Dumbfounded. Metro Last Light cpu and gpu usage - makes no sense!

[BallaTheFeared]

Ok so first screen, cpu is at 4.6GHz, gpus are at 960/1250..

44% cpu usage, 73% gpu usage... Ok makes sense, clearly I'm cpu limited.


Let's just make sure, overclock the cards a bit... 4.6GHz cpu, 1150/1650

49.6 FPS

CPU usage goes up a tad, gpu usage goes down 1%, scaling isn't great, fps go up...


What's the problem/bottleneck?

Running off a SSD, ram is at 2400MHz 11-13-12 1T


Edit: Built in benchmark..

3.8GHz cpu, 1200/1675 GPUs

4.6GHz cpu, 1200/1675 GPUs

 

[frozentundra123456]

Ok so first screen, cpu is at 4.6GHz, gpus are at 960/1250..

44% cpu usage, 73% gpu usage... Ok makes sense, clearly I'm cpu limited.


Let's just make sure, overclock the cards a bit... 4.6GHz cpu, 1150/1650

49.6 FPS

CPU usage goes up a tad, gpu usage goes down 1%, scaling isn't great, fps go up...


What's the problem/bottleneck?

Running off a SSD, ram is at 2400MHz 11-13-12 1T


Edit: Built in benchmark..

3.8GHz cpu, 1200/1675 GPUs

4.6GHz cpu, 1200/1675 GPUs

In the line right under the top picture, didnt you mean to say "gpu limited"? Also, I cant find where you say what the framerate was for those conditions.

Anyway, cant explain your results. 4.6 on Haswell is pretty remarkable though.

 

[Carfax83]

You're not bottlenecked. It's just that that particular area of the game isn't very GPU intensive. Metro Last Light isn't very system intensive at all really (unless you have PhysX and or SSAA turned on), generally speaking. Not only is it a lot more optimized than it's predecessor, but the engine is highly optimized for consoles in terms of resource usage..

It uses comparatively much lower amounts of VRAM and system memory than say Crysis 3, and even Bioshock Infinite.

I guess they took the criticism of Metro 2033's lack of optimization to heart

 

[frozentundra123456]

You're not bottlenecked. It's just that that particular area of the game isn't very GPU intensive. Metro Last Light isn't very system intensive at all really (unless you have PhysX and or SSAA turned on), generally speaking. Not only is it a lot more optimized than it's predecessor, but the engine is highly optimized for consoles in terms of resource usage..

It uses comparatively much lower amounts of VRAM and system memory than say Crysis 3, and even Bioshock Infinite.

I guess they took the criticism of Metro 2033's lack of optimization to heart

I think his point was that both cpu and gpu usage are well below 100 precent, so why is the framerate not higher. By definition, shouldn't one of them be a bottleneck, unless something else in the system is limiting the framerate.

 

[BallaTheFeared]

I think his point was that both cpu and gpu usage are well below 100 precent, so why is the framerate not higher. By definition, shouldn't one of them be a bottleneck, unless something else in the system is limiting the framerate.

Yes, except I can understand if the cpu isn't fully taxed it won't be at 100% say if it's dual threaded...

But if that was the case, than increasing my gpu clocks wouldn't increase my fps, since I was already limited by the cpu, which is not the case - fps do go up with increased gpu clocks.

Just confusing

Here is 4.8GHz

 

[Carfax83]

I think his point was that both cpu and gpu usage are well below 100 precent, so why is the framerate not higher. By definition, shouldn't one of them be a bottleneck, unless something else in the system is limiting the framerate.

Maybe he has V-sync enabled? Or perhaps the Crossfire profile for this game isn't very efficient?

 

[RussianSensation]

Ok so first screen, cpu is at 4.6GHz, gpus are at 960/1250..

CPU usage goes up a tad, gpu usage goes down 1%, scaling isn't great, fps go up...

What's the problem/bottleneck?

Your GPU is overclocked 19.8%, your memory is overclocked 32%. Your overall FPS goes up 17.5%. You are GPU limited. GPU scaling is not going to be 100% due to CF scaling and the fact that GPUs do not scale 100% with overclocking after a certain point. I would say 17.5% higher FPS with 19.8% overclock on the GPU in CF mode is very good.

I do not understand how you arrived that you are CPU limited. If you were CPU limited, at 4.6ghz, your performance wouldn't go up once you overclocked the GPUs from 960 to 1150. The reason your CPU usage is so low is because the game isn't using many threads in that scene. Actually your 2nd set of benchmarks supports this even more since there is barely a 2 fps difference despite a CPU underclock to 3.8ghz. You are mostly GPU limited in this title - specifically shaders, ROPs, TMUs, geometry engines, rather than memory bandwidth.

 

[BallaTheFeared]

Yes I'm clearly gpu limited, since overclocking the gpus yields a large gain, whereas overclocking the cpu from 3.8 to 4.8 does very little.

What's confusing is why my gpu usage is so low, 70% on both cards. That typically indicates a cpu bottleneck, which is clearly not the case... Hence the confusion, why is the gpu usage so low, yet limiting?

Thanks for replying RS!

 

[RussianSensation]

What's confusing is why my gpu usage is so low, 70% on both cards. That typically indicates a cpu bottleneck, which is clearly not the case... Hence the confusion, why is the gpu usage so low, yet limiting?

