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

Here is machine #2...

METRO REDUX BENCHMARK RESULTS
9/1/2014 2:48:42 PM

Options: Resolution: 1920 x 1080; Quality: Very High; SSAA: Off; Texture filtering: AF 16X; Motion Blur: Normal; Tesselation: Very High; VSync: Off; Advanced PhysX: On;

Run 0 (Scene 1)

• Total Frames: 14222, Total Time: 171.0895 sec
• Average Framerate: 83.16
• Max. Framerate: 238.61 (Frame: 11421)
• Min. Framerate: 11.80 (Frame: 6)

METRO REDUX BENCHMARK RESULTS
9/1/2014 3:05:43 PM

Options: Resolution: 1920 x 1080; Quality: Very High; SSAA: Off; Texture filtering: AF 16X; Motion Blur: Normal; Off: Very High; VSync: Off; Advanced PhysX: On;

Run 0 (Scene 1)

• Total Frames: 14250, Total Time: 171.3027 sec
• Average Framerate: 83.22
• Max. Framerate: 238.83 (Frame: 11479)
• Min. Framerate: 22.64 (Frame: 2326)

Alienware 18
i7-4930MX @ 4.3GHz
GTX 780M SLI @ 1006/1500/1.037V
16GB DDR3-1600 @ CL11 (stock RAM)
256GB Lite-On SSD
750GBx2 HDD RAID0
Windows 8.1.1

 

I like how you're fighting the derailing hard XD.

I thought you usually keep a 4.5-4.6GHz daily clock on your CPUs though Mr. Fox? I see you usually having 4.3GHz these days. Is it more stable or heat issues even with CLU?

 

No, 4.3GHz has been my daily driver clock speed for CPUs for a very long time. I bench a lot higher when it helps the scores, but there is no reason for higher overclocks for most things, including games. I ran MetroLL Redux on both machines at 4.5GHz and 4.6GHz and the benchmark results were essentially identical to 4.3GHz. For gaming, 4.2-4.3GHz seems like the sweet spot because FPS drop with lower clock speeds in demanding games and the FPS don't get any better when clock speeds are elevated beyond 4.3GHz in the vast majority of games.

Both machines are stable at 4.8-4.9GHz, but temps are not the greatest running clocks like that... too hot for normal use. When I do benching with those clock speeds I use AC cooling. At 4.3GHz the stock cooling (with Liquid Ultra) is excellent and does not need any special help... usually under 80°C even doing things like video rendering.

 

Mr. Fox, what I'm interested in is:

1. Original 2033 vs. 2033 Redux
2. 2033 Redux vs. LL Redux
3. Windows 7 vs. Windows 8.1

There's private voice chat in-game, so we use that. There's also text chat if your Caribbean accent is too rich for us, but I think we'll be fine.

There are only 2 US servers. Connery (US West) in Vegas and Emerald (US East) in Ashburn, VA. Vulcan and I play on Connery. TBH, you're kinda stuck in a no-man's land between the US and European servers and will have similar ping to each of them, so you might as well join us. And due to the game's netcode, a higher ping with lag compensation will actually give you an advantage.

 

Haha you'll be surprised, UK is MINIMUM 200 ping. Usually 250-350. US is far far far easier to deal with, though I can easily reach 200 ping to the USA. I'll definitely hit you guys up to try again then. As for when that is, I've no idea, but we'll see XD

Also, if 4.3GHz doesn't crack 80 for you I can probably manage OCing then. Just gotta wait for that CLU XD. All the CPU power I can get helps with my streaming; not necessarily that I'm chasing FPS numbers in games per-se. I'm perfectly fine at stock under those conditions XD. You know how we squeeze every drop of power out of our things Brother Fox, if I'm streaming and my average CPU load is below 80% I'm not using enough compression

 

That run with the Alienware 18 is Windows 8.1. I'm not dual booting that machine any more because the mSATA SSD crapped out and I haven't felt like setting it up again with a spare mSATA from my junk drawer.

I may find time to show you those comparisons with the M18xR2. I seldom use Windows 8.1 for anything except Fire Strike, so it will be interesting to me to see if Windows 8 sucks on Metro compared to Windows 7 running on the same machine like it does for almost everything else.

