Should/can I use one_degree_near_global_simulation.jl? #101
anthony-meza
started this conversation in
General
Replies: 4 comments 22 replies
-
|
We haven't been working much on the one degree model, which requires using the somewhat unfinished |
Beta Was this translation helpful? Give feedback.
2 replies
-
|
Ok! Sweet I'll look into it 👀 |
Beta Was this translation helpful? Give feedback.
-
|
Check out https://github.com/CliMA/ClimaOcean.jl/blob/main/examples/near_global_omip_simulation.jl |
Beta Was this translation helpful? Give feedback.
-
|
It's pretty awesome that there are so many example simulations that users can take and modify - it's a game-changer for accelerating progress in ocean science! Q: Are the examples in the I am trying to run the one degree simulation on throws the error Has there been a recent syntax change? |
Beta Was this translation helpful? Give feedback.
6 replies
-
|
Here's the docstring (currently!) help?> ECCORestoring
search: ECCORestoring
ECCORestoring(variable_name::Symbol, [ arch_or_grid = CPU(), ];
version = ECCO4Monthly(),
dates = all_ECCO_dates(version),
time_indices_in_memory = 2,
time_indexing = Cyclical(),
mask = 1,
rate = 1,
inpainting = NearestNeighborInpainting(Inf),
cache_inpainted_data = true)
Build a forcing term that restores to values stored in an ECCO field time series. The restoring is applied as a forcing on the
right hand side of the evolution equations calculated as
Fψ = r μ (ψ_{ECCO} - ψ)
where μ is the mask, r is the restoring rate, ψ is the simulation variable, and the ECCO variable ψ_ECCO is linearly interpolated
in space and time from the ECCO dataset of choice to the simulation grid and time.
Arguments
≡≡≡≡≡≡≡≡≡
• variable_name: The name of the variable to restore. Choices include: * :temperature, * :salinity, * :u_velocity, *
:v_velocity, * :sea_ice_thickness, * :sea_ice_concentration.
Note that `ECCOMetadata` may be provided as the first argument instead
of `variable_name`. In this case the `version` and `dates` kwargs (described below)
cannot be provided.
• arch_or_grid: Either the architecture of the simulation, or a grid on which the ECCO data is pre-interpolated when
loaded. If an architecture is provided, such as arch_or_grid = CPU() or arch_or_grid = GPU(), ECCO data will be
interpolated on-the-fly when the forcing tendency is computed. Default: CPU().
Keyword Arguments
≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡
• version: The version of the ECCO dataset. Default: ECCO4Monthly().
• dates: The dates to use for the ECCO dataset. Default: all_ECCO_dates(version).
• time_indices_in_memory: The number of time indices to keep in memory. The number is chosen based on a trade-off between
increased performance (more indices in memory) and reduced memory footprint (fewer indices in memory). Default: 2.
• time_indexing: The time indexing scheme for the field time series.
• mask: The mask value. Can be a function of (x, y, z, time), an array, or a number.
• rate: The restoring rate, i.e., the inverse of the restoring timescale (in s⁻¹).
• inpainting: inpainting algorithm, see inpaint_mask!. Default: NearestNeighborInpainting(Inf).
• cache_inpainted_data: If true, the data is cached to disk after inpainting for later retrieving. Default: true.A few things have been moving as we try to refine and streamline the user interface. In particular it was confusing that you could provide both |
Beta Was this translation helpful? Give feedback.
-
For sure! Thanks for the detailed explanation re: And, sorry, the docstring is pretty self-explanatory - I promise to check there next time first 😄! (I notice my brain getting lazier approaching the end of December...) Since we are already here, I will say that fails for another reason I will go investigate! 🤓 |
Beta Was this translation helpful? Give feedback.
-
|
seems like a bug with the show method of the |
Beta Was this translation helpful? Give feedback.
-
|
So just to say @NoraLoose, the bug is only in the show method (the method that displays information for the object in the terminal). The ECCOrestoring should have constructed. If you running the script line-by-line in the REPL, you can suppress the show method from displaying something if you add I pushed a fix in #311. |
Beta Was this translation helpful? Give feedback.
