What is tbe HMD setting actually changing?

What is the HMD setting actually changing?

I mean I like it, don't get me wrong :D. I'm using 0.65 SS and 1.5 HMD to get the best out of my 970:

LgwC1T5.png


Generally running solid 90 in space and stations and 45 ASW on planets in the tutorials.
I just wish other games had it as an option! Is it essentially downsampling a larger screen size? (Would be good to understand it a bit so I can try and replicate elsewhere. My impression is of decent improvement at 1.5)

(EDIT: Uff, can some kind mod edit the 'tbe' phone typo in the title :/)
 
Last edited:
I mean I like it, don't get me wrong :D. I'm using 0.65 SS and 1.5 HMD to get the best out of my 970:

http://i.imgur.com/LgwC1T5.png

Generally running solid 90 in space and stations and 45 ASW on planets in the tutorials.
I just wish other games had it as an option! Is it essentially downsampling a larger screen size? (Would be good to understand it a bit so I can try and replicate elsewhere. My impression is of decent improvement at 1.5)

(EDIT: Uff, can some kind mod edit the 'tbe' phone typo in the title :/)

Many games have it. It just tends to have nearly as many names as games.
If they don't you can use the oculus debug tool to force a setting.

But most games wouldn't need the same upsample/downsample trick that works for ed.

As for what it does, this is my running theory.

The hmd-q or pixel density ups the reported resolution from the drivers for the hmd.
So if you say had a 100 pixel display, set pixel density to 1.5 the drivers tell the game it's native resolution is 150 pixels.
The supersampling then tell the game to render 65% of that.
Resulting in a 97.5 pixel grid.

Why this would perform better than just running 1x across the board is well ans truly beyond me.
 
Many games have it. It just tends to have nearly as many names as games.
If they don't you can use the oculus debug tool to force a setting.

But most games wouldn't need the same upsample/downsample trick that works for ed.

As for what it does, this is my running theory.

The hmd-q or pixel density ups the reported resolution from the drivers for the hmd.
So if you say had a 100 pixel display, set pixel density to 1.5 the drivers tell the game it's native resolution is 150 pixels.
The supersampling then tell the game to render 65% of that.
Resulting in a 97.5 pixel grid.

Why this would perform better than just running 1x across the board is well ans truly beyond me.

Thanks Tor! (Can't rep you as I've doled out too many to you in the past on this sub-forum it seems :D)

Cool, I figured it would be a known technique (and also poss of specific benefit to ED - it certainly has it's own demands, re distance viewing etc etc), just couldn't see anything obviously the same in games I've checked (which is mainly just Robo Recall so far). Graphics options seem kind of thin on the ground in general.

In Oc Tool Tray I also couldn't see anything that seemed an obvious fit either. Must take a closer look at the Debug Tool (but figured Tray Tool had repackaged all the same stuff). Which setting mirrors HMD quality in those two would you say?
 
Tray tool calls it "Super Sampling", and the oculus debug tool "Pixel Per Display Override".

http://i.imgur.com/jtU23Rb.png

http://i.imgur.com/gkOfBUl.png

Oh what, so the 'Super Sampling' in ED is distinct? I assumed those two were comparable with that, not the HMD quality setting. No wonder I'm getting so confused :D

---

There's been a really cool post over on my Reddit variant of this. I think it has brought me closer to understanding...

For all I surmise, the HMD Quality setting should simply be a mutiplier on the dimensions of the rendered bitmaps, essentially the exact same thing as pixel density in the Oculus debug tool -- something that is offered by at least a few other titles.
The difference between the "Supersampling" and "HMD Quality" options, offered by Elite Dangerous, would be that the latter detemines the size of the final output, relative to what was requested by the Oculus runtime -- the bitmaps are this size when handed over by the game to the Oculus runtime, which then has those larger, more detailed images to work with, when mapping them to the headset displays (accounting for physical resolution, lens distortion, chromatic abberration, (maybe, but probably not, subpixel layout), and head motion that has occurred after the frame started rendering), whereas "Supersampling" does the downsampling on its own, before handing the pictures on, sort of "baking in" the extra detail into its x1.0 output (much like mipmaps), instead of preserving it as long as possible up to the final presentation stage.

