do you need 500 slightly overlapping Zarr datasets?
src <- "/vsicurl/https://projects.pawsey.org.au/idea-gebco-tif/GEBCO_2024.tif"
src <- "https://projects.pawsey.org.au/idea-gebco-tif/GEBCO_2024.tif"
library(purrr) ## purrr CRAN
library(mirai) ## mirai CRAN
if (!file.exists(basename(src))) {
  curl::curl_download(src, basename(src))  ## curl CRAN
}
info <- vapour::vapour_raster_info(basename(src))  ## vapour CRAN
g <- grout::grout(info$dimension, info$extent, info$block * 6) ## devtols::install_github("hypertidy/grout")
idx <- grout::tile_index(g)
dir.create("zarr")
idx$lab <- sprintf("zarr/tile_%i_%i_%i.zarr", idx$tile, idx$tile_col, idx$tile_row)
mirai::daemons(0)
mirai::daemons(31)
fun <- in_parallel(function(.x) {
  b <- 1.0
  src <- "GEBCO_2024.tif"
  
  cmd <- "gdal raster clip --input %s --bbox %s --output-format ZARR --output %s --allow-bbox-outside-source"
  e <- unlist(.x[, c("xmin", "xmax", "ymin", "ymax")])
  bbox <- e[c(1, 3, 2, 4)] + b * c(-1, 1, -1, 1)
  if (file.exists(.x$lab)) return(NULL)
  system(sprintf(cmd, src, paste0(bbox, collapse = ","), .x$lab))
  NULL
})
walk(split(idx, 1:nrow(idx)), fun)
update to prevent out of bounds clipping, and use Zarr 3 and ZSTD
plot the overlap