I can't get cudaplot to produce reproducible plots. When I use diskplot and the -i arg and I hash the plots (skipping the first 1MB) I get the same hash every time. When I do the same with cudaplot and any -z value, including 0, I get different hashes each time.
I've been trying to figure out why I've been consistently at 3.5-4x effort. Running chia plots check reveals very large numbers of plots dropped due to linepoint compression and quality mismatches. A full 1% are identified as bad. It appears that there is something wrong with my combination of hw/sw and bladebit is churning out a lot of bad plots.
System:
CPU: Opteron 16c
RAM: 256GB DDR4 ECC
GPU: RTX 1070 (x16 but PCIe2.0 due to the platform)
OS: Fedora 37
Kernel: 6.4.12-100.fc37.x86_64
Storage: Optane 905p 960GB (XFS)
Cuda: Driver Version: 535.104.05 CUDA Version: 12.2
bladebit_cuda 3.0.0 installed from rpm downloaded from github
I can't get
cudaplotto produce reproducible plots. When I usediskplotand the-iarg and I hash the plots (skipping the first 1MB) I get the same hash every time. When I do the same withcudaplotand any-zvalue, including0, I get different hashes each time.I've been trying to figure out why I've been consistently at 3.5-4x effort. Running
chia plots checkreveals very large numbers of plots dropped due to linepoint compression and quality mismatches. A full 1% are identified as bad. It appears that there is something wrong with my combination of hw/sw and bladebit is churning out a lot of bad plots.System: CPU: Opteron 16c RAM: 256GB DDR4 ECC GPU: RTX 1070 (x16 but PCIe2.0 due to the platform) OS: Fedora 37 Kernel: 6.4.12-100.fc37.x86_64 Storage: Optane 905p 960GB (XFS) Cuda: Driver Version: 535.104.05 CUDA Version: 12.2 bladebit_cuda 3.0.0 installed from rpm downloaded from github