You can think of perceived improved clarity, when supersampling in this manner, in terms of the screen door effect: You look at the virtual world through its tiny obsucring mesh, and can see visual samples through each opening. Now you turn your head minutely, and those samples become hidden behind the strands of the screen door mesh, while others that were previously occluded come into view. You might call it something like a a natural temporal supersampling that your nervous system does all the time. Doc-Ok published a nice writeup just the other week, which included the concept: http://doc-ok.org/?p=1631
However: 0.65 times 1.5 equals 0.975. Are you sure you see a marked improvement in clarity over vanilla? -Because you are rendering a smaller image than you would with just x1.0 for both parameters, and is softened twice over - once by FXAA, and once from the upsampling from 0.65.
I don't know - maybe Elite adjusts its LOD and mipmap biases in unknown ways depending on the the two values (if not just the requested bitmap size), and maybe separately for text layers, but I have not experienced any sharper anything, myself, without actually oversampling.
So, I imagine the process would be something like:
ED: Hi, my name is Cobra engine, I would like to render you some stuff. How large do you want the eye textures?
OVR: Hello, Cobby! Make them nx by ny pixels, please.
ED: Okay Ovie. Actually, I'll make that nx times HMDQ by ny times HMDQ, if it's all the same to you...
OVR: Ok, sure.
( ED renders n times HMDQ times SS, then resamples to n times HMDQ )
ED: Here you are: n times HMDQ size eye textures - enjoy.
OVR: Thank you, I will.
:p

Still digesting though. And thinking I need to experiment more...

(Or possibly I should admit I'm out of my depth and let more savvy 970/i5 owners experiment :D)
 
Oh what, so the 'Super Sampling' in ED is distinct? I assumed those two were comparable with that, not the HMD quality setting. No wonder I'm getting so confused :D

---

There's been a really cool post over on my Reddit variant of this. I think it has brought me closer to understanding...



Still digesting though. And thinking I need to experiment more...

(Or possibly I should admit I'm out of my depth and let more savvy 970/i5 owners experiment :D)

Never.
How else would you figure it out?
Try all the things. Ask questions. Try more things. Ask new questions.

As for clarity and the fact that 0.65 and and .1.5 is less than native.
The 970 is probably pushed to max as is in vr and elite.
And for vr in general high fps is a huge contributor to experienced clarity.
So if you gain 10fps by this it could very well seem clearer than if you where just running 1x.

And from what I suspect, and it has been suggested in earlier threads is that a lot of the post processing effects in ED are rendered at a lower resolution when setting the ED supersampling below 1x and thus freeing up gpu workload.

And yes. ED is a bit odd in it having both a supersampling and a HMD-Q setting.
And it doesn't help against confusion that so many other softwares call the latter for supersampling.
 
Last edited:
Never.
How else would you figure it out?
Try all the things. Ask questions. Try more things. Ask new questions.

As for clarity and the fact that 0.65 and and .1.5 is less than native.
The 970 is probably pushed to max as is in vr and elite.
And for vr in general high fps is a huge contributor to experienced clarity.
So if you gain 10fps by this it could very well seem clearer than if you where just running 1x.

And from what I suspect, and it has been suggested in earlier threads is that a lot of the post processing effects in ED are rendered at a lower resolution when setting the ED supersampling below 1x and thus freeing up gpu workload.

And yes. ED is a bit odd in it having both a supersampling and a HMD-Q setting.
And it doesn't help against confusion that so many other softwares call the latter for supersampling.

Dammit I wish I could rep you :D. And yeah I was wondering if improved FPS was playing a role in the feeling of improved clarity when in motion etc. (I was generally checking out the DBX and asteroids in the starter tutorial as my markers, seeing how they looked as I approached once the engine's were working etc).

But I shall head back to the lab and toy with some things. Think I will try a 0.75 / 1.5 ratio with my focus on menu legibility, just to see if I can get anything workable / worthwhile there. Then see if I can find a compromise on the fluid/clarity continuum that works for me :D
 
The end all, be all is that vr is exceedingly personal.
So everyone need to tweak to find what works for them.
What is a high priority for me, might be completely different for someone else.

