Metro 2033 Performance review

[cbn]

I was tempted, but I'm still trying to decide which platform to get it on. I need a new video card if I'm going to play it on PC. Definitely not buying a video card until the whole Fermi/Evergreen debate is settled.

Are the extra graphics setting really worth it for PC?

Has anyone tested how well this game scales with extra cores and hyperthreading?

 

[MarcVenice]

Maxed settings, especially outdoors, when those err, zombies would appear, fps would drop down to 20ish.

 

[Apocalypse23]

This game looks like FEAR on steroids...not very visually appealing imo.

 

[Attic]

This game is kicking my computer's butt. Q9550 @ 3.6, 8gb 1066 and a 5870. I'm trying to run it in DX11 mode and when I get in a firefight it feels like I'm swimming at the bottom of the ocean, things feel slower and floaty, not smooth enough. Anyone with a similar setup have recommended settings and tweaks?

I run DX11 1280x720 on High, using AAA and 16xAF, in DX11 I turned off tesselation and DOF. This gets me pretty steady 60fps vsync on.

First things to turn off in order of how unimportant they are according to me.

Advanced Physics (typically no FPS hit, but when a grenade explodes my FPS tanks for 10 seconds while dust settles or whatever, it's weak, I turn it off and I don't get huge frame slowdowns fom the lousy physics implementation in the game)

DX 11 Effects (turn off Tesselation and DOF, DOF is kinda cool but kills frames, Tessalation I can't tell difference)

Graphics Quality Use High instead of Very High

Turn off MSAA4X

 

[StrangerGuy]

This game looks like FEAR on steroids...not very visually appealing imo.

Shades of black, brown, grey and more black.

 

[cbn]

Tessalation I can't tell difference

Is this because you are running non-native resolution?

 

[thilanliyan]

Game looks pretty good although a bit bland...but then I'd expect post apocalypse to look bland.

 

[tviceman]

NPC models aren't the best but the environment graphics/effects going on in those screen shots look great, IMO.

 

[MrK6]

*snippity*
Maxed settings, especially outdoors, when those err, zombies would appear, fps would drop down to 20ish.

Mark, thanks for those, really informative. What are the specs of your rig?

I think the volumetric lighting in some of those shots is fantastic, as well as some textures (great resolution). The rest, on a whole, I dunno, I still don't see where it warrants the frame drop, maybe it looks a lot better in motion. It'd be nice if they put a demo out.

 

[AznAnarchy99]

Mark, thanks for those, really informative. What are the specs of your rig?

I think the volumetric lighting in some of those shots is fantastic, as well as some textures (great resolution). The rest, on a whole, I dunno, I still don't see where it warrants the frame drop, maybe it looks a lot better in motion. It'd be nice if they put a demo out.

I feel the same way. This game makes a 5870 drop to 20fps? I mean the lighting is cool but other than that, doesn't look amazing.

 

[Sylvanas]

This is probably going to be Fermi's poster boy game and being a TWIMTBP title I'm interested in what Fermi's performance will be like running this maxed considering they've probably had ample time to optimize for it.

Of interest is what the difference between 'GPU PhysX' and 'Advanced PhysX' variables are which they tested. If 'adv physics' is just the physics computed on the CPU shouldnt that be the same as 'PhysX off' ? Why make another variable?

 

[MarcVenice]

I feel the same way. This game makes a 5870 drop to 20fps? I mean the lighting is cool but other than that, doesn't look amazing.

No, this game makes a HD 5970 at 1680x1050 drop to 20fps. Ouch

Other specs are 3,4GHz q6700, 4GB ram, yada yada.

 

[cbn]

No, this game makes a HD 5970 at 1680x1050 drop to 20fps. Ouch

Other specs are 3,4GHz q6700, 4GB ram, yada yada.

The game developer claims Metro 2033 scales to 16 cores.

Have you seen any performance reviews using Core i7 or better yet Gulftown with hyper-threading enabled?

 

[blanketyblank]

The game developer claims Metro 2033 scales to 16 cores.

Have you seen any performance reviews using Core i7 or better yet Gulftown with hyper-threading enabled?

I don't see how that'd make a difference assuming it's the GPU holding it back in those circumstances. Possible, but unlikely to bring the fps down to 20 fps with an overclocked quadcore unless there's a shitload of physics calculations going on.

If the graphics settings were set down to low or medium and still had the dip though that would show it was probably CPU.

 

[toyota]

The game developer claims Metro 2033 scales to 16 cores.

Have you seen any performance reviews using Core i7 or better yet Gulftown with hyper-threading enabled?

they said it scales like that when the game is not gpu limited but clearly it is gpu limited in DX11 with all graphical features enabled. I would like to see a cpu comparison though as I bet quads and especially i5/i7 are getting utilized pretty well.

 

[cbn]

I don't see how that'd make a difference assuming it's the GPU holding it back in those circumstances.

I definitely realize the game takes a major hit with all graphics settings turned on. I am just curious to see if CPU scaling can help mitigate some of that performance hit.

they said it scales like that when the game is not gpu limited but clearly it is gpu limited in DX11 with all graphical features enabled. I would like to see a cpu comparison though as I bet quads and especially i5/i7 are getting utilized pretty well.

Well maybe with some graphics features enabled CPU core scaling will help.

I am particularly interested to see if enabling hyperthreading helps the i7.

 

[x3sphere]

No, this game makes a HD 5970 at 1680x1050 drop to 20fps. Ouch

Other specs are 3,4GHz q6700, 4GB ram, yada yada.