Test the same scene with 1 GPU. Let's see what the GPU usage is. My guess is you are getting a CF driver bottleneck. I would expect single GPU usage to be much higher than 72-73% if this theory holds true.

 

[BallaTheFeared]

That area was from the near start, can I go back on my save or will I lose my progress?

Will the benchmark suffice? It will have to wait about an hour though, baby's nap isn't until after 12 >.<

 

[Face2Face]

Not to get too off subject, but what are you seeing with your GPU usage with CFX in Crysis 3?

 

[aigomorla]

Metro had a bug with logitech Game Pannel software if i recall.

The MOST annoying bug, had to do with the applet on the logitech keyboard screen.

If your using the said game software from logitech disable it, then open metro, then reopen the logitech gaming software.

It literally doubled my FPS.

 

[BallaTheFeared]

Not to get too off subject, but what are you seeing with your GPU usage with CFX in Crysis 3?

99/99 with everything at very high MGPU SMAA when not cpu limited.

http://www.youtube.com/watch?v=1n30fFaRb3o

4.6GHz single @ 1200/1675

54 -> 76

40% CFX scaling?

Metro had a bug with logitech Game Pannel software if i recall.

The MOST annoying bug, had to do with the applet on the logitech keyboard screen.

If your using the said game software from logitech disable it, then open metro, then reopen the logitech gaming software.

It literally doubled my FPS.

I use a logitech mouse, but LCore isn't running I haven't installed the software for it.

Think I should install it, then disable it?

 

[zlatan]

The problem is not your hardware. This game is mostly use precomputed effects, and these are not useful for today's high ALU:tex ratio architectures. The ALUs are simply idling.

 

[Carfax83]

The problem is not your hardware. This game is mostly use precomputed effects, and these are not useful for today's high ALU:tex ratio architectures. The ALUs are simply idling.

That's what I said earlier in a different way. Metro Last Light is nothing like the first game, which was a real GPU buster.

Even on my GTX 580 SLI, I was rarely hitting 99% usage. The game just doesn't require much GPU power unless you're running PhysX and or SSAA..

 

[Carfax83]

Metro had a bug with logitech Game Pannel software if i recall.

The MOST annoying bug, had to do with the applet on the logitech keyboard screen.

If your using the said game software from logitech disable it, then open metro, then reopen the logitech gaming software.

It literally doubled my FPS.

They released a patch that fixed that problem.

 

[BallaTheFeared]

SSAA Single

SSAA CFX

60% CFX scaling?

 

[Skurge]

I have the same 75% GPU usage in that game. It's rather annoying. Like I'm leaving performance on the table.

 

[Carfax83]

I have the same 75% GPU usage in that game. It's rather annoying. Like I'm leaving performance on the table.

Is gameplay smooth though or jerky? If it's smooth I wouldn't worry about it.

Like I said, MLL is extremely optimized for consoles so it doesn't use the resources of high end computers effectively. As long as you're still getting playable or smooth frame rates, I wouldn't worry about it.

Of course, it could just as well be that AMD's Crossfire profile for the game isn't very efficient, or perhaps a combination of the two.

 

[escrow4]

From memory, my Titan (1100MHz/6008MHz) at 1920x1200 with PhysX turned up and every setting maxed minus SSAA typically hit 80% at the outdoor scenes, and more than likely hit 90% in some indoor scenes. FPS hovered around 50-60FPS (adaptive Vsync enabled). In the benchmark my min framerate was 20, and max was 55 or so.

 

[sxr7171]

From memory, my Titan (1100MHz/6008MHz) at 1920x1200 with PhysX turned up and every setting maxed minus SSAA typically hit 80% at the outdoor scenes, and more than likely hit 90% in some indoor scenes. FPS hovered around 50-60FPS (adaptive Vsync enabled). In the benchmark my min framerate was 20, and max was 55 or so.

Impressive O/C on your Titan. But sad to know that even it cannot run MLL at 60fps @ 1080p.

 

[tweakboy]

Gosh CPU is way enough. Video card is the bottleneck here.

 

[BallaTheFeared]

Windows 8 13.6b2 - 75.89 FPS

Windows 8.1 13.15.1 - 79.81 FPS

All other settings the same. Actual GPU usage doesn't seem to have changed however.

CPU Core 4.6GHz 1.15v, Uncore 4.5GHz 1.14v
8GB DDR3 2400MHz 11-13-12 1T
7950CFX 1200/1675

Still no closer to understanding

 

[notty22]

Windows 8 13.6b2 - 75.89 FPS

Windows 8.1 13.15.1 - 79.81 FPS

All other settings the same. Actual GPU usage doesn't seem to have changed however.

CPU Core 4.6GHz 1.15v, Uncore 4.5GHz 1.14v
8GB DDR3 2400MHz 11-13-12 1T
7950CFX 1200/1675

Still no closer to understanding

I noted behavior when running sli that might show something with your situation. Though I'm not sure what it will show (happening)
Try running both cards at 700mhz, crossfire, a slower mhz see if that then gets you 100% gpu usage. Some combination of settings might get you the same fps at closer to 100% gpu usage.
Just for kicks?

 

[BallaTheFeared]

700/1250 in CF, single card shows 99% even at 1200.

Also single gets 54 fps @ 1200/1675, faster than this run, yet still usage is ugly >.<


Edit: Ops, my CPU is actually at 4.8GHz right now, I thought I switched back before installing 8.1 preview...