Ansel and Simultaneous Multi-Projection - Two new exciting techs

[Gregster]

21:9? you need to get some 32:9 curved screen action, is pretty much double monitors but without the annoying bezel and gap in between.

Yer, I quite fancy one of those when my wallet allows (and Mrs).

 

[pmc25]

They also have a new tech which helps massively with VR rendering at no IQ cost at all. This is probably where the 2x Titan X performance thing comes from.

That statement is ridiculous.

They have new software (you seem to be implying it's a hardware feature) that provides a SMALL increase in FPS specifically at the cost of IQ.

LOL.

 

[Nexus18]

There's no master race in 21:9...stop fooling yourself mate Plenty of games with no support, new games without it, and others where you have to do some sort of tom foolery to get it working

Plenty of games do have it though and most of those that don't can be easily fixed with just tweaking a line or 2 in the user.cfg or via flawless widescreen (just 1 click to dll/activate the profile and that is it), only 1 game has required a lot of faff to get it working well and that was fallout 4.

AFAIK, the only recent AAA game that has no support at all is overwatch, developers give some BS excuse of it not being fair to 16.9 users and that it didn't match their art style lol...

Edit:

Apparently overwatch does support 21.9 now.

Besides, it is worth the little faff that you have to go through every now and then, 16.9 is just too squashed looking now imo, coming from a 21.9 screen to a 16.9 screen is like a 16.9 user going back to 4.3

Spoiler

21:9? you need to get some 32:9 curved screen action, is pretty much double monitors but without the annoying bezel and gap in between.

Too wide for me!

34" 21.9 is the absolute max that I would go for a monitor.

 

[NirK]

Hm I don't take many screenshots and I don't have multi monitor, my interest is 95% performance at 3440x1440 and also 5% VR performance which I may get in the future.

 

[stooeh]

Afaik the entire point of the VR rendering gain was that they do it at an IQ cost. They want to lower the resolution of the outer edges as you aren't looking directly at them so you won't notice it.

http://www.pcgamer.com/nvidia-gtx-1080-1070-features-detailed/

So it's not about no IQ loss, it's directly about reducing resolution to gain frame rate, it's a complete cheat and absolutely reduces IQ.

I think you misunderstand, there's no IQ loss/cheating going on here. It simply leverages their new multi projection whatever to render a smaller amount of pixels equal to the size you can actually see through the lenses.

I'm pretty sure valve already has a similar technique in software which Alex Vlachos talked about in his GDC VR rendering speech. He also mentioned a technique (foveated rendering?) that does lower the image quality/resolution of the visible outer portions of the screen but that is not what is going on here.

The article you linked says as much also.

The bottom line is the same, however: fewer pixels rendered without a loss in image quality.

 

[drunkenmaster]

I think you misunderstand, there's no IQ loss/cheating going on here. It simply leverages their new multi projection whatever to render a smaller amount of pixels equal to the size you can actually see through the lenses.

I'm pretty sure valve already has a similar technique in software which Alex Vlachos talked about in his GDC VR rendering speech. He also mentioned a technique (foveated rendering?) that does lower the image quality/resolution of the visible outer portions of the screen but that is not what is going on here.

The article you linked says as much also.

The game was running at 4K but only hitting about 50 fps, just a bit below the desired 60 fps for a 60Hz display. By flipping a switch, the game renders the outer portions of the display at a lower resolution and stretches these, leaving the main section of the display—where you're most likely focused during gaming—at full quality. Frame rates jumped from 45-50 fps without multi-projection to over 60 fps, and while there was a slight loss in quality, it was only really visible if you were stationary and carefully looking for the change.

If you read it carefully then no, what you said isn't remotely accurate. There is a specific IQ reduction, the very reason for the improved performance is reducing the resolution to parts of the screen and once again it's because they assume you are only looking dead centre and move the mouse and never your eyes.

You can't talk about performance at 4k if you're rendering 30-40% of the screen at maybe half that resolution because that simply isn't 4k. The actual performance improvements over last gen are coming from rendering parts of the screen at a lower resolution, the prewarp stuff and rendering less pixels is something they already did on Maxwell 2.0 as they specifically talk about how they did it on Maxwell already, it's just improved/different which is how all technology works. Algorithms are never perfect, you make one then improve it iteratively. SSAO>HBAO>HBAO+, trying to improve efficiency by not rendering stuff you don't actually need is one thing and it isn't a new idea in VR or for Nvidia. The actual direct increase in performance for the end result is being achieved by rendering a large option of the image at a lower resolution.

