Horizon Draw Distance performance hit (40Km, 100Km, 150Km)

[chiliwili69]

I have just run three benchmark test (with the new samuel track for 3.001) with three different values of the new parameter Horizon Draw Distance (HorDD):

 

HorDD: 40Km , Frames: 13358 - Time: 200000ms - Avg: 66.790 - Min: 40 - Max: 91

HorDD: 100Km, Frames: 13152 - Time: 200000ms - Avg: 65.760 - Min: 39 - Max: 91

HorDD: 150Km, Frames: 12753 - Time: 200000ms - Avg: 63.765 - Min: 37 - Max: 91

 

So on the avg it is just 1 or 2 fps difference, but if we plot the fps of each second of the track (200 seconds) we see the following:

 

 

The track stays most of the time at either 90 or 45 (ASW was OFF), so the final average numbers doesn´t reflect the true influence of the HorDD parameter.

To really know that, we should look at the times when the fps are just between 45 and 90, this is marked in a red elipse. There we can see that the difference is about 10 fps from 40K to 150Km.

If we calculate the difference between HorDD at 40Kw and 150Km the values are:

 

 

 

[56RAF_Roblex]

I think you also need to add some small amount of averaging (just a few ms) to that final plot because the real maximum difference in fps really is nearer 10 but a momentary difference in where the maximum load occurs is giving a false reading of 20.

[chiliwili69]

8 hours ago, 56RAF_Roblex said:

giving a false reading of 20

yes, you are right. There are some false 20 due to small differences in the sampling.

The real difference is around 10 with few peaks to 15.

[Milopapa]

FPS is useless in VR. You need to plot frametimes.

[chiliwili69]

4 hours ago, Milopapa said:

FPS is useless in VR. You need to plot frametimes.

 

Why fps is useless in VR?

 

FPS and frametimes is equivalent.

 

[A_radek]

Thank you chili. once again your systematic testing is a solid help optimizing for 90fps.

[Milopapa]

On 3/24/2018 at 11:57 PM, chiliwili69 said:

 

Why fps is useless in VR?

Because FPS is an average value by definition (Frames Per Second) whereas frametimes are shown for each and every frame. When you have a dropped frame, fps will show 89 or 90, while frametimes will go something like 8-8-9-16-8. So better for detecting framedrops.

[TUS_Samuel]

FPS is not completely useless but I agree that frametimes would give more information about performance.

I have a hypothesis at least about HUD. Even if FPS is limited to 45 then you will get enough framedrops even if avg FPS is around 45. So HUD is more about stutters then FPS. Though I have not measured this yet.

[=EXPEND=Tripwire]

I agree that frame times should be what is used to demonstrate performance.

I think it would go a long way to show possibly what is causing the stutter we see when it appears that FPS is pegged at 90fps.

[chiliwili69]

On 4/4/2018 at 10:59 AM, Milopapa said:

Because FPS is an average value by definition (Frames Per Second) whereas frametimes are shown for each and every frame. When you have a dropped frame, fps will show 89 or 90, while frametimes will go something like 8-8-9-16-8. So better for detecting framedrops

 

But the purpose of those test is not just to detect a single spare dropped frame, (which will cause a 89) but to measure continuous dropped frames within a second.

 

If you drop 13 frames in a second you achieve 67 fps in that second

if you drop 23 frames in a second you achieve 57 fps in that second, and so on...

 

So fps is fully equivalent to the dropped frames. Basically we used fps to know how far we are from 90.

 

But fps can also tell you when you have a single spare dropped frame, look at the red line (with 150Km) in the second 180, you will see that the fps are 90-90-90-90-89-90-90-90...

So, you also detect single spare dropped frames.

On 4/4/2018 at 3:21 PM, =EXPEND=Tripwire said:

I think it would go a long way to show possibly what is causing the stutter we see when it appears that FPS is pegged at 90fps

 

You can configure Fraps to report Frametimes as well.

Try to run Fraps when you say you have solid 90fps with stutters. You will see as well the Frametimes are logically 10-10-10-11-10-9-10-9-10-9-11-11...  all equal or below 11ms