 

OK, fair enough. It doesn't matter to me what test system it is. At the very least, it would be interesting to see just how much better 2033 Redux runs than the original and what kind of quantifiable increase 4 years of engine improvements have brought about. And when you're testing, don't forget that Advanced DoF is a separate setting that needs to be enabled in the original.

 

Titanfall has been fun for me to play as a lone wolf, I find it's possible to work together with a team in a loose kind of way even if you're not directly speaking to them (or hearing them - I turned off all mics, wasn't interested in hearing trash talk or order barking) - mind you I used to play Attrition Mode all the time in that game, so team work is probably not as empahsized as much in that mode.

 

On topic of Metro 2033 Redux, I've found that it does run better than the benchmark, with outside scenes having the lowest fps. Running my current settings [1600x900, Quality High, SSAA Off, AFx16, Motion Blur Normal, Tessellation Very High, Advanced PhysX On] I get 53fps in the benchmark, and so far the lowest fps I've seen with FRAPS running was outside where it seemed to be around 40-50 fps, didn't see below 40fps; whereas, in the demanding middle portion of the benchmark it would be 30-40fps. Inside tunnel gaming, as well as walking around stations is normally right around my screen refresh of 78fps.

 

2.7MB PhysX Patch Released for Metro 2033 Redux

Note: I haven't seen any patch notes, but my testing proves my created title above.

I've found out from my testing what the patch does. It addresses the PhysX 'issue' of it running on the CPU. Previously it was seen to run on the CPU, but now I can see it's running PhysX on the GPU during the benchmark - by viewing the NVidia control panel PhysX on-screen widget - now says GPU being used.

Unfortunately this made my benchmark run with Advanced PhysX On to be decreased from 53fps down to 43fps! My CPU usage dropped by 20% (from 80% down to 60%), which further validates what I'm saying. Good news is that by forcing PhysX to run on the CPU from within the NVidia Control Panel, it is now running at the previous 53fps avg that I had before. So, for me having a strong CPU and only 1 GPU it makes sense for PhysX to be run on the CPU - not a big performance hit.

So, post patch, I would advise all of you who are running Advanced PhysX, to run the benchmark with CPU PhysX forced vs GPU PhysX to see which results in best performance. For me I'll be forcing it to run on the CPU (unlike other PhysX games this works well with little performance hit).

(This does not fix the black screen issue on startup, and therefore I am still using the 'legal.ogv' fix to run the game - I tested this post patch too).

 

That's quite a high clock for your 780m's Mr Fox, do you game with them that high?

I'm really going to have to push my 4800mq see what I can get stable on my cores. Are you running 4.2-4.3 on all cores or just one?

 

Actually, that is not a very high overclock, or high voltage. There are some GPUs that run as much voltage and core clocks that high stock from NVIDIA. Even some GPU that are much weaker than 780M, such as the Maxwell 860M. I overclock and overvolt them much higher than that when I am chasing numbers. But, no... I don't game with the GPUs overclocked because there is no need for it. There are no games that require more horsepower than 780M SLI delivers at stock clock speeds. I only overclock the GPUs for benching.

 

Is there any visual difference between PhysX running on the CPU versus it running on the GPU?

 

Patch? Are we talking about a brand patch one in the last 24 hours or less, or the last one from several days ago? I cannot see any information that shows a patch was released, available or applied through Steam today or yesterday.

How can you tell the update/patch history? I don't see anything, but I might be looking in the wrong place(s).

 

Not that I can see, they both look the same to me when running the benchmark, apart from a significant drop in fps when NOT forcing it on the CPU.

I don't think there's any official information on the patch yet. I noticed that it downloaded 2.7MB to Metro 2033 redux this morning (so a brand new patch today, not the one from a few days earlier), so proceeded to do some testing to see what had changed - those were my findings, there's a big difference like I outlined in my 'patch post' earlier (post #110).


I'm letting you folks know about this because your Metro 2033 Redux will likely be running slower for you now post patch unless you force PhysX onto your CPU now. (Could be different for your system, but if your CPU is strong then I predict that it would be more efficient to do so).

 

Yes, I saw that post but I cannot see that a patch was applied in Steam. The PhysX files in my Metro Redux folder are the same files I had before. After a little poking around I think you may have been referring to a patch from NVIDIA? I may have misunderstood and thought you meant a patch from 4A Games rather than NVIDIA.

