Benchmarking framerates between 4.0 and 4.1?

[Rake]

15 hours ago, Ssnake said:

Try to open the scenario in the Mission Editor. When the warning message comes up that you're missing the map for it, click the button to "Replace map" and then pick from the stock "Hannover-Weserbergland". Steel Beasts will then proceed to load it, then you can save it, and run it.

Okay...  That worked and I was able to run the scenario.  Per the ratings, I fit into the low/medium category, but only because one time did the minimum frame rate drop into the twenties; otherwise, it's "Good" to borderline "Great":

 

 

What really surprised me was the consistency of most of the frames (excepting the smoke) with an average right at 50 FPS.  Overall, it runs well enough that I will run well over the default graphics settings.  I'll post anything that I find tipping the scales

 

My Specs:

Alienware Aurora

Win 10 Pro

i7-9700K @3.6GHZ

32 GB RAM

NVIDIA GeForce GTX 1080 Ti

Samsung SSD 860EVO - 2 TB

PC400 NVMe SK hynix - 512 GB

 

[Ssnake]

With a machine like this, SB Pro PE ought to run at least "good" if not "great".

[Scrapper_511]

Well, I just ran this older 4.0 benchmark on version 4.157 and my average frame rate went from 25 to 32! 👍👍

Edited August 5, 2019 by Scrapper_511

[inexus]

After having reading people's results and I've tested a large number of different scenarios it seems unlikely that a descent framerate can be achieved even on the fastest overclocked CPUs and GPUs today unless the missions are fairly small in terms of landscape complexity and units. I've given up the idea of upgrading to a 9900K@5Ghz as I only think it would provide a few more fps.

It's really hard to poinpoint what is causing the slowdown. I had a scenario where changing ground cover from 44 to 45 would make the fps drop 10 frames from 40 to 30. If i reverted the value to 44 it would go back up to 40... 

[Hoover]

2 hours ago, inexus said:

I've given up the idea of upgrading to a 9900K@5Ghz as I only think it would provide a few more fps.

It's really hard to poinpoint what is causing the slowdown. I had a scenario where changing ground cover from 44 to 45 would make the fps drop 10 frames from 40 to 30. If i reverted the value to 44 it would go back up to 40... 

I confirm what you wrote . I use 4K and most of the time the sim runs great at 30-50 fps. As you, I have sudden drops in fps to a one-digit. Sometimes it‘s just traversing the turret left or right or looking through the optics and 💥 from 35 to 7 fps. I played with the idea to buy a quicker cpu, but as you I came to the conclusion that this would improve the framerate just a bit and not solve the problem. Conclusion: The only options a user with 4K has is...

•  to reduce the resolution (what in my case will never happen)
• wait, for a solution from esim

Specs:

4K (3840x2160) resolution in a medium size scenario I get about a nice 40-60 fps

• System: Win 10 Pro / 64bit,

  • Ram: 32 GB

  • CPU i7-5960X 3 GHz

  • GPU NVIDIA GTX 1080 Ti

[inexus]

20 hours ago, Hoover said:

I confirm what you wrote . I use 4K and most of the time the sim runs great at 30-50 fps. As you, I have sudden drops in fps to a one-digit. Sometimes it‘s just traversing the turret left or right or looking through the optics and 💥 from 35 to 7 fps. I played with the idea to buy a quicker cpu, but as you I came to the conclusion that this would improve the framerate just a bit and not solve the problem. Conclusion: The only options a user with 4K has is...

•  to reduce the resolution (what in my case will never happen)
• wait, for a solution from esim

Specs:

4K (3840x2160) resolution in a medium size scenario I get about a nice 40-60 fps

• System: Win 10 Pro / 64bit,

  • Ram: 32 GB

  • CPU i7-5960X 3 GHz

  • GPU NVIDIA GTX 1080 Ti

Changing the resolution has almost no impact (at least if your gpu is not fully utilised at 4K which is my case). I provided screenshots earlier to demonstrate the difference in resolutions. 

 

In the weekend I tested the same scenario with differences in cpu speed. 4.4, 4.0 and 3.5ghz. I measured only a minor difference in FPS. Based on extrapolation of this data I believe that a scenario with say 30 FPS today on a 4.4ghz cpu may be 4-5 frames faster running at 5ghz. 

 

For me ‘playable’ in a pro means at least 40+ FPS. In any other sim/game I go for 60 FPS. 

