Benchmarks

Performance benchmarks for SpeedyWeather.jl across architectures. This page is auto-generated at doc-build time from SpeedyWeather/benchmark/assets/benchmark_results.json, which itself is updated by SpeedyWeather/benchmark/manual_benchmarking.jl. Running the benchmark script on a new architecture adds a column to the overview table, a series to each comparison figure, and a per-architecture section to this page.

All simulations are benchmarked over several seconds (wallclock time) without output. Benchmarking excludes initialization. Timings can vary by ±50% between runs, so treat the numbers as rough rather than precise.

Overview: PrimitiveWet resolution across architectures

Simulated years per wallclock day (SYPD) for the PrimitiveWetModel resolution sweep, one column per architecture. Each (T, L) configuration is reported for both the standard Legendre transform and fast Fourier transform (LT+FFT) and the single matrix transform (MT). Empty cells (—) mean the architecture has either not been benchmarked yet or did not run that specific configuration.

T	L	Transform	cpu-arm	cpu-x86	gpu-nvidia
32	8	LT+FFT	1400	856	5879
32	8	MT	757	107	5730
43	8	LT+FFT	564	370	3779
43	8	MT	272	28	3809
64	8	LT+FFT	147	107	1188
64	8	MT	48	3.7	1186
86	8	LT+FFT	57	39	659
86	8	MT	13	0.9	305
86	16	LT+FFT	51	49	566
86	16	MT	19	1.1	137
86	24	LT+FFT	48	38	532
86	24	MT	11	0.9	78
128	8	LT+FFT	15	11	264
128	8	MT	1.5	0.1	32
128	16	LT+FFT	21	15	236
128	16	MT	2.1	0.2	14
128	24	LT+FFT	15	11	219
128	24	MT	1.7	0.1	9.2
171	8	LT+FFT	5.5	3.9	138
171	16	LT+FFT	7.7	5.6	120
171	24	LT+FFT	4.3	4.5	110
256	8	LT+FFT	1.4	1.0	53
256	16	LT+FFT	1.9	1.3	44
256	24	LT+FFT	1.5	1.0	38

Architecture: cpu-arm

Created for SpeedyWeather.jl v0.21.1+DEV on Tue, 21 Jul 2026 17:31:54.

Machine details — cpu-arm

julia> versioninfo()
Julia Version 1.12.6
Commit 15346901f00 (2026-04-09 19:20 UTC)
Build Info:
  Official https://julialang.org release
Platform Info:
  OS: macOS (arm64-apple-darwin24.0.0)
  CPU: 8 × Apple M3
  WORD_SIZE: 64
  LLVM: libLLVM-18.1.7 (ORCJIT, apple-m3)
  GC: Built with stock GC
Threads: 1 default, 1 interactive, 1 GC (on 4 virtual cores)

Models, default setups — cpu-arm

Model	T	L	Physics	Δt	SYPD	Memory
BarotropicModel	31	1	false	1800	46826	780.58 KB
ShallowWaterModel	31	1	false	2400	35514	962.86 KB
PrimitiveDryModel	31	8	true	2400	2222	5.27 MB
PrimitiveWetModel	31	8	true	2400	1435	6.22 MB

Shallow water model, resolution — cpu-arm

Model	T	L	Rings	Δt	SYPD	Memory
ShallowWaterModel	31	1	48	2400	18732	962.86 KB
ShallowWaterModel	42	1	64	1800	15114	1.68 MB
ShallowWaterModel	63	1	96	1200	3442	3.77 MB
ShallowWaterModel	85	1	128	900	1658	6.84 MB
ShallowWaterModel	127	1	192	600	392	16.12 MB
ShallowWaterModel	170	1	256	450	133	30.33 MB
ShallowWaterModel	255	1	384	300	28	76.02 MB

