Open ali-ramadhan opened 3 years ago
Script compiles and runs without the AMD closure it seems (see diff below) but no idea why.
Trying to reproduce in vanilla Oceananigans.jl but no luck so far as it doesn't even precompile: https://github.com/CliMA/Oceananigans.jl/pull/1701
diff --git a/examples/three_layer_constant_fluxes.jl b/examples/three_layer_constant_fluxes.jl
index 042466d..1a0aedd 100644
--- a/examples/three_layer_constant_fluxes.jl
+++ b/examples/three_layer_constant_fluxes.jl
@@ -293,7 +293,7 @@
model = IncompressibleModel(
tracers = (:T, :c₀, :c₁, :c₂),
buoyancy = buoyancy,
coriolis = FPlane(f=f),
- closure = AnisotropicMinimumDissipation(),
+ # closure = AnisotropicMinimumDissipation(),
boundary_conditions = (T=θ_bcs, u=u_bcs),
forcing = (u=u_sponge, v=v_sponge, w=w_sponge, T=T_sponge,
c₀=c₀_forcing, c₁=c₁_forcing, c₂=c₂_forcing)
+ @info "Try taking a time step..."
+ time_step!(model, 0.1)
@xkykai mentioned that he does not get this error on Supercloud with the same version of packages but with CUDA 11.x with a V100. So the error might be CUDA driver version dependent (possibly also GPU dependent?).
Will see if I can isolate the cause...