I am not running GFE and it there was a patch from NVIDIA, then I am guessing that I do not have it. However, your observation on using CPU versus GPU for PhysX is valid even if I do not have it. Forcing PhysX on run on the CPU produces higher benchmark scores for me also, although less than 10 FPS. At 80+ FPS the in-game experience does not suffer either way. But, it does appear that there is more PhysX enhancement using GPU. I see see the difference right away, even as early as the initial part of the benchmark where the trolley is taking off, as it looks like there are are more snowflakes falling and where the scene shifts to all of the solders near the trolley, before going under the catwalk, everything looks sharper and sort of "sparkly" compared to using CPU. Additionally, setting the 0x080F12F5 SLI bit suggested by MaLDo makes zero difference in my benchmark scores. I have tried using NVIDIA Inspector and by manual editing and importing the new configuration using NVIDIA's SLI tool. As far as I can see, nothing changes. Benchmark results are the same and GPU utilization is unaffected using that SLI bit. See below for the benchmark showing CPU versus GPU for PhysX...

PhysX on CPU	PhysX on GPU

​

 

Actually the profile is 0x080410F6 and it seems to benefit 3-way SLI the most.

Make sure you set it in the SLI compatibility bits (DX1x) line. The line below it is for DX9 games.

 

Hmm, I'll try the other one. It's not very clear from his post which one does what. I saw in the Geforce Community where some where saying the other SLI bit works better. I'll try the other one and see if it changes anything.

 

It's the one in the update at the top of his post.

 

No meaningful difference in average framerate using SLI bit 0x080410F6, at least not on this system. Debating whether I should play with whatever GFE profile adds, but there's no compelling reason to care at this point because the game runs extremely well regardless of what PhysX processor is selected. Using CPU PhysX the GPU utilization is better balanced, with both typically 94-96%, using GPU PhysX the secondary card works harder (as expected) with a very minor reduction in average FPS, but with enhanced eye candy.

See below... benchmark is more or less identical using either SLI bit.

SLI Bit 0x080410F6 with GPU PhysX	SLI Bit 0x080410F6 with CPU PhysX

​

Edit: I think I will pass on installing GFE. It's not worth installing to squeeze a few extra FPS out of a system that is already running like a banshee.

 

Robbo was right. The PhysX patch was a game update for both Redux titles released through Steam today, not a driver or GFE update.

 

Do you have a link where that is being announced somewhere? I looked in Steam earlier and found nothing posted. When I look in my game properties, I see no indication of a patch or update being applied. If that is the case, then NVIDIA and Steam must have both released updates.

Maybe I can try validating the files and see if something gets updated. I've got my old PhysX files backed up, so if I don't like it I can probably undo it.

Edit: just validated the files for both Redux titles and no visible change was applied. In both Metro 2033 Redux and Metro Last Light Redux the only "official" word on a patches are those from 8/28, which is not new... I got both back on 8/28 and they were not applicable to PhysX.

Steam Community | Metro 2033 Redux Patch Notes and Steam Community | Metro Last Light Redux Patch Notes



Manually browsing both my Redux folders, I do see one new PhysX3Gpu_x64.dll (dated today) that was silently changed in each Redux folder. I'll run the benchmark again and see if the behavior is any different using CPU versus GPU.

 

No announcement that I could find, but there are people talking about it in the forums:

Metro 2033 Redux General Discussions :: Steam Community

Metro: Last Light Redux General Discussions :: Steam Community

If you've got automatic updates turned off in Steam, it won't update the game until the next time you launch it.

 

I'm not surprised this profile works the same as all it does is remove some unnecessary flags that were in the other one.

 

Patched, but no discernible difference... now, wonder if I can find something meaningful to burn calories on...

09.02.2014 PhysX3Gpu_x64.dll Patch - PhysX on CPU	09.02.2014 PhysX3Gpu_x64.dll Patch - PhysX on GPU

​

 

I'm definitely talking about a patch from 4A through Steam, not a driver patch.

You're referring to the Last Light benchmark - I don't have that, I'm running the Metro 2033 Redux benchmark & haven't noticed a difference in the quality or quantity of PhysX effects between CPU PhysX and GPU PhysX, apart from it running faster when forced on CPU. I'll run it again today & have a close look to see if I can see any differences between the two.