-
|
As a side note, this [2] show(io::IOContext{IOBuffer}, p::ECCORestoring{FieldTimeSeries{Center, Center, Center, Oceananigans.OutputReaders.Cyclical{Float64}, ClimaOcean.DataWrangling.ECCO.ECCONetCDFBackend{ClimaOcean.DataWrangling.ECCO.on_native_grid, true, NearestNeighborInpainting{Float64}, ECCOMetadata{StepRange{DateTimeProlepticGregorian, Month}, ECCO4Monthly}}, Tuple{Colon, Colon, Colon}, OffsetArrays.OffsetArray{Float64, 4, CUDA.CuArray{Float64, 4, CUDA.DeviceMemory}}, ImmersedBoundaryGrid{Float64, Periodic, Oceananigans.Grids.RightConnected, Bounded, OrthogonalSphericalShellGrid{Float64, Periodic, Oceananigans.Grids.RightConnected, Bounded, Oceananigans.Grids.StaticVerticalCoordinate{OffsetArrays.OffsetVector{Float64, CUDA.CuArray{Float64, 1, CUDA.DeviceMemory}}, OffsetArrays.OffsetVector{Float64, CUDA.CuArray{Float64, 1, CUDA.DeviceMemory}}}, OffsetArrays.OffsetMatrix{Float64, CUDA.CuArray{Float64, 2, CUDA.DeviceMemory}}, OrthogonalSphericalShellGrids.Tripolar{Int64, Int64, Int64}, GPU}, GridFittedBottom{Field{Center, Center, Nothing, Nothing, OrthogonalSphericalShellGrid{Float64, Periodic, Oceananigans.Grids.RightConnected, Bounded, Oceananigans.Grids.StaticVerticalCoordinate{OffsetArrays.OffsetVector{Float64, CUDA.CuArray{Float64, 1, CUDA.DeviceMemory}}, OffsetArrays.OffsetVector{Float64, CUDA.CuArray{Float64, 1, CUDA.DeviceMemory}}}, OffsetArrays.OffsetMatrix{Float64, CUDA.CuArray{Float64, 2, CUDA.DeviceMemory}}, OrthogonalSphericalShellGrids.Tripolar{Int64, Int64, Int64}, GPU}, Tuple{Colon, Colon, Colon}, OffsetArrays.OffsetArray{Float64, 3, CUDA.CuArray{Float64, 3, CUDA.DeviceMemory}}, Float64, FieldBoundaryConditions{BoundaryCondition{Oceananigans.BoundaryConditions.Periodic, Nothing}, BoundaryCondition{Oceananigans.BoundaryConditions.Periodic, Nothing}, BoundaryCondition{Oceananigans.BoundaryConditions.Flux, Nothing}, BoundaryCondition{OrthogonalSphericalShellGrids.Zipper, Int64}, Nothing, Nothing, Oceananigans.BoundaryConditions.DefaultBoundaryCondition{BoundaryCondition{Oceananigans.BoundaryConditions.Flux, Nothing}}}, Nothing, Oceananigans.Fields.FieldBoundaryBuffers{Nothing, Nothing, Nothing, Nothing, Nothing, Nothing, Nothing, Nothing}}, Oceananigans.ImmersedBoundaries.CenterImmersedCondition}, Nothing, Nothing, GPU}, Float64, FieldBoundaryConditions{BoundaryCondition{Oceananigans.BoundaryConditions.Periodic, Nothing}, BoundaryCondition{Oceananigans.BoundaryConditions.Periodic, Nothing}, BoundaryCondition{Oceananigans.BoundaryConditions.Flux, Nothing}, BoundaryCondition{Oceananigans.BoundaryConditions.Flux, Nothing}, BoundaryCondition{Oceananigans.BoundaryConditions.Flux, Nothing}, BoundaryCondition{Oceananigans.BoundaryConditions.Flux, Nothing}, BoundaryCondition{Oceananigans.BoundaryConditions.Flux, Nothing}}, CUDA.CuArray{Float64, 1, CUDA.DeviceMemory}, Nothing, Nothing, @NamedTuple{}}, Nothing, LinearlyTaperedPolarMask{Tuple{Int64, Int64}, Tuple{Int64, Int64}, Tuple{Float64, Int64}}, ClimaOcean.DataWrangling.ECCO.Temperature, Float64})
@ ClimaOcean.DataWrangling.ECCO /pscratch/sd/n/nloose/depot/packages/ClimaOcean/1uNfv/src/DataWrangling/ECCO/ECCO_restoring.jl:338was the hint that lead me to the show method. |
Beta Was this translation helpful? Give feedback.
-
|
I'm trying to run I updated the Full error |
Beta Was this translation helpful? Give feedback.
11 replies
-
These are two different runs - physics only fails at I'll see if |
Beta Was this translation helpful? Give feedback.
-
|
Oh, duh, I'm using |
Beta Was this translation helpful? Give feedback.
-
|
@jagoosw @glwagner @simone-silvestri Thanks for the suggestions. I installed my own cuda toolkit (v12.2.2, which is the highest compatible version with the installed driver). Step-by-step instructions here. With updated PTX the parameter memory error has indeed resolved in both physical and bgc cases :D TYVM! |
Beta Was this translation helpful? Give feedback.
-
|
@jagoosw what's the performance hit? |
Beta Was this translation helpful? Give feedback.
-
I hadn't looked into it before but thought the change in 12.1 worked around device parameter memory limitations (which I assumed would induce a performance hit because it would have to use slower memory), but I was wrong it doesn't cause a performance hit: https://developer.nvidia.com/blog/cuda-12-1-supports-large-kernel-parameters/ For future reference, the limit is now 32KB so we should be good for a while. |
Beta Was this translation helpful? Give feedback.
-
|
On CPU (because it took longer to run) I'm seeing an error after 1 day originating in Full error |
Beta Was this translation helpful? Give feedback.
3 replies
-
|
To solve a physics problem you may need to tweak the setup with a smaller time step, higher viscosities, etc. but it’ll be important to get GPU working first otherwise turnaround is too slow |
Beta Was this translation helpful? Give feedback.
-
|
I have added a configuration that should work in #276. For the parameter space problem at the moment, I don't think we can do much except @glwagner's idea to reduce the precision of the offsets in the offset arrays. A quick solution is to use a newer GPU which should be able to hold that parameter space |
Beta Was this translation helpful? Give feedback.
-
|
Another temporarily solution is to split up the kernels |
Beta Was this translation helpful? Give feedback.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Curious about using this code to simulate some age tracers... I wasn't sure if I should use it yet in case there was still some major ongoing debugging
Beta Was this translation helpful? Give feedback.
All reactions