#651 is big enough as-is. There are some things to do (or at least attempt) post-merge. In no particular order:
- Unify 64-bit and 32-bit zluda.exe
- Try doing RPC through shared memory:
as for bottleneck - in fluidmark i get 45-48fps at default benchmark settings with meager 13% of compute load(rx6750xt). this is on level of iirc gtx680, but rx definetly has more guts than it.
same with nvidia demos - load doesn't exceed ~30%
case in point: SampleCloth.exe scene 2 is ~150 fps on hw(compute load 30%) vs ~350 on cpu
- Fix problems with reinitializing physics
- Check for client/server version mismatch on start
- Suggest using nvidia sdk/fluidmark for troubleshooting
- Be explicit about no need for hip sdk for physx
- Make sure we use the most powerful gpu
#651 is big enough as-is. There are some things to do (or at least attempt) post-merge. In no particular order: