diff --git a/hack/calculate-cache.sh b/hack/calculate-cache.sh
new file mode 100755
index 000000000000..7f7d51e23197
--- /dev/null
+++ b/hack/calculate-cache.sh
@@ -0,0 +1,299 @@
+#!/usr/bin/env bash
+# This script calculates the expected content size, actual cached size, and cache-keys used in caching method before and after
+# implementation in https://github.com/lima-vm/lima/pull/2508
+#
+# Answer to the question in https://github.com/lima-vm/lima/pull/2508#discussion_r1699798651
+
+# usage: [DEBUG=1] ./hack/calculate-cache.sh
+# DEBUG=1 will save the collected information in .calculate-cache-collected-info-{before,after}.yaml
+#
+# This script does:
+# 1. extracts `runs_on` and `template` from workflow file (.github/workflows/test.yml)
+# 2. check each template for image and nerdctl
+# 3. detect size of image and nerdctl (responses from remote are cached for faster iteration)
+#    save the response in .calculate-cache-response-cache.yaml
+# 4. print content size, actual cache size (if available), by cache key
+#
+# The major differences for reducing cache usage are as follows:
+# - it is now cached `~/.cache/lima/download/by-url-sha256/$sha256` instead of caching `~/.cache/lima/download`
+# - the cache keys are now based on the image digest and nerdctl digest instead of the template file's hash
+# - enables the use of cache regardless of the operating system used to execute CI.
+#
+# The script requires the following commands:
+# - gh: GitHub CLI.
+#   Using to get the cache information
+# - jq: Command-line JSON processor
+#   Parse the workflow file and print runs-on and template.
+#   Parse output from gh cache list
+#   Calculate the expected content size, actual cached size, and cache-keys used.
+# - limactl: lima CLI.
+#   Using to validate the template file for getting nerdctl location and digest.
+# - sha256sum: Print or check SHA256 (256-bit) checksums
+# - xxd: make a hexdump or do the reverse.
+#   Using to simulate the 'hashFile()' function in the workflow.
+# - yq: Command-line YAML processor.
+#   Parse the template file for image and nerdctl location, digest, and size.
+#   Parse the cache response file for the cache.
+#   Convert the collected information to JSON.
+
+set -u -o pipefail
+
+required_commands=(gh jq limactl sha256sum xxd yq)
+for cmd in "${required_commands[@]}"; do
+	if ! command -v "${cmd}" &>/dev/null; then
+		echo "${cmd} is required. Please install it" >&2
+		exit 1
+	fi
+done
+
+# current workflow uses x86_64 only
+arch=x86_64
+
+LIMA_HOME=$(mktemp -d)
+export LIMA_HOME
+
+# parse the workflow file and print runs-on and template
+# e.g.
+# ```console
+# $ print_runs_on_template_from_workflow .github/workflows/test.yml
+# macos-12        templates/default.yaml
+# ubuntu-24.04    templates/alpine.yaml
+# ubuntu-24.04    templates/debian.yaml
+# ubuntu-24.04    templates/fedora.yaml
+# ubuntu-24.04    templates/archlinux.yaml
+# ubuntu-24.04    templates/opensuse.yaml
+# ubuntu-24.04    templates/experimental/net-user-v2.yaml
+# ubuntu-24.04    templates/experimental/9p.yaml
+# ubuntu-24.04    templates/docker.yaml
+# ubuntu-24.04    templates/../hack/test-templates/alpine-9p-writable.yaml
+# ubuntu-24.04    templates/../hack/test-templates/test-misc.yaml
+# macos-12        templates/vmnet.yaml
+# macos-12        https://raw.githubusercontent.com/lima-vm/lima/v0.15.1/examples/ubuntu-lts.yaml
+# macos-13        templates/experimental/vz.yaml
+# macos-13        templates/fedora.yaml
+# ```
+function print_runs_on_template_from_workflow() {
+	yq -o=j "$1" | jq -r '
+		"./.github/actions/setup_cache_for_template" as $action |
+		"\\$\\{\\{\\s*(?<path>\\S*)\\s*\\}\\}" as $pattern |
+		.jobs | map_values(
+			."runs-on" as $runs_on |
+			{
+				template: .steps | map_values(select(.uses == $action)) | first |.with.template,
+				matrix: .strategy.matrix
+			} | select(.template) |
+			. + { path: .template | (if test($pattern) then sub(".*\($pattern).*";"\(.path)")|split(".") else null end) } |
+			(
+				.template as $template|
+				if .path then
+					getpath(.path)|map(. as $item|$template|sub($pattern;$item))
+				else
+					[$template]
+				end
+			) | map("\($runs_on)\t\(.)")