 

An interesting thing is that going to the map view often brings back a good improvement of FPS so it feels like the cpu is still doing a fair amount of the graphics. 

Edited August 7, 2019 by inexus

[Badcat]

I have been running 4.157, and what has been the slow down as far as my machine goes was the Pagefile and physical RAM. I haven't bothered with getting another 8 GB of RAM but I did increase manually the Swap file in windows which gave me a few extra FPS.

This machine is pretty unique, I am running a pentium 3258 OCed to 4.2 ghz.  with 8 gigs of RAM and a Radeon 550 video card. With medium settings I am getting between 25-30 FPS. With the Benchmark scenario, 20-40.  I am assuming this game, 4.1 is using multiple cores whereas the 4.0 appeared to be using only 2 cores. 

[thewood]

So decided to run some tests to see which AA settings gave better results in quality and performance.  I was quite surprised when changing nVidia settings and SB settings.  I am pasting the spreadsheet here.  Might be hard to follow, but I'll summarize the results below.  My machine is a lenovo laptop i7-9750, gtx1660 ti, 16 Gb RAM, and an SSD.  There were no thermal issues, but RAM usage was 12 Gb.

 

Scenario 1			Scenario 2			Scenario 3	Best quality
	nVidia	SB		nVidia	SB		nVidia	SB
FXAA	Off	NA		Off	NA		Off	NA
AA Mode	Application-controlled	NA		Application-controlled	NA		nVidia -controlled	NA
AA Setting	Application-controlled	4		Application-controlled	8		4	4
Advanced AA	NA	0		NA	0		NA	0
		FPS (avg)			FPS (avg)			FPS (avg)
Scene 1 M60 observer		63			63			62
Scene 2 M60 Gunsight trees		55			45			48
Scene 3 City Smoke		45			43			45
Scene 1 - Leo looking back		57			60			56
Scenario 4			Scenario 5	Worst Quality		Scenario 6
	nVidia	SB		nVidia	SB		nVidia	SB
FXAA	Off	NA		Off	NA		On	NA
AA Mode	nVidia -controlled	NA		nVidia -controlled	NA		nVidia -controlled	NA
AA Setting	4	8		4	0		4	0
Advanced AA	NA	0		NA	0		NA	100
		FPS (avg)			FPS (avg)			FPS (avg)
Scene 1 M60 observer		62			62			63
Scene 2 M60 Gunsight trees		60			62			45
Scene 3 City Smoke		42			50			43
Scene 1 - Leo looking back		57			60			60

 

What I found is that there is some kind of interaction between the nVidia settings and SB settings.  The best settings for quality settings appear to be nVidia-controlled AA set at 4 AND SB AA set to 4.  Anything above 4 in SB appears to actually be worse quality with jagged lines on some straight shapes.  Its kind of weird.  There is just some strange interaction between the nVidia and SB settings.  The setting of nVidia at 4 and SB at 8 gave good performance, but its AA quality was not very good.  It almost looked like AA was off or very low.

Edited August 7, 2019 by thewood

[Raven434th]

Latest patch has same fps scanning tree line as previous one. 9 fps. Was there any further optimization done on this recent 159 patch?? If so, what?    

[Ssnake]

Between 4.157 and 4.159 no changes in the render code were made unless explicitly listed in the Release Notes.

[Hoover]

On 8/6/2019 at 4:54 AM, inexus said:

After having reading people's results and I've tested a large number of different scenarios it seems unlikely that a descent framerate can be achieved even on the fastest overclocked CPUs and GPUs today unless the missions are fairly small in terms of landscape complexity and units. I've given up the idea of upgrading to a 9900K@5Ghz as I only think it would provide a few more fps.

It's really hard to poinpoint what is causing the slowdown. I had a scenario where changing ground cover from 44 to 45 would make the fps drop 10 frames from 40 to 30. If i reverted the value to 44 it would go back up to 40... 

Hi inexus,

did a little bit benchmarking today. Here are the results.

Cheers Hoover

 

[REPORT]
In Benchmark 1 I had the GPU maxed out between 82% and 99%. As I could not believe it, I repeated the test 3-times. It always came down to the same result. If you look at the FPS only, all seems to be OK. The GPU usage tells another story. Before drawing a conclusion, others should confirm that result. Very interessting would be a benchmark result with a quicker GPU like the GeForce RTX 2080 Ti.