 

[andybird123]

Its a choice, dont like it dont use it.

For VR it absolutely does increase fps without reducing IQ though

And for surround it increases IQ without reducing fps

 

[drunkenmaster]

Its a choice, dont like it dont use it.

For VR it absolutely does increase fps without reducing IQ though

You are literally wrong, the article says you're wrong, everything I've seen suggests Nvidia is lowering resolution on parts of the image to increase performance. Saying this increases FPS without reducing IQ is nothing short of ridiculous.

 

[drunkenmaster]

no idea what happened there... pretend this never happened.

 

[Mauller]

You are literally wrong, the article says you're wrong, everything I've seen suggests Nvidia is lowering resolution on parts of the image to increase performance. Saying this increases FPS without reducing IQ is nothing short of ridiculous.

That is one Method they are talking about for high resolution use of SMP, where the centre of the screen can remain at full resolution while the peripheries are at a lower resolution.

But for VR and multi monitor situations the resolution are the same, the only thing that is different is a shift in the shape of the view port for the triple monitor setup and breaking down a single viewport into smaller viewports for the VR situation.

For the VR situation this means they can pre-distort the image without having to stretch it and lower its quality as they currently do with a flat viewport. instead they break the single viewport down into 4 and shape them so it pre distrorts the image without having to have a higher res viewport to make up for the lower IQ from the need to stretch the original image. So in other words it improves performance by reducing the need to render at a higher resolution.


But SMP and single pass rendering was talked about back in 2003-2009 in some research papers. especially the single pass rendering. SMP is just software, there is nothing hardware dependent about it. Many pieces of software already make use of multiple viewport rendering, just they tend to use them for completely different views, rather than a view that focuses on a single point as with the triple monitor setup Nvidia showed.

Some games also make use of Multiple view ports and have built in multi monitor configurations. Flight sims are one, also a few racing games.

 

[andybird123]

You are literally wrong, the article says you're wrong, everything I've seen suggests Nvidia is lowering resolution on parts of the image to increase performance. Saying this increases FPS without reducing IQ is nothing short of ridiculous.

Congratulations on not being able to grasp the basic fundamentals of how this works.

I suggest you read up on VR render targets

 

[CAT-THE-FIFTH]

Here is Nvidia explaining it:

https://developer.nvidia.com/sites/...works/vr/GameWorks_VR_2015_Final_handouts.pdf

Our next topic is multi-resolution shading.
The basic problem that we’re trying to solve is illustrated here. The image we present on a
VR headset has to be warped to counteract the optical effects of the lenses.

In this image, everything looks curved and distorted, but when viewed through the lenses,
the viewer perceives an undistorted image.

The trouble is that GPUs can’t natively render into a distorted view like this –
it would make triangle rasterization vastly more complicated.

Current VR platforms all solve this problem by first rendering a normal image (left) and
then doing a postprocessing pass that resamples the image to the distorted view (right).
If you look at what happens during that distortion pass, you find that while the center of
the image stays the same, the edges are getting squashed quite a bit.

This means we’re over-shading the edges of the image. We’re generating lots of pixels that
are never making it out to the display–they’re just getting thrown away during the
distortion pass. That’s wasted work and it slows you down.

The idea of multi-resolution shading is to split the image up into multiple viewports
–here, a 3x3 grid of them.

We keep the center viewport the same size, but scale down all the ones around the edges.
This better approximates the warped image that we want to eventually generate, but
without so many wasted pixels. And because we shade fewer pixels, we can render faster.
Depending on how aggressive you want to be with scaling down the edges, you can save
anywhere from 25% to 50% of the pixels. That translates into a 1.3x to 2x pixel shading
speedup.

Another way to look at what’s going on here is as a graph of pixel density aross the image.
The green line represents the ideal pixel density needed for the final warped image. With
standard rendering, we’re taking the maximum density – which occurs near the center
– and rendering the whole image at that high density.
Multi- resolution rendering allows us to reduce the resolution at the edges, to
more closely approximate the ideal density while never dropping below it.

