Client - Canary takes 4-5 minutes to start up

  1. Client (production/canary) and FXServer version

Client: Canary
FXServer Version: FXServer-master SERVER v1.0.0.5445 win32

  1. What you expected to happen
    FiveM to instantly start.
  2. What actually happens
    Took 4-5 minutes to start.
  3. Category of bug (eg. client, server, weapons, peds, native)
    Client
  4. Reproducible steps, preferably with example script(s)
  • Start FiveM (in production client)
  • Switch to Canary
  • Restart Client (quite obvious)
  • Start FiveM.

When starting FiveM, it opens the ‘splash screen’ where it says that ‘we’re getting there’, but once that’s done it takes 4 to 5 minutes before FiveM actually opens and you can join a server.

Seems to work normal on production, but as soon as I switched to canary is caused this.

Also seems to happen in Beta :thinking:

Can you record a trace for this? Current canary/beta starts pretty much instantly here, as always.

The guide in the docs should work fine for this.

Followed the guide step by step so should be good;
Tried to upload it twice through the forums, seems to fail, probably the file size? Just uploaded it on workupload.

Trace was done on canary.

Can you check if this is resolved if deleting the contents of %localappdata%\NVIDIA\DXCache? In newer versions we removed our custom ‘hack’ for this which was suspected to be the cause of Slow FiveM start (+1 minute) but cl2 instance is fine (-10 seconds) - #5 by manzarek, but if this apparently also corrupts with it moved back to the original place, that’s… weird, and would mean the underlying issue is likely unfixable (as it’s not reproducible for us, and we don’t have NVIDIA driver symbols, nor does NVIDIA even give a single fuck about bugs we report).

… even more sadly, the latest version of the NvShaderDiskCache code to us (thanks to a specific source that’s been in tech news a few months ago) seems to predate the latest NVIDIA refactor where they moved NVIDIA Corporation\NV_Cache to NVIDIA\DXCache. I’m similarly suspecting that this bug was introduced for some people as of this refactor, as the filename format is different as well.


Yeah, the code that’s failing here 100% is related to new stuff since this refactor.

This fixed it, I assume this will be a ‘temporarily’ fix until NVIDIA gets their guts together?

Alternately, you can make a disable.txt in this directory which will disable the shader cache, note that this may slow down loading on some other games (since this is meant to speed up loading, actually - however this bug instead makes it loop for an eternity or so).

Will try, and if it does slow down other games I will just remove it.
Thanks.