Graphics Update... BGolus and Varrak cleanup rampage

Thanks for posting that! I can't wait to see those epic improvements ingame!

Hopefully we'll see one of the guys post an internal test with them if it doesnt release after PAX.

Yeah, it would be cool if @bgolus or @varrak could do a short video highlighting these features

Will this fix the shadow issues on intel hd 4000 chipset's ?p

I get bad checkerplate shadows effects on the ground on the leading and trailing edge of the main shadow....
Posted pic's





edit..
the main body/bulk of the shadow is fine, it only has the artifacts around the edge...

bloody good read!

that's the whole SAO part bgolus wrote about, this will fix it.

^^

LOL

You just made my French class interesting.

Keep up the awesome work

From my experience with intel hd and intel chipsets in general , the drivers seem to handle shader's differently to the other chipsets (ati,nVidia) with varying results in the rendered effects.
an example of one of those effects being HDR bloom, which is more prominent on the nVidia and ATI's, and has less of a visual impact/bleeding effect on intel..
Shadows on my nVidia GT 320m GPU based mac mini seem fine, i barely notice the artifacts around the shadows..
So you can see why i asked about it fixing the issue on intel hd in specific.

.....


@bgolus
You mentioned about the GBuffer normals being horrible,, Is that causing the banding / posterization of the skybox ???

SAO and the shadow "fix" are two completely separate things, unrelated to each other.

Shadow acne is what maxpowerz and superouman posted images and video of. It's banding or streaking in the shadows, usually most obvious at the edge of shadows where faces are parallel to the light direction. It is not specific to Intel, but occurs everywhere, it's just the luck of the planet mesh and light position you get that determines if it's noticeable or not.

Now, this next comment is important.
Shadow acne is an unsolved issue in rasterized computer graphics!

No one, literally no one has solved it. The various "solutions" aren't really solutions at all but ways to hide the issue as much as possible, always at the cost of artifacts (light bleeding, peter panning, shadow shrinking, blurred or bright edges, etc.) or significant performance loss. But any solution you pick never fully solves the problem, just hides it to be less objectionable.

Here's a before and after of the technique I implemented.

Before: Note the bands / stripes circled in the center and upper right. There's also some light bleed on the ice on the right (above the circled area) where the shadow doesn't quite cover the base edge.


After: The bands in the shadow are gone! ... almost. If you look closely at the top of the ice chunks in the center circle you can still see a little bit of saw-tooth like artifacting. Also, notice the shadows on the little rocks on the left are now smaller and shorter in length! That's the trade off we made to reduce the acne, shadows shrink a little.

That's what the shadows look like on my nVidia based mac,
Did you screen shot from an nVidia GPU based PC??.
I noticed in the pic that the shadow artifacts are no where near as prominent as what they are on my intel chipset.

the shadows get acne regardless of ati nvidia or intel currently. I have a litteral ton of them, much more than in any of your screenshots and I'm running nvidia.

is it another one of those "big check" cases?

I can confirm that. It's happening on my nVidia based mac too.

Also since when is the ice so shiny like that?

GBuffer normals are for geometry lighting. Lighting in computer graphics is most commonly done by comparing the direction (or normal) of a surface against the position or direction of a light. In deferred rendering the normals of a surface are stored in a texture (GBuffer) that gets used later rather than doing the comparison when actually drawing that surface.

The skybox has no normals; it has no "surface", so no normals, it's just a texture. The banding comes from several things, mainly from the fact it's a normal regular 8 bit texture of very subtle gradients (which will be banded due to the limited colors available) compressed with DXT1 (which can also add banding) that's then used in an HDR rendering pipeline which can try to brighten the image further (which will show off the banding worse) or add bloom (which will can cause banding).

So ... yeah. There's banding.

* edit for clarity
* edit for accuracy, we aren't DXT1 compressing the skybox anymore

Is there a form of quantization that could solve that issue?
Maybe dithered before compressing to dxt1 or using some form of spatial color quantization before or during the compression process...

I was thinking of things like you mentioned to me ages ago like a film-grain filter or noise filter, maybe even some form of Rectangular Probability Density Function or Triangular Probability Density Function to make the noise to help reduce this posterized effect.

i don't understand the technical aspect, but i do understand this will make the game function and/or look better

I do actually want to add some form of dithering in the HDR pipeline when down converting the HDR frame buffer during tone mapping, but for it to fix the sky boxes they would have to be authored dithered. You'd just have noisy bands if you did it as part of the DXT compression.

Someone give this man a raise!

Wow, it's looking absolutely awesome

Will these effect the lava planets in any sort of way too?
Like, will dark sides of lava planets get a sort of lava-ish glow if land is close to a pool of lava?
If so, can we maybe get a screenshot maybe possibly?

Very cool stuff guys!

oh boy that is looking damn bloody close to the original concept video now.... I cant wait to see this in action.

Have you guys played around with the Bloom around Lava yet? was a question I asked awhile back in one of the livestreams and the main problem was that when you added Bloom to lava, it was adding bloom to everything? and started to white out things like the sun so I was wondering if you had made any progress on this? have you found a different technique to perhaps make the lava "pop out" like it does on the concept pictures, right now its still too orange with almost zero affects or brightness to it. How does the new effect discussed on this thread affect lava?