I personally think any benefit from running high hmd-q is questionable. I much rather have as high an fps as I can get.
So even with a 1080ti I don't use a higher setting than 1.25. Or lately use the debug tool to set a density of 1.33. To me this performs as well as 1.25 and looks as good as 1.5.

But a little upsampling does a lot, but my main goal is 90fps as much as possible. That and if possible I want shadows at ultra :)

If the performance improvements that seem to be in current beta stick it could be possible.
(Currently now, I drop shadows to high, yeah I know. Poor me)

As for readability.
It sort of clicked together for me by just relaxing, and not trying to make out details that simply isn't there.
Now of course you can always lean in a little, but I haven't needed to.

That's not saying for instance the commodities screen isn't blurring some places but it's readable enough to see what's what.

Then again I haven't really done any trading since I made elite in it pre-horizons and long before I got vr...
 
If the performance improvements that seem to be in current beta stick it could be possible.

Every little helps!


You are a scholar and a gentleperson!

Although worryingly they seem to say going above 1.0 in pixelsPerDisplayPixel is pointless... :D

Although you can set the parameter to a value larger than 1.0 to produce a higher-resolution intermediate render target, Oculus hasn't observed any useful increase in quality and it has a high performance cost.
 
Last edited:
Every little helps!



You are a scholar and a gentleperson!

Although worryingly they seem to say going above 1.0 in pixelsPerDisplayPixel is pointless... :D

Yeah its quite amusing. That part of the reference was written back in the day !
 
Last edited:
Every little helps!



You are a scholar and a gentleperson!

Although worryingly they seem to say going above 1.0 in pixelsPerDisplayPixel is pointless... :D

I have a been saying it for a bit, the native resolution of the panel is also the final resolution.
And even great increases in supersampling does not actually increase detail.
And my definition of increased detail means means 'new information'.
But it is impossible to actually display that because the panels are the fixed maximum resolution that can be displayed, But rendering to a higher than your displays resolution is also considered a very high end AA (AntiAliasing).
you might not get much new information displayed in the image, but it will be smoother and certain things might overlap slitghtly better.

It's a bit like a 1080p blueray displayed on a 720p tv. It will look slightly better than native 720p content but that's because the internals can do more edge detection and you have a less aliased source material to go with. Now add to this you have a little constant headmovent to shift which of the rendered pixels are displayed on the panel.

One of the techniques I use subconsciously after a year in the rift to increase readability is to glide my view over what I'm looking at, the brain pieces the motion together and removes the pixelated effect and I see a much higher detail than is actually there.
This works regardless levels of supersampling.
Only going below 1x detracts from it.

Now as you might suspect already.
Elite : Dangerous, is a very special snowflake.

First and foremost. The AA used in elite is simply put. Very poor, and costly when it comes to rendering power.
So costly I actually get better performance running pixel density at 1.33 than activating the in game AA.
(Insert facepalm gif)
And it looks better to boot.

I don't know where but I believe I have seen somewhere (possibly not VR or gaming related) stated that benefits of a higher resolution image, displayed on a certain panel is virtually null when you go over 33%, that is until you almost quadruple the source resolution.
Hence why I'm testing and feel rather happy using exactly 1.33x for my pixel density setting.

Anyhoo.
I'm starting to ramble. It's been two beers and I'm fast approaching midnight here ;)
 
Last edited:
Yeah its quite amusing. That part of the reference was written back in the day !

Yeah was looking for timestamp, kinda assumed it was DK2 days from the way they only referenced 'our first generation DK1' etc.

First and foremost. The AA used in elite is simply put. Very poor, and costly when it comes to rendering power.
So costly I actually get better performance running pixel density at 1.33 than activating the in game AA.
(Insert facepalm gif)
And it looks better to boot.

Oh damn, now I'm going to have to try a SS 1.3 vs no AA comparison...

(All of this is just a long prelude to me buying a 1080, but I figure I have a year until both their prices drop and my wallet recovers enough ;))
 
Yeah was looking for timestamp, kinda assumed it was DK2 days from the way they only referenced 'our first generation DK1' etc.



Oh damn, now I'm going to have to try a SS 1.3 vs no AA comparison...