+
+		) | flatten |.[]
+	'
+}
+
+# returns the OS name from the runner equivalent to the expression `${{ runner.os }}` in the workflow
+# e.g.
+# ```console
+# $ runner_os_from_runner "macos-12"
+# macOS
+# $ runner_os_from_runner "ubuntu-24.04"
+# Linux
+# ```
+function runner_os_from_runner() {
+	# shellcheck disable=SC2249
+	case "$1" in
+	macos*)
+		echo macOS
+		;;
+	ubuntu*)
+		echo Linux
+		;;
+	esac
+}
+
+# check the remote location and return the http code and size.
+# The result is cached in .calculate-cache-response-cache.yaml
+# e.g.
+# ```console
+# $ check_location "https://cloud-images.ubuntu.com/releases/24.04/release-20240725/ubuntu-24.04-server-cloudimg-amd64.img"
+# 200 585498624
+# ```
+function check_location() {
+	location="$1"
+	readonly cache_file="./.calculate-cache-response-cache.yaml"
+	# check response_cache.yaml for the cache
+	if [[ -f ${cache_file} ]]; then
+		cached=$(yq -e eval ".[\"${location}\"]" "${cache_file}" 2>/dev/null) && echo "${cached}" && return
+	else
+		touch "${cache_file}"
+	fi
+	http_code_and_size=$(curl -sIL -w "%{http_code} %header{Content-Length}" "${location}" -o /dev/null)
+	yq eval ".[\"${location}\"] = \"${http_code_and_size}\"" -i "${cache_file}"
+	echo "${http_code_and_size}"
+}
+
+# print image location, digest, size, hash, containerd, containerd_location, containerd_digest, containerd_size from the template
+# e.g.
+# ```console
+# $ print_location_digest_size_hash_from_template "templates/default.yaml"
+# https://cloud-images.ubuntu.com/releases/24.04/release-20240725/ubuntu-24.04-server-cloudimg-amd64.img sha256:d2377667ea95222330ca2287817403c85178dad397e9fed768a9b4aec79d2a7f 585498624 49aa50a4872ded07ebf657c0eaf9e44ecc0c174d033a97c537ecd270f35b462f true https://github.com/containerd/nerdctl/releases/download/v1.7.6/nerdctl-full-1.7.6-linux-amd64.tar.gz sha256:2c841e097fcfb5a1760bd354b3778cb695b44cd01f9f271c17507dc4a0b25606 237465717
+# ```
+function print_location_digest_size_hash_from_template() {
+	readonly template=$1
+	case "${template}" in
+	http*)
+		template_yaml=$(curl -sSL "${template}")
+		;;
+	*)
+		template_yaml=$(<"${template}")
+		;;
+	esac
+	readonly yq_filter="
+		[
+			.images | map(select(.arch == \"${arch}\")) | [.[0,1].location, .[0,1].digest],
+			.containerd|[.system or .user],
+			.containerd.archives | map(select(.arch == \"${arch}\")) | [.[0].location, .[0].digest]
+		]|flatten|.[]
+	"
+	if command -v limactl &>/dev/null; then
+		parsed=$(limactl validate <(echo "${template_yaml}") --fill 2>/dev/null | yq eval "${yq_filter}")
+	else
+		parsed=$(yq eval "${yq_filter}" <<<"${template_yaml}")
+	fi
+	# macOS earlier than 15.0 uses bash 3.2.57, which does not support readarray -t
+	# readarray -t arr <<<"${parsed}"
+	while IFS= read -r line; do arr+=("${line}"); done <<<"${parsed}"
+	readonly locations=("${arr[@]:0:2}") digests=("${arr[@]:2:2}")
+	readonly containerd="${arr[4]}" containerd_location="${arr[5]}" containerd_digest="${arr[6]}"
+	declare location digest size hash
+	for ((i = 0; i < ${#locations[@]}; i++)); do
+		[[ ${locations[i]} != null ]] || continue
+		http_code_and_size=$(check_location "${locations[i]}")
+		read -r http_code size <<<"${http_code_and_size}"
+		if [[ ${http_code} -eq 200 ]]; then
+			location=${locations[i]}
+			digest=${digests[i]}
+			break
+		fi
+	done
+	if [[ -z ${location} ]]; then
+		echo "Failed to get the image location for ${template}" >&2
+		return 1
+	fi
+	hash=$(sha256sum <<<"${template_yaml}" | cut -d' ' -f1 | xxd -r -p | sha256sum | cut -d' ' -f1)
+	declare containerd_size
+	containerd_http_code_and_size=$(check_location "${containerd_location}")
+	read -r _containerd_http_code containerd_size <<<"${containerd_http_code_and_size}"
+	echo "${location} ${digest} ${size} ${hash} ${containerd} ${containerd_location} ${containerd_digest} ${containerd_size}"
+}
+
+# format first column to MiB
+# e.g.