[TEST]
 I used MSI Afterburner for testing. Load Afterburner after you entered the station you like to test. Avoid switching to the map. Always take you position first then start Afterburner.

[REMARKS]
From minute 02:00 the weather gets worse and the visibility gets lower, felt proportinal goes the GPU load down from 99% to about 57%.

 

[UPDATE]

In my results the GPU was maxedout to 99%. But this finding seems to be an edge case and happens only when you look in direction of the wood in daysight and in a short distance to the wood. I could reproduce this behaviour (GPU maxedout 99% ) in another scenario. But the FPS stayed in these cases within the (for me) playable (> 30 FPS) range. When I pan away from the wood the GPU loadout drops dramatically.

 

[temp. Conclusion]

In my case with my hardware the GPU seems to be NOT the limiting factor.

 

 

Edited August 18, 2019 by Hoover
Update:

[Ssnake]

How do you determine the "min" framerate?

From the statistics that the screenshot/Alt+F12 frame counter shows?

My way of determining the min rate is from the actual screenshot values, not from those statistics.

[Hoover]

12 minutes ago, Ssnake said:

How do you determine the "min" framerate?

From the statistics that the screenshot/Alt+F12 frame counter shows?

My way of determining the min rate is from the actual screenshot values, not from those statistics.

I take the Afterburner values.

[ssidiver]

On 8/3/2019 at 5:11 PM, Ssnake said:

We are using parallelization in some cases, but not generally/all the time.

Given some comments that have been made about multi core use, I have been watching their use.

 

Loading maps definitely utilizes all cores; while playing a scenario also increases the use of all cores but leans predominately on two. The two process using most of the CPU capacity are SBProPEcm.exe (111 Threads) and DecodeProcess.exe (28 Threads).

 

Seems to me reasonable amount of multi core use.

[Funduro]

All in all very happy with the latest update. My PC is somewhat limited with regards performance but I am getting 30-40 fps which is playable. The only real problem area seems to be zoom views using the binoculars or gun sight where I am getting single digit frames especially in wooded areas. Just wondering if there are any plans to have a look at this in a future patch? Could it be related to the tree models? too many polygons? Is it LOD related perhaps? Not an expert but it would be nice to see a fix for this.

[Ssnake]

I can but recommend that you work with the terrain detail sliders, as they also control the distance at which 3D trees get rendered as simple billboards. The deep forest is a test of the "max overdraw" situation where a gazillion of triangles are all overlapping each other, so reducing the number of triangles by way of LOD balance in the detail sliders should help. At the end of the day however the ability of your graphics card to process 3D geometry data is the bottleneck. I suspect that your graphics card is either old, or low end, or both. We can squeeze what's possible from it, but where there's a hardware limit, only better hardware can solve it. For example, my GTX 980 has no problems whatsoever maintaining a framerate above 60 with the deep forest scene.

[Funduro]

Thanks for that. I have an AMD VaporX 7950 graphics card with 3GB memory which I admit is getting a bit old now. I will have a look at those detail sliders.

[Raven434th]

1 hour ago, Ssnake said:

I can but recommend that you work with the terrain detail sliders, as they also control the distance at which 3D trees get rendered as simple billboards. The deep forest is a test of the "max overdraw" situation where a gazillion of triangles are all overlapping each other, so reducing the number of triangles by way of LOD balance in the detail sliders should help. At the end of the day however the ability of your graphics card to process 3D geometry data is the bottleneck. I suspect that your graphics card is either old, or low end, or both. We can squeeze what's possible from it, but where there's a hardware limit, only better hardware can solve it. For example, my GTX 980 has no problems whatsoever maintaining a framerate above 60 with the deep forest scene.

If its geometry,why such a big difference between thermal and day sights? In my case 9 in day and 20-21 in thermal? That's scanning a treeline.

Edited August 29, 2019 by Raven434th

[Ssnake]

The thermal image is rendered at a much lower resolution (that of the thermal sensor).

[Rotareneg]

That's not all, the free camera's thermal view is full resolution and gets better FPS too, I think it's mainly that the thermal view doesn't use a lot of shaders, so while it is still a lot of overdraw on the trees for example, they're just a plain texture without bump and specular maps (I have to wonder why we need bump and specular mapped leaves, grass, etc., but I suppose that's another discussion.)