Metro: 2033 Redux and Metro: Last Light Redux - Performance Discussion

Ah, I thought initially you were gonna show benchmark videos back-to-back of Metro 2033 Vanilla vs Metro 2033 Redux & Metro Last Light Vanilla vs Metro Last Light Redux. That would have been pretty cool to have seen the contrast - for me anyway, because I don't have the original version of Metro 2033 installed at the moment to compare the difference.

 

I'll probably end up playing Last Light Vanilla again after I finish the Metro 2033 Redux, to get a greater feeling of continuation of the story. Last Light Vanilla ran well for me (& looked good), so I can't really justify buying that one.

 

Well I just have original Last Light without any of the DLC (Ranger Mode as paid DLC, dafuq?!?), so getting Redux will be a better deal for me anyway.

 

I think I already have benchmark videos for both vanilla Metro titles in my YouTube channel. You might be able to stitch them together in a playlist for a similar review.

 

so no DOF in the Metro 2033 redux?

 

My sewing skills are terrible! ;-)

 

Results show heavily optimized PhysX but what about all other game? Do you, Fox, plan on benchmarking 4 games without Advanced PhysX turned on?

 

It's integrated into the main graphical presets like in LL and no longer a separate setting. And optimized so enabling it no longer drops FPS by 50% or more.

 

Hi James. I'm not sure what the first part of your question means ( in blue). Are you asking me to demonstrate what the benchmark results are without Advanced PhysX enabled on each of the 4 Metro titles?

 

interesting.....will consider the redux versions

 

If that is what you are asking, here's the deal... the Redux titles are optimized for improved Advanced PhysX performance. If you disable that, performance is better for the old titles, but you lose one of the optimization advantages that comes with having the Redux versions. They changed a number of things for the better, including things with the gameplay itself. It would be interesting to see how the old and new Metro titles differ with AMD GPUs. I am only guessing, but I would speculate that the Redux titles run comparative better than they do if running one of the original Metro titles if your machine AMD graphics. I would expect to see less of a benefit with NVIDIA cards since PhysX was never intended for AMD.

Here is a glance at how disabling Advanced PhysX is not necessarily a good thing to do, depending on which Metro game you are running. This can be compared with the previous chart. As you can see, the difference between game versions is not as remarkable. Feel free to extrapolate and draw your own conclusions.


(click graphs to enlarge in a new browser window)

Spoiler

Click to enlarge image in a new browser window...

 

Thank you Fox. Yes I meant all other video effects, textures, processing... all game without Advanced PhysX Looks like without Adv.PhysX Metro LL is way better on Redux (90 vs 72) while Metro 2033 is somewhat better on classic version. Thanks.

 

Sure... here's a better format for the graph. Easier to compare, I think. On a system with plenty of horsepower to drive it, things do look better with Advanced PhysX enabled.

click graph to enlarge in new browser window​

 

And actually it is obvious why Metro 2033 Classic is faster... because it is on older engine with way worse looking, less effects and textures. Not like Metro LL which got way less effects on top of non-Redux version.

 

Yes, that is correct, but it is only faster depending on how you have it configured. If not set up optimally it can be much slower as well as not looking as nice as 2033 Redux.

The part I do not know is if the results with be similar with other hardware. All of these test are with GTX 780M SLI, so something else may behave somewhat differently.

Metro Last Light Redux has them all beat... excellent game. I have very happy with how both Redux titles turned out, other than the initial bugs with it not wanting to launch.

 

Metro LL was cool and better and now Redux... I coudn't run LL with Advaned PhysX at all. I recall myself very pitty because I had 3920XM CPU with a midrange mobile GPU and CPU PhysX was same BAD as GPU. All was 10 fps. That is why I am surprised about PhysX optimisation and maybe will play in future in LL Redux.

Atmosphere was not full without PhysX)

 

What about performance?Do both metro redux run better than original versions?

 

Yes they do... smoother and look better. You may need to adjust the settings to suit your hardware, but I consider both Redux titles to be a major improvement over the originals. These are totally remastered, not just minor game engine changes, but also changes to content and the inclusion of previously released DLC. The changes to Metro 2033 Redux are bigger than those to Last Light Redux because Last Light already had the newer and better game engine. Watch the video below for a comprehensive explanation of everything that has changed.