May be a CPU limitation? I can run this fine under DX11 with a 5970 at 1920x1200. Never seen the frames drop to 20. Usually hover around 30-40. Gameplay vid here..

http://www.youtube.com/watch?v=Eg_RcB91x2o

It's impossible to run it at 2560x1600 though, so bad news for Eyefinity users and those with tri or quadfire setups. Hits a VRAM wall and starts stuttering at 1FPS. ATI really should've stuck more than a 1GB framebuffer on these cards.

 

[Grooveriding]

I definitely realize the game takes a major hit with all graphics settings turned on. I am just curious to see if CPU scaling can help mitigate some of that performance hit.



Well maybe with some graphics features enabled CPU core scaling will help.

I am particularly interested to see if enabling hyperthreading helps the i7.

My i7 is at 4.2 with HT on, 5870 crossfire. 2560x1600, DX11, tessellation on, depth of field off, settings to very high, the game is -barely- playable. High action scenes it enters unplayable territory.

It does look good, I'd say this is the best looking game out bar Crysis, which looks better. The lighting and smoke effects are impressive, there is way too much use of motion blur though, to the point it looks silly and overdone, and there is no option to disable motion-blur ??? Why ?!

I'm betting performance on 5970 and crossfire setups will improve when ATI implements a driver update with a profile for the game. Albeit this is clearly nvidia's go-to marketing game for Fermi's release. Expect to see the game bundled with the 470 & 480 and review sites and forum marketers pushing benchmarks of this game on Fermi hardware like nobody's business. Batman AA, anyone ?

The game is playable at max at 1920x1200 on my setup though, unfortunately it looks not so hot on my panel at non-native res.

 

[Lonyo]

The game developer claims Metro 2033 scales to 16 cores.

Have you seen any performance reviews using Core i7 or better yet Gulftown with hyper-threading enabled?

Might be interesting to see a Gulftown with extra PhysX on running on the CPU with high detail settings on some NV card vs same but PhysX running on GPU. If the game can have extra PhysX enabled and running on the CPU.

If it's GPU limited, then turning on extra PhysX stuff would (should) drop frame rates, but what if you shift that onto a CPU which isn't limiting you... especially one with lots of cores.

 

[Lonbjerg]

Might be interesting to see a Gulftown with extra PhysX on running on the CPU with high detail settings on some NV card vs same but PhysX running on GPU. If the game can have extra PhysX enabled and running on the CPU.

If it's GPU limited, then turning on extra PhysX stuff would (should) drop frame rates, but what if you shift that onto a CPU which isn't limiting you... especially one with lots of cores.

Only if you oversimplifly things and disregard the SIMD disadvantages of the CPU vs the GPU :hmm:

 

[BFG10K]

Well maybe with some graphics features enabled CPU core scaling will help.

It won't because enabling those features will load the GPU even more. This is game is GPU limited, plain and simple.

It's likely any decent dual-core CPU can still push a 5870 into oblivion.

 

[Lonyo]

Only if you oversimplifly things and disregard the SIMD disadvantages of the CPU vs the GPU :hmm:

What.
Don't be such a goddamn fanboy every time you post something.


If you have no spare GPU power, how can the spare GPU power do the PhysX? It can't really.
You do though have spare CPU power.

This is a GPU limited game. If it's GPU limited, the PhysX on the GPU would make performance even worse.
Since there is clearly spare CPU power, I was wondering whether a new high end CPU that costs over $1000 (more than a second graphics card which could do the PhysX) would give worse, better, or equal performance to a high end graphics card which is the limiting factor when you turn on the extra PhysX.


Don't take everything to heart and think everyone is trying to claim that PhysX will run as fast on a CPU as a GPU. But in this case, it might, since you don't have much GPU left.

 

[ibex333]

The game runs like crap on my rig.. Even with low setting!!! The lag is unbearable... And to think that I am already past the "MARKET" with this kind of performance.. I'll have to OC more or something... Maybe it's because of Windows 7? Maybe I should run in DX9?

 

[MrK6]

This discussion piqued my curiosity so I bit the bullet and got the game. I tinkered around with the settings and the game runs great on 5850 Crossfire @ 2560x1600 as long as you disable Advanced DOF and Tesselation. I get 45-70FPS so far (only played about 5 minutes), running completely smoothly on the 10.3a Cats. If I turn on Advanced DOF or Tesselation, FPS crashes to about 20-30FPS. If I turn on both, it's unplayable at 1-3FPS. What's even more interesting is if I turn all settings (via the "Quality" option) down to low, but leave the game in DX11 and with Advanced DOF and Tesselation on, the performance doesn't improve. This isn't a frame buffer issue - it's the drivers, the implementation, etc. - something isn't playing nice here with these two features.

What's also interesting is the minimal impact that Tesselation and and Advanced DOF have on image quality. Here's two screenshots from the first minute of the game. I tried to line up the NPC's head to make it easier, but the motion blur kind of screwed up his helmet in one shot. Either way, Tesselation and Advanced DOF are on on the left screen shot, and they're off on the right. Like I said, it's 1-3FPS vs. 55FPS (in this shot) for performance. I also zoomed in on the NPC model and did a side by side in GIMP so you can see the effects of Tesselation on a character model better.



I'm not impressed, considering the performance penalty. Looking at the side-by-side comparison, the arm pack is a little more puffy and the shoulder armor plate is rounder with tesselation on. However, the textures also appear more muddied/blurry. I can't really see any effect of Advanced DOF, but maybe that's more prominent outside. In either case, totally not worth it. I'll turn the two options on here and there when I'm playing to keep checking, but these just look more like unoptimized throw-ins rather than game changing enhancements.

 

[Udgnim]

there's a difference in lighting between the PC and console version: http://translate.google.de/translat...che-Unterschiede-festgestellt/PC/&sl=de&tl=en

no idea if it's specific to a certain DX version though