i will test to confirm

@doitsujin possibly another Blackwell specific bug? might not show on Ada and before.

Yeah, I worry it's a Blackwell thing. It's failing very very early, before the main window even appears, as I said. So I think the actual game configuration is irrelevant, but I'll try some experiments this evening if no one else can independently repro. But I was definitely able to confirm that reverting the commit allowed it to run.

@philipl does it happen on 595 or the 595 Vulkan Beta Drivers?

I experience a similar issue with MH Wilds with 610 but not on 595 with an RTX5070 mobile

@Eckoa I need to try. I'm going to downgrade anyway because there's a significant performance regression with these drivers. Will report if it makes a difference.

Never change, Nvidia.

Can't reproduce on master + RTX 5090 + 610 driver + VKD3D_CONFIG=descriptor_heap
Perhaps try without any mods, mangohud, etc.

I didn't get around to looking at the actual proton log until just now, but it shows an access violation in glcore. The log also tells me that the game is fully crashing but the crash reporter isn't starting properly so that's why it doesn't terminate cleanly. That's not important.

The logs remind me that installed the latest streamline and dlss dlls which is generally needed to get smooth frame gen, so I will try reverting back to shipped dlls and see if that makes a difference.

Ok. Make of it what you will, but with the streamline libraries reverted to the shipped ones, it does not crash, but FG is also broken (frames out of order).

I grabbed the latest dlls from https://github.com/NVIDIA-RTX/Streamline/releases/ and replaced whichever ones were present in the game binaries.

And well, turns out I wasn't using the latest 2.11.1 streamline files. When I installed those, everything worked. I guess the bad interaction was due to the specific older streamline release (2.10.3). I assume that's not really worth investigating. I guess we can close this. Thanks!

@HansKristian-Work Sorry, can we reopen this.

It turns out that my "works fine with shipped streamline libs" test was invalid. When I revalidated the game in steam to download the old streamline files, it also refreshed Proton Experimental and so undid the updated vkd3d-proton. I reran the tests today with 595 (30% faster than 610 in a single location comparison, but business as usual) and it failed in the same way. I did the full bisect from scratch twice and made sure to test the original streamline files and got exactly the same result.

I have to revert the commit for descriptor_heap to work with the game, regardless of streamline version.

I recognise that other folks have not been able to reproduce, so I'm going to do more testing with different in-game configuration to see if I can find what the difference is.

Whelp. After doing some poking around locally, I started wondering if it was a build problem. So I downloaded the latest master artefact from github and tried running that - and it worked. So somehow the binary produced by my local build is problematic.

I built using ./package-release.sh and installing whatever dependencies and tooling were required from standard ubuntu 26.04 packages.

If anyone would like to repro independently, doing a build on Ubuntu 26.04 with the distro's mingw should produce the same results. I'm also happy to throw my binaries up somewhere but I can understand a reluctance to just run a random person's random build...