Metro Redux Review | Attack of the Fanboy

<iframe width='853' height="480" src="//www.youtube.com/embed/IZwVvl71JnM" frameborder='0' allowfullscreen=""></iframe>

 

i buyed metro 2033 redux cause i finished metro last light already.I have a weird issue?game runs better with 0.5 SSAA than SSAA off?anyone getting the same problem?

 

i talk about almost 50% decrease if i turn ssaa from 0.5 to off

 

DOH! That's because 0.5 SSAA renders your game at half native resolution (0.5!), and then it's upscaled to your screen. x2 SSAA on the other hand renders your game internally at twice your output resolution & then it's scaled down to your output resolution. This is all just the way SSAA works. (Therefore, 0.5 SSAA is worse visual quality than SSAA off!)

 

Honestly, I don't understand why it even exists. SSAA is garbage as far as I am concerned. Even with an insanely powerful machine it carries so much overhead that it is utterly worthless in my opinion.

 

I agree, but it is the purists unadulterated form of AA! Very costly though, but I have had sucess in the past downsampling some games like F1 2012 and Call of Juarez (a bit like SSAA) using the NVidia Control Panel & creating custom resolutions beyond my panel's resolution and choosing the correct option in the NVidia Control Panel to downsample it using the GPU scaling option. (not sure if I've used all the right terminology there!). It's not worth it though, x16 CSAA and x16AF is less costly & looks better I think.

 

SSAA is there because a lot of games nowadays don't play nice with MSAA (or CSAA, just another form of MSAA) due to the popularity of deferred shading techniques. It can look amazing, but it's probably harder to tell in Metro games due to the dark atmosphere and post process AA that's already on by default. Pixar movies and such look so good partly because they have extremely high levels of supersampling. (And they're ray traced too, but that's another discussion altogether.)

LOL Dmitri (yes I called you Dmitri, this is the Metro after all ) 0.5 SSAA is subsampling, that's why your FPS went through the roof. Are you sure you don't need new glasses or something? Because the severe blurriness should've jumped out at you right away.

 

Yeah, SSAA is same as RAW in video. It may be twice the overkill about performacne but the Best in its kind... if you ever dare to compare 4x SSAA with 4xMSAA LOL.

It is really good for me for HALO game for example (which is offtopic, I know). Also it may show you new details on the monitor while all other AA techniques just either blur or reconstruct picture from existing pixel information in GPU.

 

For some reason forcing MSAA in Halo PC has never worked? Anyway, I agree, SSAA and downsampling old non-demanding games can really make them pop.

LOL great analogy. FXAA would be YouTube then.

 

Nothing is working on HALO PC except SSAA and SMAA/SweetFX

 

SMAA is the best form of AA available right now. It's efficient; 1x and 2Tx being a max hit of like 2fps, with 4x being less demanding than 4x MSAA, but looking like FXAA without the blur, MSAA, and TXAA without the muddying of textures in motion all rolled into one. It's not as good as 8x MSAA or above (16xQ CSAA, 32xQ CSAA, etc) and it's not as good at killing running black lines as TXAA is, but with the way it rolls everything into one like it does, combined with the low performance hit (negligible if 1x or 2Tx are enough for you) without much if any downsides? It's the way to go.

 

Yep, but did you know you can enable AF in config.txt? I only learned that this year after doing it wrong (forcing it in GPU control panel) for 11 years.

Pffft best AA? I'd have to go with 256x SSAA of quad-SLI Quadros.

SMAA 1x is pretty nice and injectable into most games, but S2x, T2x, and 4x all require support in the game or engine itself. AMD-exclusive MLAA 2.0 is also very good. Compared to SMAA 1x, it looks just as sharp and trades better performance for slightly worse edge smoothing and sub-pixel artifacting. Nvidia driver FXAA performs like MLAA 2.0 but is the blurriest of the three. However, it handles motion the best (well duh, it blurs the most).

 

Yeah, injecting it is a bit of an annoyance. Also, nVidia can run MLAA if it's present in a game, just like AMD can run FXAA (as far as I know). Also, how does MLAA 2.0 perform better than SMAA 1x? As far as I've ever seen, SMAA 1x is basically at MOST a 1fps hit. What, does MLAA 2.0 increase FPS? O_O

I wasn't talking about injecting SMAA though, more about what devs should be using. Honestly, if you're gonna put 2x or 4x MSAA in a game, you might as well replace it with SMAA 1x, 2Tx and 4x. There is zero reason not to. The only thing I would really consider a worthwhile improvement over 4x SMAA in a game (in terms of looks) would be 16xQ CSAA or higher. It has almost the same hit as 8x MSAA (maybe a little higher) but is quite a bit sharper. It is, however, limited to nVidia. I think most people forget that CSAA is a nVidia only tech.

Anyway, I was basically talking about its efficiency while keeping great results. You can get better fine-tuned performance from other methods, like SSAA improving on... everything, really. Or CSAA surpassing sharpness/jaggies killing. Or TXAA being better at killing running black lines etc. But you try turning on all of those at once and see how low your FPS gets. I think you can actually DO it in Serious Sam 3 too; to date it is *THE* most demanding PC game I have ever seen. If you set the game's preset to "ultra" it actually leaves off a large number of effects. You can turn on SSAA, MSAA and at least 1 other kind of AA as well as a few other actual effects in-game. Running fulllscreened on my 780Ms at stock I would get barely 35fps at 1080p with all the bells and whistles turned up. Usually dropping below that count, too. And that was just in a map where nobody was around; far less if it had about 40 enemies rushing at me on-screen. Even BF4 with ultra and 200% resolution scale runs better LOL

And now I have inspiration to write a basic guide as to what one can expect from different AA methods. >_<. And the guides in my sig shall live on I guess.

 

I'm not talking about per-game implementations of MLAA, e.g. Witcher 2 (its custom MLAA looks amazeballs BTW), which both brands can run. I'm talking about MLAA 2.0, which you can only force through AMD's driver. The performance hit of SMAA 1x is typically much more than 1 FPS. You can find out for yourself by comparing injected SMAA 1x with Nvidia driver FXAA.

Actually there many reasons no to. Most deferred engines have little to no support for MSAA, which is why the higher SMAA modes and TXAA are off the table and can't be universal. Temporal AA isn't supported on multi-GPU configurations either, so there goes SMAA T2x, 4x, and TXAA, leaving only SMAA 1x and S2x (assuming MSAA support) and of course FXAA and MLAA.

CSAA is an extension of MSAA and only nominally Nvidia-exclusive as AMD has an equivalent called EQAA.

 

Fox gonna kill me... Fox gonna kill me... Fox gonna kill me... Fox gonna kill me for the way where this thread went now

P.S. octiceps, I know how to turn on AF in general in config (just create a custom DevID of your card there and write down USE AF or smth like that) but not about x16 AF

 

You remove the double slash in front of the //UseAnisotropicFilter line under your GPU brand (ATi or Nvidia). Only looks like 8x AF to me though. 16x needs to be forced through the driver.

 

Nobody is going to die, but maybe 30 lashes with a wet noodle is appropriate.

Back on topic... this is interesting. I used DDU and stripped everything out. Installed GeForce 327.23 and used my NVIDIA Inspector NIP file. Metro Last Light Redux now runs only about 4 FPS less on the average, but CPU and GPU PhysX performance are essentially identical now. This is using the same CPU and GPU settings as all of the other benchmarks.

Further interesting, using the latest GeForce drivers, using AFR2 as tbe SLI Rendering Mode gives me an extra 8 to 10 FPS on average. With 327.23 using "NVIDIA Recommended" produces equivalent results as AFR2 on the latest drivers and using AFR2 reduces average FPS about 8 to 10. So, exactly the opposite behavior.

 

Did PhysX indicator show that game actually runs from GPU PhysX?

 

I will have to check. I don't normally enable that.

Edit #1: I just checked and PhysX runs on the CPU automatically with NVCP is set to "Auto-select (recommended)" for PhysX processor selection. So, it runs on the CPU no matter whether you choose that or manually select the CPU. In practical application, the older driver (327.23) makes the best choice for Metro Redux performance without a native Redux profile when set to "Auto-select" for PhysX and the latest driver does not choose correctly using "Auto-select." That also explains why the average FPS matches the latest drivers when CPU is manually selected as the PhysX processor.

Now I wonder if there are other games that are not running as well as they could for exactly the same reason using the latest and "greatest" GeForce drivers.

This might get more interesting. I thing I may try the latest drivers and PhysX, then overwrite the profile with the modded profile I am using with 327.23 and see what happens.

Here is what that looks like right now...

Code:

Profile "Metro Redux" UserSpecified=true    ShowOn All
    ProfileType Application
    Executable "metro.exe" UserSpecified=true
    Executable "metro_benchmark.exe" UserSpecified=true
    Setting ID_0x00a06946 = 0x080410f6 UserSpecified=true
    Setting ID_0x106d5cff = 0x00000000 UserSpecified=true
    Setting ID_0x10f9dc81 = 0x00000011 UserSpecified=true
EndProfile

Wonder if this will still be true after doing that?

Edit #2: Here is what GeFart Experience adds for the Redux profile...

Code:

Profile "Metro Redux" UserSpecified=true    ShowOn All
    ProfileType Application
    Executable "metro.exe" UserSpecified=true
    Executable "metro_benchmark.exe" UserSpecified=true
    Setting ID_0x00a06946 = 0x080000f6
    Setting ID_0x1033cec1 = 0x00000003
    Setting ID_0x1033cec2 = 0x00000002
    Setting ID_0x1033dcd2 = 0x00000004
    Setting ID_0x1033dcd3 = 0x00000003
    Setting ID_0x106d5cff = 0x00000000
    Setting ID_0x1095def8 = 0x02400005
    Setting ID_0x10ecdb82 = 0x00000005
    Setting ID_0x10f9dc81 = 0x00000011
    Setting ID_0x701eb457 = 0x2241ab21 InternalSettingFlag=V0
    SettingString ID_0x7049c7ec = "웮ꑌ" InternalSettingFlag=V0
    SettingString ID_0x704d456e = "쪳⃟쀰�㒺僪娟簹鹼瀣딅ᤶ厘�빁᱌뼝褧籝譯⑉贼㱅矻粁ꗪ熟胕㖩뫨뚄ᮈ쐹" InternalSettingFlag=V0
    SettingString ID_0x7051e5f5 = "籮鸙" InternalSettingFlag=V0
    Setting ID_0x708db8c5 = 0x5da16cc0 InternalSettingFlag=V0
    Setting ID_0x709a1ddf = 0x4b1cd968 InternalSettingFlag=V0
    SettingString ID_0x70b5603f = "榻鳈⏹ꢗ" InternalSettingFlag=V0
    Setting ID_0x70edb381 = 0x24208b6c InternalSettingFlag=V0
EndProfile

Wonder what the deal is with the goofy characters? The Geforce 3D Profile Manager tool also will not re-import the profile with goofy foreign characters like that in the TXT file.

 

I bet that on R327 LL Redux will work on CPU PhysX even if you set GPU mode and not Auto-select. I guess it's the Bug which further is called Feature.

 

Well, I was not referring to SMAA injected. I was talking about in-game applications; in even the unoptimized Dead Rising 3 and stuff like Crysis 3, it's barely 3fps even with 2Tx. And yeah I know about temporal not being useful for multiGPU support, but 4x still is usable and since TXAA can use temporal filters in SLI, I think they can figure it out with SMAA. And if they can't, somebody needs to update SMAA >_>.

If something has no MSAA support, then ok. But I was mainly talking about in games where MSAA *IS* selectable, but no SMAA option is ever used; far less FXAA/MLAA. Like in Battlefield 3/4, how there is 2x MSAA and 4x MSAA but no support for other types of AA.

I have never heard of EQAA. What magic is this

 

Did some benchmarks of Metro 2033 Redux and I must say that playing it on Windows 10 gives way better results than on Seven.
Especially there are no drops on 10 even without using Advanced PhysX.

And sorry for being somewhat offtopic (that is the main reason I'm here), does anyone knows how to use secondary melee attack by the volt driver? I tried all kind of keycombos but tight rlick = zoom, left = shoot, R+Left = recharge, Melee attack = Knife melee attack.
What am I missing? Should I come close to an enemy to attack melee?