Game Engine Convergence And The Problem With UE5

Not quite sure where to put this - According to this article, the next iteration Assassins Creed / Anvil will sport ray traced GI and virtual geometry. They'd both be particularly good wins for the settings of these games. Hopefully we'll see more not UE engines adopt virtual geometry.

No mension of VSM though, but then the article's quite slight on graphics talk.

AFAIK it isn't recommended to have one (virtual shadow maps) without the other (virtualized geometry) ...

If you want shadows at all with virtualized geometry, regular shadow maps or multi-resolution cascaded shadow maps aren't ideal due to their uniform sampling property which gets very wasteful in terms of work as the resolution increases so other shadowing methods with non-uniform sampling such as SDSM (sample distribution shadow maps) or VSM can skip having to sample parts of the shadow map at full resolution which is the behaviour that we do want when going to higher resolutions ...
 
AFAIK it isn't recommended to have one (virtual shadow maps) without the other (virtualized geometry) ...

If you want shadows at all with virtualized geometry, regular shadow maps or multi-resolution cascaded shadow maps aren't ideal due to their uniform sampling property which gets very wasteful in terms of work as the resolution increases so other shadowing methods with non-uniform sampling such as SDSM (sample distribution shadow maps) or VSM can skip having to sample parts of the shadow map at full resolution which is the behaviour that we do want when going to higher resolutions ...

Yup, that's why I called it out, even if at this point we don't know how they're approaching shadows.
 
Yup, that's why I called it out, even if at this point we don't know how they're approaching shadows.
They could always just use plain old cascaded shadow maps but it'll look really ugly with aliasing artifacts everywhere ...

I seriously doubt they're considering ray traced shadows for the levels or other high quality assets since not even the leading vendor can handle the overhead of representing virtualized geometry in their acceleration structure. Their solution to anyone using their special UE5 branch was to straight up use screen space shadows ...
 
Not quite sure where to put this - According to this article, the next iteration Assassins Creed / Anvil will sport ray traced GI and virtual geometry. They'd both be particularly good wins for the settings of these games. Hopefully we'll see more not UE engines adopt virtual geometry.

No mension of VSM though, but then the article's quite slight on graphics talk.


Anvil/AC has been running GPU driven meshlets since AC Unity, first or second? ever to ship in a game. Maybe they've got progressive meshlet LODs instead of per model LOD snapping now.

For GI it'll be interesting to see if they bother with animated meshes at all or it's a hybrid static with screenspace dynamic a bit like UE5's software mode. They need to ship on Series S and like crowds too much to spend all that time rebuilding the BVH each frame, at least if they want to maintain performance.
 
Game Maker's Notebook had a chat with the Penny's Big Breakaway folks, who obviously chose to build their own engine for that game. When they came to talking about hiring people, they said a custom engine wasn't the issue they thought. Unity/UE are so prevalent, that there were enough people who've essentially exhausted their interest curve for it. They wanted something else to flex their muscles on.

This clearly doesn't work for every project, especially ones with hundreds of people working on them, but I thought it was an interesting counterpoint the accepted wisdom that custom engines make it hard to find talent.
 
Back
Top