***EDIT: Just ran CPU PhysX vs GPU PhysX (2 benchmark runs for each type) to scrutinize where the PhysX effects happen & the quality & quantity of these effects - they're identical (sparks, rubble destruction, smoke, explosions). Certainly as far I can tell. Note: this is the Metro 2033 Redux title - I don't have the Last Light Redux. Looks like forcing PhysX to run on the CPU is the smart thing to do for most people - assuming they have a high end CPU (e.g. i7 Sanybridge & up)

 

Yeah, I saw that it was from 4A after comparing before and after file versions. The update was done silently on my end. The only thing that I can identify that is different is the PhysX3Gpu_x64.dll file has a newer date, but it had zero affect on my system as far as I can tell. This update is on both Redux titles, not just 2033. I saved the old PhysX3Gpu_x64.dll and manually swapped them back and forth and cannot see where it made any difference. On my system the difference in PhysX effects using CPU versus GPU is subtle, but not difficult to recognize. Not sure why... perhaps something with having dual-GPU? It could also be that I am looking at Last Light Redux. I will have to look at 2033 Redux to see if there is a visible difference... there may not be any with 2033.

 

Even some sli users on Steam Forums are saying they have worse performance post patch, and that by forcing PhysX onto CPU produced smoother & better results, one system I remember was an i7-2600K + GTX 690, so I'm surpised the new patch hasn't decreased your benchmark average fps unless you force PhysX CPU.

EDIT: just looked back at your post & screenshots - so you are getting higher avg fps with PhysX forced on CPU. Well, that's inline with my discoveries so far.

 

LOL. None of the CPU which is strong for PhysX Use shadowplay to record a video of benchmark with CPU and GPU turned on for Advanced PhysX and you will see the difference. Especially about crashing concrete walls. I played a little bit with these settings in Metro LL and the reason of lower fps is that when CPU is forced to work for PhysX you still will NOT get all physX effects. If not almost all of them.

For those who have problems with advanced physX forced on CPU try to delete 3 physX dlls from Metro folder. Something like wrapper, loader etc.

 

I think you're a bit confused & not keeping up to date with the discussions on Metro 2033 Redux (& sounds like you don't own the Redux version of the game & sounds like you haven't tested it). You do know we're talking about the Redux version of 2033 right? PhysX is fine & runs better on the CPU for most people (but only for the Redux titles, not other games) - no difference in quality or quantity of PhysX effects that I can discern with close scrutiny. So, I think your last paragraph is nonsensical & the advice is not necessary - (I think it's just that you've been quickly skimming this thread).

(I can only speak for the Metro 2033 Redux, not the Last Light Redux - as I only own the orginal version of Last Light)

 

Yes, I actually said "However, your observation on using CPU versus GPU for PhysX is valid even if I do not have it" in a post you quoted, LOL. Before the patch I think (not 100% certain) PhysX only ran on the CPU even if it was set to the recommended "automatic" setting in NVIDIA CP. Some customers complained that their dedicated PhysX GPU was not being used at all. After the patch you need to actually manually select CPU in the NVCP to get the best performance. I had it set to the default automatic selection before the patch and get the same FPS as setting CPU manually after the patch. I think the patch is beneficial for those wanting to use a dedicated PhysX GPU instead of their CPU in a non-SLI configuration.

 

Yes, I misunderstood you. And you're right about the things you're saying in the part I underlined - that's what I saw too.

 

OK, for those that wanted to see my W7 versus W8 Metro Last Light Redux benchmarks, see below. As you can see, essentially no difference in average FPS.

Now, with regard to comparing the old 2033 and Last Light benchmarks to the new Redux versions, I cannot do it. Apparently, something with Redux or the Redux patches that followed has broken my benchmark programs for the older non-Redux versions. They will no longer run at all and Google Searches for the error produced nothing. The games still work, but not the benchmarks. Anecdotally, I can confirm both Redux titles are much smoother and have higher framerates than both of the original Metro titles.