Primitive wet model, resolution — cpu-arm

Model	T	L	Rings	Transform	Δt	SYPD	Memory
PrimitiveWetModel	31	8	48	default	2400	1400	6.22 MB
PrimitiveWetModel	42	8	64	default	1800	564	10.51 MB
PrimitiveWetModel	63	8	96	default	1200	147	22.34 MB
PrimitiveWetModel	85	8	128	default	900	57	38.87 MB
PrimitiveWetModel	127	8	192	default	600	15	85.50 MB
PrimitiveWetModel	170	8	256	default	450	5.5	151.57 MB
PrimitiveWetModel	255	8	384	default	300	1.4	343.69 MB
PrimitiveWetModel	85	16	128	default	900	51	67.81 MB
PrimitiveWetModel	127	16	192	default	600	21	148.26 MB
PrimitiveWetModel	170	16	256	default	450	7.7	261.34 MB
PrimitiveWetModel	255	16	384	default	300	1.9	586.14 MB
PrimitiveWetModel	85	24	128	default	900	48	96.80 MB
PrimitiveWetModel	127	24	192	default	600	15	211.09 MB
PrimitiveWetModel	170	24	256	default	450	4.3	371.20 MB
PrimitiveWetModel	255	24	384	default	300	1.5	828.73 MB
PrimitiveWetModel	31	8	48	matrix	2400	757	48.11 MB
PrimitiveWetModel	42	8	64	matrix	1800	272	133.88 MB
PrimitiveWetModel	63	8	96	matrix	1200	48	582.80 MB
PrimitiveWetModel	85	8	128	matrix	900	13	1.75 GB
PrimitiveWetModel	127	8	192	matrix	600	1.5	8.19 GB
PrimitiveWetModel	85	16	128	matrix	900	19	1.78 GB
PrimitiveWetModel	127	16	192	matrix	600	2.1	8.24 GB
PrimitiveWetModel	85	24	128	matrix	900	11	1.80 GB
PrimitiveWetModel	127	24	192	matrix	600	1.7	8.30 GB

Primitive Equation, Float32 vs Float64 — cpu-arm

Model	NF	T	L	Δt	SYPD	Memory
PrimitiveWetModel	Float32	31	8	2400	1227	6.22 MB
PrimitiveWetModel	Float64	31	8	2400	1232	11.35 MB

Grids — cpu-arm

Model	T	L	Grid	Rings	Δt	SYPD	Memory
PrimitiveWetModel	63	8	FullGaussianGrid	96	1200	80	32.40 MB
PrimitiveWetModel	63	8	FullClenshawGrid	95	1200	109	32.13 MB
PrimitiveWetModel	63	8	OctahedralGaussianGrid	96	1200	158	22.34 MB
PrimitiveWetModel	63	8	OctahedralClenshawGrid	95	1200	146	22.06 MB
PrimitiveWetModel	63	8	HEALPixGrid	95	1200	219	16.42 MB
PrimitiveWetModel	63	8	OctaHEALPixGrid	95	1200	171	19.89 MB

Number of vertical layers — cpu-arm

Model	T	L	Δt	SYPD	Memory
PrimitiveWetModel	31	4	2400	2324	3.87 MB
PrimitiveWetModel	31	8	2400	1404	6.22 MB
PrimitiveWetModel	31	12	2400	1003	8.58 MB
PrimitiveWetModel	31	16	2400	771	10.94 MB

PrimitiveDryModel: Physics or dynamics only — cpu-arm

Model	T	L	Dynamics	Physics	Δt	SYPD	Memory
PrimitiveDryModel	31	8	true	true	2400	2368	5.27 MB
PrimitiveDryModel	31	8	true	false	2400	4046	5.27 MB
PrimitiveDryModel	31	8	false	true	2400	2614	5.27 MB

PrimitiveWetModel: Physics or dynamics only — cpu-arm

Model	T	L	Dynamics	Physics	Δt	SYPD	Memory
PrimitiveWetModel	31	8	true	true	2400	1423	6.22 MB
PrimitiveWetModel	31	8	true	false	2400	3055	6.22 MB
PrimitiveWetModel	31	8	false	true	2400	1558	6.22 MB

