Terra Therma and Siege maps like Emerald Taiga because those are known to cause certain issues.

Edited by G4LV4TR0N, 02 September 2017 - 12:35 AM.

[NARC BAIT]

visionGT4, on 30 August 2017 - 11:27 PM, said:

LRM spam is the only time i see framerate drops (6700k @ 4.7 + 980m SLI) with everything on low except textures. Disabling 1 card makes no differance in this situation - its all about the cee pee u's

RAM speed has a noticeable impact at my end ... different clocks essentially set different 'low points' .... anything described as asynchronous makes the fork of the engine crash ...

xWiredx, on 01 September 2017 - 08:27 AM, said:

Eh? Being 'bound' is a function of the end result. You don't say something is not CPU-bound just because the code is not optimized for multi-threading. If it isn't, being CPU-bound is still a product of that lack of optimization. The game is indeed CPU-bound

it just seems to me, that its like describing your cars maximum speed by the reference to the limit of the road your driving on ... anyway, its kind of splitting hairs, but allowing them to describe the problem as being someone elses fault, excuses them from having to fix, someone elses fault ...

xWiredx, on 01 September 2017 - 08:27 AM, said:

My VRAM measurement was, again, at 1080p and on a very old version of community warfare. Things have changed a lot. If you change the resolution, of course more VRAM is going to be used. Not sure what your point is there. People playing MWO at 4K are a very very tiny minority.

the resolution has very little impact to the texture memory used, which tends to the majority of where the VRAM is burned here, the settings in the menu have a big impact, but they don't really describe the whole range very well, as in, very high isn't the absolute maximum, and low is far above the absolute minimum ...

I've scaled down from 4K, to 3072x1728, with the main difference being how far I have to move the mouse to cover all pixels on the screen ... scaling downwards is easy, and scaling upwards is possible, but if you set your resolution to 1.5x within a full window, you end up losing visual information along the way ... the same as DSR in effect ... what you should do, is you should disable the LOD system ... and watch the distance get MUCH cleaner than what AA will do for you ... that's 'e_Lods = 0' into a user.cfg/tuning.cfg ... your cross map gauss accuracy will thank me later ..

xWiredx, on 01 September 2017 - 08:27 AM, said:

Though I was planning when building my next system to get a nice 2K screen... more for productivity than anything else.

that's why I got a 4K screen ... its 4x 1920x1080p ... with the native windows 10 window location shortcuts, why would you only have one task running ... I had multiple monitors for years, and have been pretty happy to squish four into one, and saving on desk space ... that's more space, for more mess, looking at the desk in front of me ...

G4LV4TR0N, on 02 September 2017 - 12:05 AM, said:

You might need to try another game on Forest Colony to compare results. I do wonder were those frame drops there caused by day/night shifts. I'd also like to see Terra Therma and Siege maps like Emerald Taiga because those are known to cause certain issues.

yeah, those graphs are more for baselining, than absolute numbers ... and I am going to do more, but well ... have you ever had that time you cleaned something, which made the rest of it look worse, and your suddenly doing 1000% more than you had planned ... and essentially I've identified a few other issues present during the graphing .... yanked the hard drive that I was dumping video captures to for underperforming, re-organised a shitload of data between drives, been working on the memory timings .... and in true style, I declared to be done with it, the morning after poked one more thing, and then spent a whole day messing, teasing and testing ... but for now I've backed off going for even higher memory clocks, and am evaluating my current user.cfg ... the shadow system is doing my head in a bit ... it looks mostly good at the moment, but comes at a reasonable cost ... in saying that, a few times I've decided to change directions because I could see the shadow of a SNV before I could see the mech ... that's the point right ? gain legitimate advantages ?

anyway, your not likely to catch me in FP any time soon ... and the reason is three fold, firstly, almost all my 'preferred' mechs are light IS, secondly I cant have a valid drop deck with four lights and be over 160t, and thirdly, theres IS, in FP ... they tend to make the PUG queue look good and organised ... maybe being in a 12 man roller is another story, but is that really a GG ...

in terms of TOD change, I've seen it plenty, I'd consider a drop to 45-48 fps 'normal' for my system during the init or end of the sun cycle, but that's more like my long term average than the current ... the sun is not a deferred light source, at sunrise it adds load, and at sunset it 'frees' its allocations, generally you will only see one side of the change during a game, in terms of testing it ... on the testing grounds, on frozen city/river city the sun sets about 20 minutes or so after spawn .. on terra therma, will mostly cause issues with the particle system ... which runs in a serial fashion, that is, one effect after the other, and if there are more than X, run them in half rendering mode, which brings its own performance hit too btw ... fog effects over the view distance can get a bit mental around the middle looking in any direction, and may bring people close to the limit, before anything has happened ...

[NARC BAIT]

second batch of graphs, testing the shadow system ... which killed performance on tourmaline desert, and had some big impact on forest colony ....

[G4LV4TR0N]

So Forest Colony is dropping frame rates, not surprised as it's using some different terrain/effects draw system. It's also one of those maps, along River City, that supports day/night shifts which many people consider fps killers. While I never measured it, almost sure that day/night shifts are definitely more CPU- than GPU-bound.

