Oceananigans bump CI

[Pangoraw]

No description provided.

[wsmoses]

tpu flops are decreased to 0.00011567750560185194 from 0.0008782576127735523, so this is definitely not mergeable atm

[wsmoses]

@Pangoraw are the sizes different?

from gpu run:

┌ Info: [0] allocations
│ GordonBell25.allocatorstats() =
│ AllocatorStats
│ --------------
│ Num Allocs: 72
│ In Use: 11.006 MiB
│ Peak In Use: 26.153 MiB
│ Largest Alloc Size: 15.146 MiB
│ Limit: 29.620 GiB
│ Reserved: 0 bytes
│ Peak Reserved: 0 bytes
│ Reservable Limit: nothing
│ Largest Free Block: 0 bytes
│ Pool: 29.620 GiB
│ Peak Pool: 29.620 GiB
└

this is way too small

[Pangoraw]

yeah default size of simulation was lowered in https://github.com/PRONTOLab/GB-25/pull/258/changes#diff-dcc6b106f84fbd7b01b06d52dd3148ff3706665ece587c0666a90f5a749d47e6. Let's try with previous size

[wsmoses]

infinite compile time seemingly?

[Pangoraw]

TPU:
[0] first time step: 5.692765 seconds (10.16 M allocations: 525.449 MiB, 1.24% gc time, 98.52% compilation time)
[0] loop: 12.117740 seconds (9.92 M allocations: 513.606 MiB, 0.79% gc time, 45.86% compilation time)
[0] second loop: 6.550102 seconds (210 allocations: 7.375 KiB)
    metrics = Metrics(
        flops = 0.000802745539381693,  # [flops / (peak flops * program time)], capped at 1.0


GPU:
[0] first time step: 12.030048 seconds (10.16 M allocations: 525.487 MiB, 0.71% gc time, 98.59% compilation time)
[0] loop: 24.587343 seconds (9.92 M allocations: 513.605 MiB, 0.34% gc time, 48.20% compilation time)
[0] second loop: 12.568419 seconds (210 allocations: 7.375 KiB)
    metrics = Metrics(
        flops = 1.0,  # [flops / (peak flops * program time)], capped at 1.0