Individual dynamics functions — cpu-arm

PrimitiveWetModel | Float32 | T31 L8 | OctahedralGaussianGrid | 48 Rings — cpu-arm

Function	Time	Memory	Allocations
pressure_gradient_flux!	39.666 μs	31.98 KiB	200
linear_virtual_temperature!	2.056 μs	0 bytes	0
geopotential!	7.562 μs	384 bytes	6
vertical_integration!	14.500 μs	0 bytes	0
surface_pressure_tendency!	11.875 μs	15.66 KiB	96
vertical_velocity!	23.333 μs	0 bytes	0
linear_pressure_gradient!	2.065 μs	0 bytes	0
vertical_advection!	112.500 μs	2.44 KiB	32
vordiv_tendencies!	224.375 μs	231.83 KiB	284
temperature_tendency!	291.583 μs	344.77 KiB	401
humidity_tendency!	279.375 μs	344.09 KiB	396
bernoulli_potential!	93.667 μs	114.30 KiB	129

Architecture: cpu-x86

Created for SpeedyWeather.jl v0.21.1+DEV on Tue, 21 Jul 2026 16:59:17.

Machine details — cpu-x86

julia> versioninfo()
Julia Version 1.12.2
Commit ca9b6662be4 (2025-11-20 16:25 UTC)
Build Info:
  Official https://julialang.org release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 128 × AMD EPYC 9554 64-Core Processor
  WORD_SIZE: 64
  LLVM: libLLVM-18.1.7 (ORCJIT, znver4)
  GC: Built with stock GC
Threads: 1 default, 1 interactive, 1 GC (on 128 virtual cores)
Environment:
  LD_LIBRARY_PATH = /usr/local/lib:/usr/local/lib:

Models, default setups — cpu-x86

Model	T	L	Physics	Δt	SYPD	Memory
BarotropicModel	31	1	false	1800	10654	780.69 KB
ShallowWaterModel	31	1	false	2400	19394	962.97 KB
PrimitiveDryModel	31	8	true	2400	1467	5.27 MB
PrimitiveWetModel	31	8	true	2400	819	6.22 MB

Shallow water model, resolution — cpu-x86

Model	T	L	Rings	Δt	SYPD	Memory
ShallowWaterModel	31	1	48	2400	19308	962.97 KB
ShallowWaterModel	42	1	64	1800	8616	1.68 MB
ShallowWaterModel	63	1	96	1200	2297	3.77 MB
ShallowWaterModel	85	1	128	900	960	6.84 MB
ShallowWaterModel	127	1	192	600	231	16.12 MB
ShallowWaterModel	170	1	256	450	86	30.33 MB
ShallowWaterModel	255	1	384	300	18	76.02 MB

Primitive wet model, resolution — cpu-x86

Model	T	L	Rings	Transform	Δt	SYPD	Memory
PrimitiveWetModel	31	8	48	default	2400	856	6.22 MB
PrimitiveWetModel	42	8	64	default	1800	370	10.51 MB
PrimitiveWetModel	63	8	96	default	1200	107	22.34 MB
PrimitiveWetModel	85	8	128	default	900	39	38.87 MB
PrimitiveWetModel	127	8	192	default	600	11	85.50 MB
PrimitiveWetModel	170	8	256	default	450	3.9	151.57 MB
PrimitiveWetModel	255	8	384	default	300	1.0	343.69 MB
PrimitiveWetModel	85	16	128	default	900	49	67.81 MB
PrimitiveWetModel	127	16	192	default	600	15	148.26 MB
PrimitiveWetModel	170	16	256	default	450	5.6	261.34 MB
PrimitiveWetModel	255	16	384	default	300	1.3	586.14 MB
PrimitiveWetModel	85	24	128	default	900	38	96.80 MB
PrimitiveWetModel	127	24	192	default	600	11	211.09 MB
PrimitiveWetModel	170	24	256	default	450	4.5	371.20 MB
PrimitiveWetModel	255	24	384	default	300	1.0	828.73 MB
PrimitiveWetModel	31	8	48	matrix	2400	107	48.11 MB
PrimitiveWetModel	42	8	64	matrix	1800	28	133.88 MB
PrimitiveWetModel	63	8	96	matrix	1200	3.7	582.80 MB
PrimitiveWetModel	85	8	128	matrix	900	0.9	1.75 GB
PrimitiveWetModel	127	8	192	matrix	600	0.1	8.19 GB
PrimitiveWetModel	85	16	128	matrix	900	1.1	1.78 GB
PrimitiveWetModel	127	16	192	matrix	600	0.2	8.24 GB
PrimitiveWetModel	85	24	128	matrix	900	0.9	1.80 GB
PrimitiveWetModel	127	24	192	matrix	600	0.1	8.30 GB

Primitive Equation, Float32 vs Float64 — cpu-x86

Model	NF	T	L	Δt	SYPD	Memory
PrimitiveWetModel	Float32	31	8	2400	850	6.22 MB
PrimitiveWetModel	Float64	31	8	2400	767	11.35 MB

Grids — cpu-x86

Model	T	L	Grid	Rings	Δt	SYPD	Memory
PrimitiveWetModel	63	8	FullGaussianGrid	96	1200	71	32.40 MB
PrimitiveWetModel	63	8	FullClenshawGrid	95	1200	74	32.13 MB
PrimitiveWetModel	63	8	OctahedralGaussianGrid	96	1200	100	22.34 MB
PrimitiveWetModel	63	8	OctahedralClenshawGrid	95	1200	100	22.06 MB
PrimitiveWetModel	63	8	HEALPixGrid	95	1200	163	16.42 MB
PrimitiveWetModel	63	8	OctaHEALPixGrid	95	1200	119	19.89 MB

Number of vertical layers — cpu-x86

Model	T	L	Δt	SYPD	Memory
PrimitiveWetModel	31	4	2400	1412	3.87 MB
PrimitiveWetModel	31	8	2400	851	6.22 MB
PrimitiveWetModel	31	12	2400	603	8.58 MB
PrimitiveWetModel	31	16	2400	467	10.94 MB

PrimitiveDryModel: Physics or dynamics only — cpu-x86

Model	T	L	Dynamics	Physics	Δt	SYPD	Memory
PrimitiveDryModel	31	8	true	true	2400	1456	5.27 MB
PrimitiveDryModel	31	8	true	false	2400	2293	5.27 MB
PrimitiveDryModel	31	8	false	true	2400	1609	5.27 MB

PrimitiveWetModel: Physics or dynamics only — cpu-x86

Model	T	L	Dynamics	Physics	Δt	SYPD	Memory
PrimitiveWetModel	31	8	true	true	2400	852	6.22 MB
PrimitiveWetModel	31	8	true	false	2400	1773	6.22 MB
PrimitiveWetModel	31	8	false	true	2400	951	6.22 MB

Individual dynamics functions — cpu-x86

PrimitiveWetModel | Float32 | T31 L8 | OctahedralGaussianGrid | 48 Rings — cpu-x86

Function	Time	Memory	Allocations
pressure_gradient_flux!	70.633 μs	31.98 KiB	200
linear_virtual_temperature!	3.278 μs	0 bytes	0
geopotential!	10.260 μs	384 bytes	6
vertical_integration!	13.680 μs	0 bytes	0
surface_pressure_tendency!	18.721 μs	15.66 KiB	96
vertical_velocity!	59.542 μs	0 bytes	0
linear_pressure_gradient!	3.179 μs	0 bytes	0
vertical_advection!	160.256 μs	2.44 KiB	32
vordiv_tendencies!	412.104 μs	218.48 KiB	284
temperature_tendency!	532.669 μs	324.75 KiB	401
humidity_tendency!	507.748 μs	324.08 KiB	396
bernoulli_potential!	174.696 μs	107.62 KiB	129

Architecture: gpu-nvidia

Created for SpeedyWeather.jl v0.22.1+DEV on Fri, 28 Aug 2026 13:51:09.