[SCHLIMMER BESTIMMER XXX]

xWiredx, on 30 August 2017 - 04:48 AM, said:

I have not had time to play MWO in many many months now, and it doesn't look like I will for weeks still (or maybe ever).

DRAMA QUEEN

xD

[Thorqemada]

On some maps are slow down spots like Caustic E5 if i remember right, where for no reason i drop 20 or 30 fps.

Map with much Water are also ressource hogs where FPS Drops likely happen - or Maps that have already a high object count and with 24 Mechs you come to the Limit of your PC having set the Object Detail Level to very high.

Usually i set Environment to low as i dont like Particles below High as PPC Trails and Missiles become ugly - normally i should set Particles to the lowest Settings bcs it is the very most ressource hog of this game aside of the HUD which as improved over time.

But i cant settle myself down with PPC Trails that loo like a flying spermatozoid...an no Missile Trials - it is unacceptable on the medium Setting - it would be acceptable on low Settings.

Edited by Thorqemada, 05 September 2017 - 07:29 AM.

[NARC BAIT]

Thorqemada, on 05 September 2017 - 07:28 AM, said:

On some maps are slow down spots like Caustic E5 if i remember right, where for no reason i drop 20 or 30 fps.

cant say I'm overly familiar with that one ... if you go from E5 to D5 over the terrain then that could cause some issues with the stuff on the other side, but it shouldn't have a big impact ...

Thorqemada, on 05 September 2017 - 07:28 AM, said:

Map with much Water are also ressource hogs where FPS Drops likely happen - or Maps that have already a high object count and with 24 Mechs you come to the Limit of your PC having set the Object Detail Level to very high.

the only reason water tends to kill FPS is because of the reflections ... its not the actual water itself at all, its because someone from pgi was smoking crack one day and decided that water should reflect everything, honestly the dude probably should have gotten a decent 5mw laser and gone out to a body of water and had a look at what it looks like ... and then tried to get 20m in the air ( like a mech ) and repeated the process ... you wouldn't have decided to make the water the way it is at the moment ....

Thorqemada, on 05 September 2017 - 07:28 AM, said:

Usually i set Environment to low as i dont like Particles below High as PPC Trails and Missiles become ugly - normally i should set Particles to the lowest Settings bcs it is the very most ressource hog of this game aside of the HUD which as improved over time.

But i cant settle myself down with PPC Trails that loo like a flying spermatozoid...an no Missile Trials - it is unacceptable on the medium Setting - it would be acceptable on low Settings.

I honestly cant stand the 'trails' from the higher particles settings ... especially the way they fade in blocks, and it makes it super obvious that the engine has no degree of wind, ever, at all, at any height .... the trails from the aircraft delivering an airstrike really does my head in, not to mention the amount of times I've watched those planes fly straight into terrain, because yknow, that's how planes work ... in terms of PPC's, I don't mind the higher setting, but it really doesn't 'add' to the in game immersion, I'd honestly think that the higher settings look less believable and actually leave you in a tactically worse off place, at the lower settings the trail behind the PPC highlights where it came from, and at the higher settings, that tactical data is blurred out, assuming you had the spare time to absorb the info ...

and then there is the HUD ... which, hasn't really gotten better at all ... there was a day that it was, but everyone complained, and they hot fixed it, giving us worse performance than prior to it all ... oh it was funny to hear everyones complaints about the disorientation it caused on that day ... the biggest impact is the way the map is done, or more so, the way the data is processed to generate the map ... which happens, wether its visible or not ... even in third person where the map is not shown, CPU cycles are wasted with it ...

[NARC BAIT]

another dump of graphs, in two groups

[NARC BAIT]

new round of testing ... well, its always a new round, buy anyway .... for the majority of these I had dropped the resolution into puke mode 1080p ... that's what you guys like right ? 1080p ? I seriously nearly vomited ...

very high settings - no user.cfg - 1080p
polar / domination / min = 37 / avg = 76 / max = 173


very high settings - no user.cfg - 1080p
caustic / domination / min = 22 / avg = 91 / max = 181


low settings - no user.cfg - 1080p
polar assault / min = 46 / avg = 118 / max = 200


low settings - no user.cfg - 1080p
frozen / skirmish / min = 64 / avg = 124 / max = 246


low settings - basic user.cfg - 1080p
grim / skirmish / min = 63 / avg = 138 / max = 243


low settings - basic user.cfg - 1080p
crimson / escort / min = 49 / avg = 97 / max = 174


low settings - most user.cfg - 1080p
hpg / conquest / min = 63 / avg = 123 / max = 212


low settings - most user.cfg - 1080p
tourmaline / escort / min = 55 / avg = 109 / max = 226


low settings - most user.cfg - vsync - 1080p
grim / domination / min = 56 / avg = 60 / max = 61


low settings - most user.cfg - vsync - 3072x1726
crimson / escort / min = 45 / avg = 60 / max = 60