Peter Chamberlain wrote:In the release version of 20 the optimisation is automatic if you are on the minimum CUDA version listed in the read me.
Does that mean we can delete the 4GB+ tensor cache file and it will make no difference to performance or is Resolve still using it if you installed a beta where the generation was required and the performance increase is best using the the cache file?
With this rushing to release version I could see why BMD may get rid of a step that many people complained about due to the hour it took to run and also crashing possibly related to it (for some people) and still others where each run Resolve wanted to regenerate the cache.
Is not optimising the tensor cache actually a step backwards as far as performance but a step forward as far as users complaining less?
If the cache file is no longer being used or needed my MagicMask2 results are the same, and did not get worse.