Machine details — gpu-nvidia

julia> versioninfo()
Julia Version 1.12.2
Commit ca9b6662be4 (2025-11-20 16:25 UTC)
Build Info:
  Official https://julialang.org release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 128 × AMD EPYC 9554 64-Core Processor
  WORD_SIZE: 64
  LLVM: libLLVM-18.1.7 (ORCJIT, znver4)
  GC: Built with stock GC
Threads: 1 default, 1 interactive, 1 GC (on 128 virtual cores)
Environment:
  LD_LIBRARY_PATH = /usr/local/lib:/usr/local/lib:

julia> CUDA.versioninfo()
CUDA toolchain: 
- runtime 13.3.0, artifact installation
- driver 580.126.9 for 13.3
- compiler 13.3.33, artifact installation

CUDA libraries: 
- cuBLAS: 13.6.0
- cuSPARSE: 12.8.2
- cuSOLVER: 12.2.6
- cuFFT: 12.3.0
- cuRAND: 10.4.3
- CUPTI: 2026.2.1 (API 13.3.1)
- NVML: 13.0.0+580.126.9

Julia packages: 
- CUDACore: 6.2.2
- GPUArrays: 11.5.13
- GPUCompiler: 1.23.0
- KernelAbstractions: 0.9.42
- CUDA_Driver_jll: 13.3.1+0
- CUDA_Compiler_jll: 0.4.4+1
- CUDA_Runtime_jll: 0.23.0+1
- NVPTX_LLVM_Backend_jll: 22.1.7+1

Toolchain:
- Julia: 1.12.2
- LLVM: 18.1.7

1 device:
  0: NVIDIA H100 80GB HBM3 (sm_90, 64.265 GiB / 79.647 GiB available)
     compiles to sm_90a / PTX 9.3 (LLVM: sm_90a / PTX 9.0)

Models, default setups — gpu-nvidia

Model	truncation	L	Physics	Δt	SYPD	Memory
BarotropicModel	32	1	false	1800	7932	429.08 KB
ShallowWaterModel	32	1	false	2400	8926	432.58 KB
PrimitiveDryModel	32	8	true	2400	6909	576.66 KB
PrimitiveWetModel	32	8	true	2400	5888	581.80 KB

Shallow water model, resolution — gpu-nvidia

Model	truncation	L	Rings	Δt	SYPD	Memory
ShallowWaterModel	32	1	48	2400	8931	432.58 KB
ShallowWaterModel	43	1	64	1800	2948	745.16 KB
ShallowWaterModel	64	1	96	1200	942	1.63 MB
ShallowWaterModel	86	1	128	900	695	3.07 MB
ShallowWaterModel	128	1	192	600	292	6.70 MB
ShallowWaterModel	171	1	256	450	126	11.74 MB
ShallowWaterModel	256	1	384	300	50	26.06 MB

Primitive wet model, resolution — gpu-nvidia

Model	truncation	L	Rings	Transform	Δt	SYPD	Memory
PrimitiveWetModel	32	8	48	default	2400	5879	581.80 KB
PrimitiveWetModel	43	8	64	default	1800	3779	994.73 KB
PrimitiveWetModel	64	8	96	default	1200	1188	2.17 MB
PrimitiveWetModel	86	8	128	default	900	659	4.07 MB
PrimitiveWetModel	128	8	192	default	600	264	8.88 MB
PrimitiveWetModel	171	8	256	default	450	138	15.56 MB
PrimitiveWetModel	256	8	384	default	300	53	34.53 MB
PrimitiveWetModel	86	16	128	default	900	566	5.12 MB
PrimitiveWetModel	128	16	192	default	600	236	11.24 MB
PrimitiveWetModel	171	16	256	default	450	120	19.76 MB
PrimitiveWetModel	256	16	384	default	300	44	43.96 MB
PrimitiveWetModel	86	24	128	default	900	532	6.17 MB
PrimitiveWetModel	128	24	192	default	600	219	13.60 MB
PrimitiveWetModel	171	24	256	default	450	110	23.95 MB
PrimitiveWetModel	256	24	384	default	300	38	53.40 MB
PrimitiveWetModel	32	8	48	matrix	2400	5730	581.80 KB
PrimitiveWetModel	43	8	64	matrix	1800	3809	994.73 KB
PrimitiveWetModel	64	8	96	matrix	1200	1186	2.17 MB
PrimitiveWetModel	86	8	128	matrix	900	305	3.81 MB
PrimitiveWetModel	128	8	192	matrix	600	32	8.50 MB
PrimitiveWetModel	86	16	128	matrix	900	137	4.86 MB
PrimitiveWetModel	128	16	192	matrix	600	14	10.86 MB
PrimitiveWetModel	86	24	128	matrix	900	78	5.91 MB
PrimitiveWetModel	128	24	192	matrix	600	9.2	13.22 MB

