Oculus Performance HUD

[Roger Wilco Jr]

A search didn't display this topic, so I thought I'd mention it.
I'm sure many of you Oculus veterans already know about it.

Anyway, after spending 3 days using the default settings for the CV1 in ED, I decided to try the SDK debug tool to set a higher pixel density (apparently using the debug tool is more efficient than using in-game super sampling). ED has been looking better each day, as I get more used to both the hardware and the view, but increasing the pixel density really makes a big difference. Things were looking great until I was driving around in an SRV in a canyon. During tight turns the canyon walls would blur and I'd start feeling a little funny. I tried using the in-game FPS counter, but it didn't work, and neither did the Nvidia Experience software. While looking for a way to measure the FPS, I found the Performance HUD in the Oculus SDK Debug Tool. If you enable the Performance HUD, and choose to display the Performance Summary, it will overlay a couple graphs showing the frame rate and the overhead.

It was pretty eye opening playing with various settings. With my GTX 1080 and 2.0 PD, I could maintain 90 FPS in space. It would drop in stations, but it was acceptable. But on planets in an SRV, it would drop way too low to be comfortable. With a 1.5 PD and Ultra settings, I could maintain 90 FPS just about anywhere around a station or planet, with just occasional drops. But then I headed to a RES. With a bunch of ships, missiles, chaff, and etc., even on VR low, the FPS can drop into the 50's - maybe even lower. It wasn't really much worse using VR High or Ultra settings.

Anyway, the FPS drop in the RES really doesn't bother me very much. As long as I can maintain a decent FPS in an SRV, I'll live with it. So now I'll be spending days trying to figure the maximum PD I can use, and what graphics settings I can lower.

[TorTorden]

A lot of people also downsample in game as well.
Inow they set a multiplier of less than 1 in the game menu and upsample with dk tool.

I haven't tried with the rift yet but the vive actually looks and perform better using ingame sampling 0.65 and a steamvr upsampling setting of 2x.

In short the ED rendering engine honestly that great.

The difference between low and high vr is fairly negligible (high vr is actually mostly medium settings) and you can easily turn off any anti aliasing (at least when upsampling out of game)
And with a card like the 1080 you can probably get away with increasing a lot of to near Ultra levels.

As for the fps counter the one in game works. But shows only on the monitor window, apparently the number is grabbed early in the rendering pipeline so is actually good for what is being sent to the hmd.

[Roger Wilco Jr]

A lot of people also downsample in game as well.

Although some say it's a terrible idea, I should at least try it to see the affect for myself.

As for the fps counter the one in game works.

Cool - so that's what I see people peeking at. Of course, I'd need to put on my glasses to read it.
I think I'd prefer the HUD for testing graphics settings changes. Too bad they don't just display the FPS in VR.

[thebs]

Delete

[thebs]

And with a card like the 1080 you can probably get away with increasing a lot of to near Ultra levels.

I'm actually running a GTX 980 Ti in the Rift and getting 75fps+ constantly, with 90fps typical, everything Ultra, by turning down Shadows to Medium. Turning down Shadows really makes a heavy, post-processing performance difference.

If anyone changes the sampling with the Rift, please post your results. It's something I'm willing to try as well.

[Roger Wilco Jr]

I must admit, I'm still more than a little confused about exactly what is happening. You've got the Oculus Debug Tool Pixels Per Display Pixel Override, the in-game Supersampling, and also in-game anti-aliasing. Apparently some of this is pre-processing and some is post-processing - I assume. And it doesn't help that some of the settings don't do anything, like I think most of the anti-aliasing options (or is it all of them). And worst of all, you can't get an actual Oculus screenshot for side by side comparisons - you have to rely on memory!

So I think we are trying to achieve two goals: decrease jaggies while maintaining a decent FPS.