(All of this is just a long prelude to me buying a 1080, but I figure I have a year until both their prices drop and my wallet recovers enough ;))

Well, I was running no AA, and no supersampling at all with my 980ti so depends how that performs with your gpu.

And after upgrading my i5 4670k, to an i7 4790k, and my 980ti to a 1080ti, I'm still left drooling over the upcoming release of the eight series of intel cpu's.
Six physical cores, more cache, and 12 threads. (current 'reliable' rumours say 11% boost in single thread performance, and 50% boost with multithreaded performance compared to i7 7700k)

And these cpu's will be announced within a month, two at most.

As for the 1080 (or TI), I don't expect another price drop, not until Nvidia's next gen chip, 2080 aka Volta (rumoured model number) release in the first or second quarter of 2018.
That could very well be the GPU that knocks this generation of VR HMD's out of the park, as in stable 90fps with SS and maxed settings, if the optimistic pundits are right.

Especially now that AMD's VEGA is out and disappointing en-masse, with benchmarking establishing it in the 1070 performance range for practically the same price as a 1080, and the psu wattage needed for a 1080ti or Titan Xp.
So Nvidia has no real interest in lowering prices on the 1080 by my guestimates.

Anyhoo. If you have about a year to save up, I would definitely keep an eye out for Volta or 2080 or whatever they will call it.

The beta these past few days was promising, so perhaps I can wait for Volta and build a new monster rig....
Hmm, there's a thought.

Dear lord VR is turning into something expensive....
 
Last edited:
Interesting discussion! Anyone got any thoughts on the suggestion over here that changing this setting using the Oculus debug or tray tools produces a better effect than setting it via the in-game HMD Quality setting?

https://forums.frontier.co.uk/showt...of-questions?p=5840002&viewfull=1#post5840002

It really shouldn't matter, but I suspect it kind of does.

I go with the debug tool for now, since I can set pixel density to 1.33 directly. HMD-Q has no steps between 1.25-and 1.5, and 1.5 doesn't seems to drop into ASW a bit more than I like.
So 1.33 is perfect as far as I am concerned.

One difference between, 1.25 and 1.33 is for instance the box surrounding the current heading in the compass when in orbit or in the SRV, one of the sides of this little box gets clipped out, with 1.33 I can actually see all four sides of this little box.

Not a big deal, but I such things sometimes irritate me :p
 
Last edited:
It really shouldn't matter, but I suspect it kind of does.

I go with the debug tool for now, since I can set pixel density to 1.33 directly. HMD-Q has no steps between 1.25-and 1.5, and 1.5 doesn't seems to drop into ASW a bit more than I like.
So 1.33 is perfect as far as I am concerned.

One difference between, 1.25 and 1.33 is for instance the box surrounding the current heading in the compass when in orbit or in the SRV, one of the sides of this little box gets clipped out, with 1.33 I can actually see all four sides of this little box.

Not a big deal, but I such things sometimes irritate me :p

I get that the debug tool and tray tool might allow finer/different granularity steps to be selected (and that 1.33 might be preferable to 1.25 for a given user's setup) but what I don't get is the suggestion that setting, say, 1.5 in debug tool would give different results to setting 1.5 in the in-game HMD Quality.
 
I get that the debug tool and tray tool might allow finer/different granularity steps to be selected (and that 1.33 might be preferable to 1.25 for a given user's setup) but what I don't get is the suggestion that setting, say, 1.5 in debug tool would give different results to setting 1.5 in the in-game HMD Quality.
Me neither, and since ED is so erratic when it comes to performance, outside of simply trying it out and continuously keep tweaking there is no definite answer.

I daily see settings that perform fine someplace, go for a break, restart game and the same area and same settings starts triggering ASW.
Or more likely, I go finish missions and come back to an artefacting mess.

It's just too much weirdness really, and regardless of how beefy a PC anyone gets, where there is the most benefit to be had is FD optimising their engine code :\
And for this we simply need to cross our fingers and hope.

And of course find some settings that works, and focus more on playing the game than the FPS counter..
 
I turn the in game HMD Setting to 1.0 and use the Oculus Tray Tool at 1.6. If I did the opposite (never use them both at the same time, lowers performance considerably for no visual gain for me) I found the writing more blurry.
 
Back
Top Bottom