Metro Last Light Redux - W8 GPU PhysX	Metro Last Light Redux - W8 CPU PhysX	Metro Last Light Redux - W7 GPU PhysX	Metro Last Light Redux - W7 CPU PhysX
		View attachment 115381	View attachment 115382

 

And, Metro 2033 Redux... again, no meaningful difference in performance with W8 versus W7, and 2033 Redux handles PhysX better using the CPU the same as we see with Last Light Redux.

I like both Redux titles better than the non-Redux. The Redux version of both plays better and looks better in my opinion. I like Last Light best, but the 2033 Redux is a big improvement over the original 2033 in my opinion.

Metro 2033 Redux - W8 GPU PhysX	Metro 2033 Redux - W8 CPU PhysX	Metro 2033 Redux - W7 GPU PhysX	Metro 2033 Redux - W7 CPU PhysX

 

Ah, your Last Light Vanilla benchmark no longer works, my Last Light benchmark works OK (non-redux version) - bizarre.

I've compared Last Light Vanilla and Metro 2033 Redux, and I think Last Light Looks better (same settings on both, bar Tessellation Off for Last Light Vanilla). Looks like they did a good job on the performance front for Last Light Redux, you got higher benchmark performance in that title than the Metro 2033 Redux.

 

Well, what is even stranger about that is now both of the non-Redux benchmarks started working again, LOL. The error message is gone. Both got updated by Steam when I logged in this morning with new VDF files dated today.

After running them the improvement with the Redux titles is more than obvious... huge improvement with the new releases as far as performance and enhanced visual quality, especially for Metro 2033 Redux. Running PhysX on CPU with both of the older titles is like a slide-slow with spots where PhysX enhancements are heaviest having framerate dips into single digits if DOF is enabled. The cynical part of me wonders if maybe they did this to accommodate AMD users, because running PhysX on CPU has always been horrible in my experience. Part of me doesn't like that, but the idea of optimizing PhysX where it is no longer proprietary and works well for AMD would be a good thing, just as it would be to do the same thing with Mantle. If AMD and NVIDIA would focus their efforts on trying to outdo one another on the hardware side and share the love where software is concerned that would be better for everyone.

 

I do think that PhysX just has been straight up poorly optimised in the past - especially on the earlier titles like Metro 2033 and Batman Arkham City. It seems PhysX is getting more optimised as times go by - Arkham Origins was well optimised for a PhysX title with good consistent high GPU usage & good framerates, and now the latest iteration of PhysX with the Redux titles; for me it's a good omen for PhysX for the future! (As long as the effects aren't diluted & dumbed down then I'm happy with the extra performance).

 

OK, here is a slide show to get a feel for performance across all four Metro titles. I am not going to re-run all of them on Windows 8 for comparison because it's a waste of time. The four examples provided already that show there is no appreciable difference in these games for better or worse using Windows 8, so this is irrelevant at least as far as both Metro Redux titles are concerned.

Sorry these are not in the right sequence. I've tried fixing it several times and the sort by original file name feature is not working correctly for some reason. I will deal with it later because I have work that needs to be done. (Or better yet, I'll make an Excel graph later on showing average FPS so it is easier to make sense of it.)

<iframe class='imgur-album' width='100%' height="850" frameborder='0' src="//imgur.com/a/Z3RQS/embed"></iframe>

 

Yes, I know that you are talking about Redux version of Metro LL and yes, I don't have Redux version and won't in closest future. Although I did follow this topic once by once and saw that some people had Advanced PhysX bug (running on CPU) and in addition that in quality/performance Redux means just newer engine with somewhat better effects and hopefully better optimizations meant that I can make assumptions based on what I knew about Metro LL (not Redux) and what I know about PhysX theory in general where Nvidia made it hellfully not optimized for running on CPU.

Looks like optimizations in Redux were way better than I expected and therefore first half of my post can be considered as wrong guess at 50/50 chance and not as "nonsence". If you call wrong guess with not less than half chance as nonsense then nonsense is practically almost all guesses by almost every kind of person.
I don't have screenshots which I had in past about benchmarks with Advanced PhysX with lower results on CPU but all Internet is full of screams "Metro LL's Advanced PhysX runs on CPU and gives horrible performance!"

