Sharpening: (too many) options?

[mrsylvestre]

Hello,

Nowadays, there are 3 or 4 places where sharpening options can be set for VR: in DCS graphics settings (sharpening slider), in OXRTK, in QuadViews (settings file) and, if using a non-DP headset, in the streamer (Virtual Desktop, Pico Connect, Oculus link,...). Any advice on how to best set the sharpening settings, and at which stage of the rendering chain? Is it similar to supersampling which is best set at one place, not spread out between different stages, or is sharpening better applied in stages?

I did a few tests and clearly sharpening applied at multiple stages does seem to add up. This got me wondering about what the best strategy could be.

[Nodak]

Safest bet is to use the games sharpen filters first above all else, no one knows the specific game graphics like the games coders.  There's always a point of diminishing returns when sharpen is overlaid and over done, than it's 'Welcome to Shimmer City.'

[mbucchia]

1 hour ago, Nodak said:

Safest bet is to use the games sharpen filters first above all else, no one knows the specific game graphics like the games coders.  There's always a point of diminishing returns when sharpen is overlaid and over done, than it's 'Welcome to Shimmer City.'

That's the right answer, since the game one is done at the proper stage of rendering.

I'll add for the streaming options (like Virtual Desktop) it's also nice to use the sharpening in the streaming app (done on the headset side) since it happens after the encoding+decoding, where some of the sharpeness was lost. Note that sharpening before the encoding will also increase bandwidth but it's quite necessary as it produces better results.

[Dangerzone]

Nodak got it right. Only thing I'll add - you can't have too many options. You don't need to use them, but there could be a reason that someone else wishes to use a different sharpening reason that's not relevant to us. I'll always accept more options than less. 

[WipeUout]

I wanted to test this and kinda forgot about it, so I gave it a go.

I as expecting that using quadview CAS sharpening would provide best performance since it is only the focus are that is getting sharpen?  I was wrong, something else is at play here.

Results show not a whole lot of difference between sharpening options but clearly, CAS in OXRTK has the edge not only for a lower GPU frametime but also for lower CPU frametime.  Also of note, OXRTK sharpen offers a slightly better performance than not using any sharpen option!

I repeated my testing several times and results are consistant at +/- 0.02 ms.

It is entirely possible that you could get different results for other HMDs. 

As far as quality of visual, OXRTK CAS and QV CAS offer similarly sharp visuals at 100% and DCS a little less sharp at 100%.

Note: DLSS B = balanced setting, FT= Frametime in milisecond.

Edited July 8, 2024 by WipeUout
Typo

[mrsylvestre]

I also experimented a bit, although in a less systematic way than @WipeUout

Performance wise (frame times, fps) , it doesn't seem to make a big difference indeed. Quality wise, reasonable sharpening before encoding and a bit on top on the headset side seems to be a good recipe on my system. Thank you all for the advice and tests above.