Possible race condition in PSO precache collection tasks

Hi there,

thanks for including the callstack. To help debug this crash further, can you please try the following (please note that none of these are fixes and are just meant for debugging purposes):

  • Clear the driver cache before every playtest using the “-clearPSODriverCache” command-line argument. This will delete all the cached PSOs, which will result in longer load times as the PSOs will need to be rebuilt from scratch on every run. This may also result in more hitches during gameplay and potentially trigger the crashes more frequently.
  • Can you try setting r.PSOPrecaching to 0. This will disable PSO precaching to confirm if the crashes are PSO precaching related. This should only be disabled for debugging however (by default this should be enabled).
  • With r.PSOPrecaching enabled again, please try setting the r.PSOPrecache.ProxyCreationWhenPSOReady to 0 and check if this changes the frequency of the crashes
  • Please try setting r.PSOPrecache.ProxyCreationWhenPSOReady to 1 and r.PSOPrecache.ProxyCreationDelayStrategy to 1. This will render meshes which PSO is not ready with a default material. This may also result in less crashes=
  • You may also bring up the PSO stats by setting r.PSOPrecache.Validation to 2 and r.PSOPrecache.Validation.TrackMinimalPSOs to 1, then use the command “stat psoprecache”

You can find more information on these CVars in this link.

Please let me know how that goes.

Thanks,

Sam