Spoiler

this is review of Metro LL where on the bootom you can see performance of Advanced PhysX. This review doesn't look like very trusted but still

AS for my second half of post about bug. For those who has that bug (or had if that steam patch worked) the problem indeed was caused by PhysX files in Metro folder as I wrote. Proof for Metro Redux is here

 

Sounds like some philosophical ramblings there around what is nonsense & what isn't, but your second post is a close contender for 2nd most non-sensical! No, but putting away a little bit of my joking nastiness aside: poor performance when running PhysX on the CPU on other titles other than Redux is not really in question, so you don't really need to refer to that. And I don't get your last paragraph about PhysX bugs. From what I've seen the choice is to run Advanced PhysX on the CPU or GPU, and you choose the one that gives you the best performance - for most people it's the CPU option unless you have an old CPU. So, I don't understand the complications of your last paragraph, I don't understand the problem you are referring to, is it something different?

 

My last paragraph was about a bug which some people had and how to deal with it.

http://steamcommunity.com/app/286690/discussions/0/34096318872601975/?l=english

 

Awesome. Thanks for all the testing, Mr. Fox! Looks like 2033 Redux runs about 50% faster than the original and LL Redux runs over 100% (!) faster than LL. And yeah, CPU PhysX is a killer in both of the original titles. As is Advanced DoF in the original 2033, which easily cuts my frame rate in half or less. Better graphics and much better performance. It's amazing what a bit of optimization can do, eh? Ubisoft should take a page out of this book.

 

Cool, I understand what you're getting at now. That 'problem' is a thing of the past since the patch a couple of days ago - now it runs as default on the GPU, with users having to force PhysX to run on the CPU for best performance in most cases - this is what I had said way back in post #110 & was the reason for that posting, which happens to be the first post you replied to (looks like we've come full circle!).

 

The answer is quite apparent: It's due to the next-gen consoles and 4A games having to rework the PhysX effects to run on their AMD APU's. Not that I consider it a bad thing, at all, as it benefits us all as PC gamers and I have a general disdain for proprietary standards unless they're a necessary evil to achieve the best results.

 

It's not a bad thing as long as it ends there. If it makes it possible to have an acceptable experience playing games with lame hardware, there could be an unfortunately dumbing down of everything deemed "more than necessary" to achieve playable results. It's kind of pathetic when standards need to be lowered just to make room for the bottom feeders. Oh wait... it's already happening. *flashes dirty look at Ultrabook and tablet gamers*

 

I wanna point out that while there are a lot of extra features all around in the redux for 2033, the original game's textures were almost always far nicer. So there WAS some dumbing down. I more consider the 2033 redux a sidegrade; better animations/performance/etc but worse fire effects, textures and the game's a lot brighter (though not always a bad thing). I've not played 2033 Redux, but I've seen a bit of it from videos/comparisons/etc.

So we sacrificed textures and gritty atmospheric "feel", but we got a MUCH better running game apparently, as well as PhysX on CPU. So... yeah. I'll take it as a win; if for nothing else than to say "finally, a dev learned to optimize".

 

OK, these charts should be a lot easier to look at... based on the benchmarks already posted...

Spoiler

 

I never managed to get into the original Metro 2033 - I found the gameplay too clunky, it feels a lot more refined in the Redux version, more like Last Light Vanilla - which I really enjoyed. I'm enjoying the Redux version, and will definitely complete it this time! Funny that you mention textures, I could have sworn that I'd read on game reviews that textures had been re-worked in Redux 2033 for the better, but I did hear that they dumbed down some aspects of the lighting, and I have a suspicion that the PhysX effects are of less quantity in Redux 2033. All in all, for me it's definitely an improvement over the original Metro 2033, I even like the atmosphere better - the original was just TOO grimy and grey (I know it was supposed to be like that, but it was just too oppressive I thought).

 

Here's a screen recording of both Redux benchmarks...

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

 

Yeah, besides the removal of volumetric lighting in some scenes, which I'm not entirely displeased with as it looked out of place and absolutely tanked performance in the original game, I think Redux is an all-around graphical improvement over 2033. And the main difference in lighting can be attributed to the global illumination in Redux. Well, you could force GI through the config file in the original 2033, but it had serious visual side effects. And here you can see that the texture work in Redux is clearly superior: original vs. Redux.

In any event, I'll be getting both Redux titles as Mr. Fox has convinced me that the performance improvements in LL Redux are worth having. :thumbsup: