Post your Metro Exodus benchmark HERE!

[Jig McGalliger]

I am more than willing to advise those who need a better understanding of how the benchmark and 4A engine works as well as how to better run the game on their system. That is what this thread is for. Have fun melting your GPU

Choose your settings and post the result. Post your system specs too. The benchmark executable is available from the game's directory.

System Specs:
CPU: AMD Ryzen 5 1600X 4.0GHz
Cooler: Corsair Hydro Series H110
Motherboard: ASRock Fatal1ty X370 Gaming K4
Memory: G.Skill Trident Z Series 2x8GB DDR4-2933
GPU: ASUS ROG STRIX RTX 2070 O8G-GAMING @2070MHz
Physx: ASUS GTX660Ti-DC20 2GB
Case: Fractal Design R6
Power Supply: Corsair CXM 750W 80+ Bronze

I have motion blur disabled from the config file.

Updated Benchmark #2 Driver 431.36

Ultra + RTX/DLSS 1440p (current in game settings)
Options: Resolution: 2560 x 1440; DirectX: DirectX 12; Quality: Ultra; Texture filtering: AF 16X; Motion Blur: Low; Tesselation: Full; Advanced PhysX: On; Ray Trace: High; DLSS: On; Hairworks: On; Shading Rate: 100;

Total Frames: 4951; Total Time: 104.4621 sec
Average Framerate (99th percentile): 47.88
Max. Framerate (99th percentile): 68.70 (Frame: 41)
Min. Framerate (99th percentile): 28.03 (Frame: 4007)

Ultra + RTX @ 1440 w/ shader at .8
Options: Resolution: 2560 x 1440; DirectX: DirectX 12; Quality: Ultra; Texture filtering: AF 16X; Motion Blur: Low; Tesselation: Full; Advanced PhysX: On; Ray Trace: High; DLSS: Off; Hairworks: On; Shading Rate: 80;

Total Frames: 4762; Total Time: 104.1915 sec
Average Framerate (99th percentile): 46.19
Max. Framerate (99th percentile): 70.20 (Frame: 104)
Min. Framerate (99th percentile): 27.48 (Frame: 3828)

Extreme @ 1440p
Options: Resolution: 2560 x 1440; DirectX: DirectX 11; Quality: Extreme; Texture filtering: AF 16X; Motion Blur: Low; Tesselation: Full; Advanced PhysX: On; Ray Trace: Off; DLSS: Off; Hairworks: On; Shading Rate: 100;

Total Frames: 4005; Total Time: 104.248 sec
Average Framerate (99th percentile): 38.91
Max. Framerate (99th percentile): 57.70 (Frame: 613)
Min. Framerate (99th percentile): 23.50 (Frame: 3197)

^ So no better implementation of DLSS

Updated Benchmark v1.0.0.8 (March 19)

Ultra + RTX/DLSS 1440p (current in game settings)
Options: Resolution: 2560 x 1440; DirectX: DirectX 12; Quality: Ultra; Texture filtering: AF 16X; Motion Blur: Low; Tesselation: Full; Advanced PhysX: On; Ray Trace: High; DLSS: On; Hairworks: On; Shading Rate: 100;

Total Frames: 5053; Total Time: 104.3851 sec
Average Framerate (99th percentile): 48.95
Max. Framerate (99th percentile): 73.44 (Frame: 1902)
Min. Framerate (99th percentile): 25.84 (Frame: 4089)

Ultra + RTX @ 1440 w/ shader at .8
Options: Resolution: 2560 x 1440; DirectX: DirectX 12; Quality: Ultra; Texture filtering: AF 16X; Motion Blur: Low; Tesselation: Full; Advanced PhysX: On; Ray Trace: High; DLSS: Off; Hairworks: On; Shading Rate: 80;

Total Frames: 4891; Total Time: 104.4164 sec
Average Framerate (99th percentile): 47.33
Max. Framerate (99th percentile): 68.19 (Frame: 37)
Min. Framerate (99th percentile): 28.20 (Frame: 3937)

Extreme @ 1440p
Options: Resolution: 2560 x 1440; DirectX: DirectX 11; Quality: Extreme; Texture filtering: AF 16X; Motion Blur: Low; Tesselation: Full; Advanced PhysX: On; Ray Trace: Off; DLSS: Off; Hairworks: On; Shading Rate: 100;

Total Frames: 4015; Total Time: 104.2851 sec
Average Framerate (99th percentile): 39.00
Max. Framerate (99th percentile): 58.33 (Frame: 188)
Min. Framerate (99th percentile): 23.21 (Frame: 3368)

[Jig McGalliger]

Not yet

[Jig McGalliger]

Extreme on v1.0.0.8
Options: Resolution: 2560 x 1440; DirectX: DirectX 11; Quality: Extreme; Texture filtering: AF 16X; Motion Blur: Low; Tesselation: Full; Advanced PhysX: On; Ray Trace: Off; DLSS: Off; Hairworks: On; Shading Rate: 100;

Total Frames: 4015; Total Time: 104.2851 sec
Average Framerate (99th percentile): 39.00
Max. Framerate (99th percentile): 58.33 (Frame: 188)
Min. Framerate (99th percentile): 23.21 (Frame: 3368)

So it no longer looks at max/min the 1% low/high.

[Sparhawk122]

Origineel geplaatst door Jig McGalliger:

Extreme on v1.0.0.8
Options: Resolution: 2560 x 1440; DirectX: DirectX 11; Quality: Extreme; Texture filtering: AF 16X; Motion Blur: Low; Tesselation: Full; Advanced PhysX: On; Ray Trace: Off; DLSS: Off; Hairworks: On; Shading Rate: 100;

Total Frames: 4015; Total Time: 104.2851 sec
Average Framerate (99th percentile): 39.00
Max. Framerate (99th percentile): 58.33 (Frame: 188)
Min. Framerate (99th percentile): 23.21 (Frame: 3368)

So it no longer looks at max/min the 1% low/high.

New updated benchmark = Worse performance 😐🤨

[Dave3d]

Funny, I no longer see the drops to teens (they removed them from the benchmark), AND they are falsifying their numbers (it says tops for me was 117 fps, when I didnt see more than 102 fps).
What a joke.
Whats that old song? How low can you go?

[Jig McGalliger]

Origineel geplaatst door Sparhawk122:

New updated benchmark = Worse performance 😐🤨

Less than 2 fps is well within the margin of error.

Origineel geplaatst door Dave3d:

Funny, I no longer see the drops to teens (they removed them from the benchmark), AND they are falsifying their numbers (it says tops for me was 117 fps, when I didnt see more than 102 fps).

The old version that told you that a single frame was high or low was always a bad indicator of performance; however, it did tell you which frame that was. Now I'm not sure how they are doing the calculation to show a single frame when they are are showing 99th percentile (should be equal to a 1% low/high). The Metro benchmarks have ALWAYS been this way so it should be no surprise.

The benchmark times each frame, so how are you seeing the on ~10ms frame where they actually display that max value of 117? You won't. The human eye will see too many flashes of numbers to be able to discern which of 4000 numbers flashing before you at over 60 times per second is the highest. Like sure, maybe, but don't ever expect this or rely on it.

Also, I always do 2 runs, because the first run always has an outrageuously higher/lower max/min with the second being more rounded out as it doesn't have to manage asset loading. This doesn't happen on the newer version which is fine for me, I still do two runs.

The new one is obviously a better indicator of overall performance as showing the 1 lowest/highest frame is ♥♥♥♥ and no reviewer/benchmarker does it for a specific reason. Anyone who does benchmarking professionally is using FCAT, spreadsheet analysis, and reviewing at most an accuracy of 0.1% lows. 1% is more than accurate for this.

[Sparhawk122]

Origineel geplaatst door Jig McGalliger:

Origineel geplaatst door Sparhawk122:

New updated benchmark = Worse performance 😐🤨

Less than 2 fps is well within the margin of error.

Origineel geplaatst door Dave3d:

Funny, I no longer see the drops to teens (they removed them from the benchmark), AND they are falsifying their numbers (it says tops for me was 117 fps, when I didnt see more than 102 fps).

The old version that told you that a single frame was high or low was always a bad indicator of performance; however, it did tell you which frame that was. Now I'm not sure how they are doing the calculation to show a single frame when they are are showing 99th percentile (should be equal to a 1% low/high).

The benchmark times each frame, so how are you seeing the on ~10ms frame where they actually display that max value of 117? You won't. The human eye will see too many flashes of numbers to be able to discern which of 4000 numbers flashing before you at over 60 times per second is the highest. Like sure, maybe, but don't ever expect this or rely on it.

The new one is obviously a better indicator of overall performance as showing the 1 lowest/highest frame is ♥♥♥♥ and no reviewer/benchmarker does it for a specific reason. Anyone who does benchmarking professionally is using FCAT, spreadsheet analysis, and reviewing at most an accuracy of 0.1% lows. 1% is more than accurate for this.

The patch notes said improved stability and optimisation. I don't see any improvement in the optimisation category.

[Jig McGalliger]

Origineel geplaatst door Sparhawk122:

The patch notes said improved stability and optimisation. I don't see any improvement in the optimisation category.

They state "General Stability fixes" and "Optimization for Lower spec PC's" which means the low/medium settings are easier to run, thus allowing more lower end hardware to run the game at a more constant framerate.

None of this affects my benchmark as I am maxing out my system. General stability also doesn't exactly affect me either as I have no major crashing issues in game.

[Dave3d]

Origineel geplaatst door Jig McGalliger:

Origineel geplaatst door Sparhawk122:

New updated benchmark = Worse performance 😐🤨

Less than 2 fps is well within the margin of error.

Origineel geplaatst door Dave3d:

Funny, I no longer see the drops to teens (they removed them from the benchmark), AND they are falsifying their numbers (it says tops for me was 117 fps, when I didnt see more than 102 fps).

The old version that told you that a single frame was high or low was always a bad indicator of performance; however, it did tell you which frame that was. Now I'm not sure how they are doing the calculation to show a single frame when they are are showing 99th percentile (should be equal to a 1% low/high). The Metro benchmarks have ALWAYS been this way so it should be no surprise.

The benchmark times each frame, so how are you seeing the on ~10ms frame where they actually display that max value of 117? You won't. The human eye will see too many flashes of numbers to be able to discern which of 4000 numbers flashing before you at over 60 times per second is the highest. Like sure, maybe, but don't ever expect this or rely on it.

Also, I always do 2 runs, because the first run always has an outrageuously higher/lower max/min with the second being more rounded out as it doesn't have to manage asset loading. This doesn't happen on the newer version which is fine for me, I still do two runs.

The new one is obviously a better indicator of overall performance as showing the 1 lowest/highest frame is ♥♥♥♥ and no reviewer/benchmarker does it for a specific reason. Anyone who does benchmarking professionally is using FCAT, spreadsheet analysis, and reviewing at most an accuracy of 0.1% lows. 1% is more than accurate for this.

Yeah, but INCLUDING that .1% HIGH in the numbers make the bench look better right?
So, drop the .1%, take the .1% high out of the site so we dont see it BUT include it in the results = higher values/fps.
I didnt see a single scale included over 100/102, but sure enough, there it was in MAX FPS: 117.
Makes it better when the low is 40 instead of 17.5, and the high is 117 instead of 102 right?
And, yes, the .1% IS good to include, because they SHOW where the drop is, and where the stutters are, and actually SHOW stutters.
But, ok, they can fudge. +8 fps to them for the 117, also +11 fps for not showing the 17 fps.
GG

[Jig McGalliger]

I disagree, their 99th percentile is an average of the highest/lowest 1% of frames recorded, thus removing the outlier from our results because we as humans would not perceive it anyways. It's making the data we're seeing from the results more attributable to how we experience the game, just like every professional tester/reviewer out there.

How they're trying to show which single frame (EG "25.84 (Frame: 4089)") is a percentile is beyond me because that is NOT how percentiles work. Multiple values could be in the 99th percentile as there are over 4000 frames drawn for runs that average over 45fps.

I think this could be just an oversight and still I agree the new method is superior at showing how we might experience the game and help mitigate people freaking out over a single outlier like so:

Origineel geplaatst door Sparhawk122:

That's... 🤢🤮

Minimum 7 FPS.

LOLOLOL

Unless they tried to make this specific frame value (EG "(Frame: 4089)" that really has no effect on the percentiles or overall average) is in fact the average of those 1% slowest frames drawn to give you an idea where in the benchmark it happened, but that is a bad indication and looking at the graph would show you where the 1 or 2 out of 4000 frames outlier actually happened. A timestamp would be better because this is forcing you to divide that frame by the overall to see where along the time of the bench this percentile average was located.

Either way, I do 2 or more runs and the average has never had any strong outliers on both versions (notice how my old min/max bench values vary greatly from others' posted here?). The low outlier almost ALWAYS happens when the deer first fully enters the scene and the high is often the first frame of the first run, essentially when nothing is even being rendered.

I'm not going to fight the devs on their decisions for a change of the benchmark (their decisions aren't based on me or what I think as an individual), I'm saying that the changes I saw were made with some reasoning (time will ultimately tell if they are warranted). Evening the bell curve while also making the overall statistics plot and data points revealed is almost always a good thing for getting more accurate statistics. To break this vicious cycle, one must do more than just act without any thought or doubt.

TLDR, don't get hung up on the 1-2 outliers out of 4-6 thousand frames . This is clearly what they intended with the update.

[harmonlpoehl]

DX12 CRASHES, DX11 runs great, if crossfire would work well ,my 290x cfx is Itching for a go at it.
1080Ti
AMD 8370E


METRO EXODUS BENCHMARK RESULTS
2/22/2019 11:47:35 PM

Options: Resolution: 3840 x 2160; DirectX: DirectX 12; Quality: Ultra; Texture filtering: AF 16X; Motion Blur: Normal; Tesselation: Full; Advanced PhysX: Off; Ray Trace: Off; DLSS: Off; Hairworks: Off; Shading Rate: 100;
Run 0



Total Frames: 3924; Total Time: 104.4608 sec
Average Framerate (99th percentile): 38.09
Max. Framerate (99th percentile): 59.56 (Frame: 51)
Min. Framerate (99th percentile): 23.00 (Frame: 701)
View larger version
Average Results



Average Framerate (99th percentile): 38.09
Max. Framerate (99th percentile): 59.56
Min. Framerate (99th percentile): 23.00
https://www.youtube.com/watch?v=WhzMiefL4mk&t=34s

[Mauricius]

Tested with latest drivers and patch:


Options: Resolution: 1920 x 1080; DirectX: DirectX 12; Quality: Ultra; Texture filtering: AF 16X; Motion Blur: Normal; Tesselation: Full; Advanced PhysX: On; Ray Trace: Off; DLSS: Off; Hairworks: On; Shading Rate: 100;
Run 0

Total Frames: 7993; Total Time: 105.022 sec
Average Framerate (99th percentile): 76.75
Max. Framerate (99th percentile): 120.24 (Frame: 1538)
Min. Framerate (99th percentile): 40.53 (Frame: 6525)


Average Results

Average Framerate (99th percentile): 76.75
Max. Framerate (99th percentile): 120.24
Min. Framerate (99th percentile): 40.53

[Moe]

Options: Resolution: 1920 x 1080; DirectX: DirectX 12; Quality: Ultra; Texture filtering: AF 16X; Motion Blur: Normal; Tesselation: Full; Advanced PhysX: On; Ray Trace: Off; DLSS: Off; Hairworks: On; Shading Rate: 100;
Run 0

Total Frames: 4528; Total Time: 104.3018 sec
Average Framerate (99th percentile): 43.91
Max. Framerate (99th percentile): 69.69 (Frame: 725)
Min. Framerate (99th percentile): 24.52 (Frame: 3685)



Options: Resolution: 1920 x 1080; DirectX: DirectX 12; Quality: High; Texture filtering: AF 16X; Motion Blur: Low; Tesselation: Full; Advanced PhysX: On; Ray Trace: Off; DLSS: Off; Hairworks: On; Shading Rate: 100;
Run 0

Total Frames: 5366; Total Time: 104.4448 sec
Average Framerate (99th percentile): 51.90
Max. Framerate (99th percentile): 82.32 (Frame: 767)
Min. Framerate (99th percentile): 27.81 (Frame: 4384)


Ryzen R5 2600x (stock clocks)
16gb DDR4 3200mhz (14-14-14-34)
GTX 1070 (stock clocks)


I need a new GPU

[Nite69]

Origineel geplaatst door Sparhawk122:

Origineel geplaatst door SyberWolf™:

Options: Resolution: 2560 x 1440; DirectX: DirectX 12; Quality: Extreme; Texture filtering: AF 16X; Motion Blur: Normal; Tesselation: Full; Advanced PhysX: On; Ray Trace: Ultra; DLSS: On; Hairworks: On; Shading Rate: 100;
Run 0

Total Frames: 6005, Total Time: 104.2163 sec
Average Framerate: 58.10
Max. Framerate: 123.44 (Frame: 4115)
Min. Framerate: 7.79 (Frame: 4750)


Average Results

Average Framerate: 58.00
Max. Framerate: 123.44
Min. Framerate: 7.79


9900K
2080Ti
16GB

That's... 🤢🤮

Minimum 7 FPS.

LOLOLOL

its because the benchmark will stutter during the first run, its why you need to run at least 2 runs or more of the benchmark to get a better minimum

in metro 2033 benchmark they had it set to 3 runs by default but for some reason they set it to 1 run on exodus by default

[Jig McGalliger]

Origineel geplaatst door Nite:

in metro 2033 benchmark they had it set to 3 runs by default but for some reason they set it to 1 run on exodus by default

More runs certainly did a better job of hindering outliers. 99th percentile should also do the same though maybe a little too brazen for my liking. I'd prefer .1% accuracy.

I'm fine with it's current built as recording the single min/max frame time is obviously a horrible metric.

[macatak]

Origineel geplaatst door Nite:

Origineel geplaatst door Sparhawk122:

That's... 🤢🤮

Minimum 7 FPS.

LOLOLOL

its because the benchmark will stutter during the first run, its why you need to run at least 2 runs or more of the benchmark to get a better minimum

Having hairworks enabled causes the stutter/hitch on the 1st run (close to where the deer appears), turn hairworks off and this won't happen on the 1st run.