In other words, we never have less than one rendered pixel per display pixel, anywhere in the image.
This setting lets us save about 25% of the pixels, which is equivalent to 1.3x improvement
in pixel shading performance.

It’s also possible to go to a more aggressive setting, where we reduce the resolution in the
edges of the image even further. This requires some care, as it could visibly affect image
quality, depending on your scene. But in many scenes, even an aggressive setting like this
may be almost unnoticeable – and it lets you save 50% of the pixels, which translates into a
2x pixel shading speedup.

Remember,this is only for VR headsets,and is taking advantage of the fact our periphery vision is not as good.

 

[stooeh]

But isn't that the 'old' method multi-resolution shading and not the new feature 'Multi projection'?

 

[CAT-THE-FIFTH]

But isn't that the 'old' method multi-resolution shading and not the new feature 'Multi projection'?

I personally don't see a problem with something like MRS,or its adaptation into other areas.

Like I said human periphery vision is poorer,so its better the processing power is more intelligently utilised.

The PC has a tendency to brute force things,so maybe better use of the processing power is a good step forward.

It is why consoles tend to punch above their weight.

 

[drunkenmaster]

The problem is as discussed before, you can move your head and eyes separately, changing the quality at the edge of the screens locks you to looking only at the centre of the screen, to see something on the edge in full quality you have to move the mouse to centre it rather than your eyes. That is the problem with DoF, it effectively chooses for you where you were focusing rather than let you choose.

As I said it's much less bad as it saves performance rather than uses more as with DoF which is what made DoF such a ludicrously awful thing to put into games, but it's still unrealistic. I don't know how you can at the same time say woo, surround gaming improves, more realism with a bigger field of view but on the other hand say, you don't even look at anything but the very centre of your middle screen so lets reduce the IQ on the outside of the middle screen let alone 2 further screens.

The message is literally opposite, you need more peripheral vision with better quality, but peripheral vision is pointless, reduce IQ and save performance, it's important or it's unimportant, it can't be both.

Now marketing surround as increased IQ and reduced resolution at edge of the screen as performance at the cost of IQ makes sense, but marketing them both as the way forward at the same time is just marketing nonsense and misleading.

 

[AlamoX]

multi-projection look very interesting, makes you think why nobody did it before, and the VR numbers sounds amazing, need more info about it thou, how consistant and so on, i hope it's not patented and AMD could make something similar.
Hansel on the other hand i found it kinda lame, there is a reason game Devs do not bother putting stuff like these in their games, nobody cares about it, and i think one of the reasons i find it so lame is the reaction of that chick at nvidia event, sounded like she was about to have an orgasm hearing about it... i dont know the contrast of the situation, just added to the lameness of the feature for me

 

[Klo]

Don't really care about the screenshot technology, and how many people in the wider market have multiscreen set ups or VR? That technology is very much for the 1%.

 

[D.P.]

Multiprojecection has nothing to do with lowering resolution for the peripheral vision.

A naive VR rendering solution is to render the left eye view-port and then independently render the right eye view-port in a complete new rendering pass. Multiprojection render allows both eyes to render simultaneously, sharing resources and optimizing rendering between the 2 views.

Resolution isn't changed in the slightest anywhere in the scene, merely redundancy is removed between rendering the left and right views.

 

[Gregster]

You are literally wrong, the article says you're wrong, everything I've seen suggests Nvidia is lowering resolution on parts of the image to increase performance. Saying this increases FPS without reducing IQ is nothing short of ridiculous.

The amount of times you have claimed this and that lately and got it wrong, you will have to excuse me if I don't take your word for it.

 

[Mauller]

Multiprojecection has nothing to do with lowering resolution for the peripheral vision.

A naive VR rendering solution is to render the left eye view-port and then independently render the right eye view-port in a complete new rendering pass. Multiprojection render allows both eyes to render simultaneously, sharing resources and optimizing rendering between the 2 views.

Resolution isn't changed in the slightest anywhere in the scene, merely redundancy is removed between rendering the left and right views.

Not entirely right, you are talking about single pass rendering which allows a single pass to be used to render multiple viewport frustrum.

Multi-projection does allow the creation of multiple viewports with some being lower resolution than others. But the main benefit of multi-projeciton is fixing the arrangement of the viewports so they represent either the monitor layout correctly or are used to remove the need for having a higher resolution flat viewport and pre distorting an image for VR.