Some neat info about improved VRS tech
Microsoft Game Stack VRS update (Series X|S) - Doom Eternal, Gears 5 and UE5 - 33% boost to Nanite Performance - cut deferred lighting time in half
Interesting uses for VRS. That saving for the deffered lighting is massive and definitely something developers will be interested in when it comes to optimisation.
Can someone translate this into something closer to English?
instead of calculating each pixel, you clump pixels in groups depending on necessary granularity. So you can calculate (light, shaders) once to solve for 1 to 4 pixels, increasing efficiency.
Increased efficiency means more overhead for the GPU which means with dynamic resolution, higher resolution where it matters most.
It also means by saving on rendering costs (milliseconds frame times) you can push the GPU to render more advanced features like volumetrics, particle effects, lighting, etc.
Basically it’s a good optimization technique and is supported directly by the hardware and is fairly easy to implement.
much better. Thank you! Im not THAT technically savvy
Here’s an old blog about Gears moving to VRS Tier 2. If you scroll down and look at Hivebusters with the green-orange-yellow-red blocks it visualizes what VRS is doing. 2x2 means it’s only calculating a 2x2 square once (aka one calculation for 4 pixels), whereas the red is still 1:1, as if VRS wasn’t on. But notice how much is green, and in the normal screenshot it doesn’t look bad at all. Big performance saving.
I could be misunderstanding the 2x2 stuff (it seems as though it’s actually more pixels than just 4, and the number refers to the “shading rate”) but it’s the same concept. Less work for visually imperceptible changes = better framerates or more headroom for better graphics elsewhere.
Its pretty expensive to compute the lighting for every pixel even in parts of the picture with not much of contrast or with low frequency, like the sky. So the idea is to calculate these expensive things for some blocks of pixels only one time. VRS allows you to do such a thing. You can specify there for every region of the picture how big these blocks are. Some areas of the picture you want to render full res, like your character in a 3rd person game. Some areas you can live with lover lighting resolution because its not really noticable.
Now comes the important part: VRS is only concerned about the light shading equation, not the geometry resolution. Geometry resolution is always full res, so these blocks of shading don’t manifest themselves like macroblocks in video compression. With VRS you always get clean geometry edges.
So now VRS is being implemented for lighting as well?
So now it will save farme time during calculating lighting and the usual shading as well
With shading i mean calculating the lighting equation per fragment (1x1 to 2x2 pixel block).
So this is the best part
Implementation of VRS
Left image has the lighting being rendered at 1x1 tile (old school)
Right image has the lighting being rendered at 2x1 or 2x2 tile (new vrs implementation for lighting)
Question is when will ALL games start implementing it so we don’t get to see the frame rate gap anymore?
The presenter mentioned that ID tech were impresend with Reinhardlumax2 results and will consider it in next ID tech games
No i believe that was an example of HDR tone mapping with vrs. Reducing rendering in extreme highlights to save on costs rather than only sdr range
The Variable Rate Compute Shading talk on the game stack youtube is about lighting.
You are right
Relax. It’s up to the devs to use the tools provided to them. So, no all games won’t use it especially since these features aren’t present in the lead platform. That’s an unrealistic expectation.
Never until Sony is the market leader.
You could do all of this on the lead platform, PC, unless you mean the #1 console the Nintendo switch… then yeah the switch isn’t doing any of this. the number two console in PlayStation can’t do it either.
Yup, stuff like VRS Tier 2.0 are part of DX 12.0 Ultimate so id expect that to come with time.