I've heard, and read here, that the supersampling in Elite is not very good. What is meant by that? Is it because it does a lousy job? I didn't notice that on my monitor. Is it because it requires too much processor power? If so, why? Does it actually do a better job of getting rid of jaggies because it uses a more complex pattern and/or algorithm? If so, I'd think "one pass" with the in-game supersampling would do a better job than "two passes" using the debug tool and in-game downsampling.

Is the Oculus debug tool PPDPO "better" because it uses less performance? Does it actually do a worse job of getting rid of jaggies because it only uses a simple pattern/algorithm?

I assume that if in-game supersampling is set to default 1.0, it doesn't do anything extra - no extra passes? But if you either upsample or downsample, it performs at least one extra pass?

And if you use the debug tool and in-game up/down supersampling, doesn't it add even more passes? I'd almost think that would require 4 passes in total. You might get a better image, but wouldn't that kill the FPS?

And how does the in-game anti-aliasing setting affect this whole process? Isn't it even just more processing power that should probably be applies to using a higher supersampling setting? Shouldn't it just be turned off?

I swear, a little knowledge (or basically none at all) is a dangerous thing.

I guess all I can do is continue experimenting. Hopefully, somebody with a lot of technical knowledge can separate some of the facts from the fiction. Technically, the FPS results should be facts. My problem will be the subjective review of image quality based on my memory.

[Roger Wilco Jr]

I can verify anti-aliasing does have some effect - actually, quite good.

Both FXAA and SMAA smooth jaggies on shallow angles, but don't do much at 45 degrees compared to supersamling. FXAA is about a 5% performance hit, and SMAA is about a 7% performance hit, with 1.5x supersampling about a 46% performance hit and 2.0x supersampling about a 68% performance hit. I think FXAA and SMAA are done after the frame is rendered and only work on the edges. Oddly, supersampling helped reduce jaggies on 45 degree angles, but not so much on shallower angles and circles. You really seem to need both enabled if you want to minimize jaggies all around. I couldn't tell much of a difference between FXAA and SMAA, so I'd recommend testing with both and see what looks best. I've read that FXAA can blur more than SMAA, but I didn't notice that in my very limited testing.

I read the MLAA anti-aliasing is ATI specific. I did test them anyway, and they did not perform as well and had higher performance hits, so I'd definitely recommend against them for Nvidia cards.

Sadly, taking high res screen shots didn't help, as the images do not look like what I was looking at on screen, and it takes like 10 clicks and 10 seconds to go through the setting changes, so I couldn't really do side-by-side comparisons.

Btw, I performed these tests on my monitor with ultra settings and 2560x1440 resolution using an Nvidai GTX 1080.

Now I'm going to perform the same tests in the HMD and see if I can see the same differences, and then I'll bring the Oculus debug tool into the mix and see what happens.

[Roger Wilco Jr]

OK, here is some more info that is contrary to what I've been reading on the main forums. As far as I can tell, and at least the visuals are subjective, the in-game supersampling and the debug tool have the same effect on FPS, and I didn't notice a difference in quality.

So if I have in-game supersampling set to 1.0x, 1.5x, or 2.0x, and the debug tool is either off or set to 1.0x, I get the same FPS as when I have the debug tool set to 1.0x, 1.5x, or 2.0x, and the in-game supersampling set to 1.0x. And while higher supersampling give a better quality image on objects, the difference in text quality was minimal.

The other interesting result came when testing in-game downsampling with the debug tool. For example, using a setting of 1.3x in the debug tool and 1.0x in-game, I get about a 40-45% increase in FPS vs. using 2.0x in the debug tool and 0.65x in-game (=1.3x). And again, everything looked identical to me.

So for the best image, I recommend setting the debug tool to a value somewhere between about 1.1x and 1.5x, with in-game supersampling set to 1.0x, and make sure anti-aliasing set to FXAA or SMAA (for Nvidia cards - MLAA may be better for AMD). Then consider reducing some graphics settings to increase FPS even more.

Note: all my testing was performed sitting in an underground small hangar with ultra quality settings - and I'm not done yet.