Primitive Equation, Float32 vs Float64 — gpu-nvidia

Model	NF	truncation	L	Δt	SYPD	Memory
PrimitiveWetModel	Float32	32	8	2400	5781	581.80 KB
PrimitiveWetModel	Float64	32	8	2400	5091	582.38 KB

Grids — gpu-nvidia

Model	truncation	L	Grid	Rings	Δt	SYPD	Memory
PrimitiveWetModel	64	8	FullGaussianGrid	96	1200	766	2.26 MB
PrimitiveWetModel	64	8	FullClenshawGrid	127	1200	481	3.95 MB
PrimitiveWetModel	64	8	OctahedralGaussianGrid	96	1200	1191	2.17 MB
PrimitiveWetModel	64	8	OctahedralClenshawGrid	127	1200	766	3.78 MB
PrimitiveWetModel	64	8	HEALPixGrid	127	1200	1088	3.71 MB
PrimitiveWetModel	64	8	OctaHEALPixGrid	127	1200	872	3.76 MB

Number of vertical layers — gpu-nvidia

Model	truncation	L	Δt	SYPD	Memory
PrimitiveWetModel	32	4	2400	5931	508.07 KB
PrimitiveWetModel	32	8	2400	6154	581.80 KB
PrimitiveWetModel	32	12	2400	5900	655.52 KB
PrimitiveWetModel	32	16	2400	6330	729.25 KB

PrimitiveDryModel: Physics or dynamics only — gpu-nvidia

Model	truncation	L	Dynamics	Physics	Δt	SYPD	Memory
PrimitiveDryModel	32	8	true	true	2400	6860	576.66 KB
PrimitiveDryModel	32	8	true	false	2400	8572	576.66 KB
PrimitiveDryModel	32	8	false	true	2400	11334	576.66 KB

PrimitiveWetModel: Physics or dynamics only — gpu-nvidia

Model	truncation	L	Dynamics	Physics	Δt	SYPD	Memory
PrimitiveWetModel	32	8	true	true	2400	6153	581.80 KB
PrimitiveWetModel	32	8	true	false	2400	8112	581.80 KB
PrimitiveWetModel	32	8	false	true	2400	9668	581.80 KB

Individual dynamics functions — gpu-nvidia

PrimitiveWetModel | Float32 | T31 L8 | OctahedralGaussianGrid | 48 Rings — gpu-nvidia

Function	Time	Memory	Allocations
pressure_gradient_flux!	79.750 μs	14.42 KiB	348
linear_virtual_temperature!	12.640 μs	2.77 KiB	52
geopotential!	18.800 μs	3.73 KiB	101
vertical_integration!	27.230 μs	7.17 KiB	141
surface_pressure_tendency!	N/A	N/A	N/A
vertical_velocity!	24.330 μs	8.45 KiB	191
linear_pressure_gradient!	12.570 μs	2.34 KiB	50
vertical_advection!	24.320 μs	9.09 KiB	142
vordiv_tendencies!	N/A	N/A	N/A
temperature_tendency!	N/A	N/A	N/A
humidity_tendency!	N/A	N/A	N/A
bernoulli_potential!	N/A	N/A	N/A