+# ```console
+# $ echo 585498624 | size_to_mib
+#   558.38 MiB
+# ```
+function size_to_mib() {
+	awk '
+		function mib(size) { return sprintf("%7.2f MiB", size / 1024 / 1024) }
+		int($1)>0{ $1=" "mib($1) }
+		int($2)>0{ $2=mib($2) }
+		int($2)==0 && NF>1{ $2="<<missing>>" }
+		{ print }
+	'
+}
+
+# actual_cache_sizes=$(gh cache list --json key,createdAt,sizeInBytes|jq '[.[]|{"key":.key,"value":.sizeInBytes}]|from_entries')
+# e.g.
+# ```console
+# $ echo "${actual_cache_sizes}"
+# {
+#   "Linux-1c3b2791d52735d916dc44767c745c2319eb7cae74af71bbf45ddb268f42fc1d": 810758533,
+#   "Linux-231c66957fc2cdb18ea10e63f60770049026e29051ecd6598fc390b60d6a4fa6": 633036717,
+#   "Linux-3b906d46fa532e3bc348c35fc8e7ede6c69f0b27032046ee2cbb56d4022d1146": 574242367,
+#   "Linux-69a547b760dbf1650007ed541408474237bc611704077214adcac292de556444": 70310855,
+#   "Linux-7782f8b4ff8cd378377eb79f8d61c9559b94bbd0c11d19eb380ee7bda19af04e": 494141177,
+#   "Linux-8812aedfe81b4456d421645928b493b1f2f88aff04b7f3171207492fd44cd189": 812730766,
+#   "Linux-caa7d8af214d55ad8902e82d5918e61573f3d6795d2b5ad9a35305e26fa0e6a9": 754723892,
+#   "Linux-colima-v0.6.5": 226350335,
+#   "Linux-de83bce0608d787e3c68c7a31c5fab2b6d054320fd7bf633a031845e2ee03414": 810691197,
+#   "Linux-eb88a19dfcf2fb98278e7c7e941c143737c6d7cd8950a88f58e04b4ee7cef1bc": 570625794,
+#   "Linux-f88f0b3b678ff6432386a42bdd27661133c84a36ad29f393da407c871b0143eb": 68490954,
+#   "golangci-lint.cache-Linux-2850-74615231540133417fd618c72e37be92c5d3b3ad": 2434144,
+#   "macOS-231c66957fc2cdb18ea10e63f60770049026e29051ecd6598fc390b60d6a4fa6": 633020464,
+#   "macOS-49aa50a4872ded07ebf657c0eaf9e44ecc0c174d033a97c537ecd270f35b462f": 813179462,
+#   "macOS-8f37f663956af5f743f0f99ab973729b6a02f200ebfac7a3a036eff296550732": 810756770,
+#   "macOS-ef5509b5d4495c8c3590442ee912ad1c9a33f872dc4a29421c524fc1e2103b59": 813179476,
+#   "macOS-upgrade-v0.15.1": 1157814690,
+#   "setup-go-Linux-ubuntu20-go-1.23.0-02756877dbcc9669bb904e42e894c63aa9801138db94426a90a2d554f2705c52": 1015518352,
+#   "setup-go-Linux-ubuntu20-go-1.23.0-6bce2eefc6111ace836de8bb322432c072805737d5f3c5a3d47d2207a05f50df": 936433302,
+#   "setup-go-Linux-ubuntu24-go-1.22.6-02756877dbcc9669bb904e42e894c63aa9801138db94426a90a2d554f2705c52": 1090001859,
+#   "setup-go-Linux-ubuntu24-go-1.23.0-02756877dbcc9669bb904e42e894c63aa9801138db94426a90a2d554f2705c52": 526146768,
+#   "setup-go-Windows-go-1.23.0-02756877dbcc9669bb904e42e894c63aa9801138db94426a90a2d554f2705c52": 1155374040,
+#   "setup-go-Windows-go-1.23.0-6bce2eefc6111ace836de8bb322432c072805737d5f3c5a3d47d2207a05f50df": 1056433137,
+#   "setup-go-macOS-go-1.23.0-02756877dbcc9669bb904e42e894c63aa9801138db94426a90a2d554f2705c52": 1060919942,
+#   "setup-go-macOS-go-1.23.0-6bce2eefc6111ace836de8bb322432c072805737d5f3c5a3d47d2207a05f50df": 982139209
+# }
+actual_cache_sizes=$(
+	gh cache list --json key,createdAt,sizeInBytes \
+		--jq 'sort_by(.createdAt)|reverse|unique_by(.key)|sort_by(.key)|map({"key":.key,"value":.sizeInBytes})|from_entries'
+)
+
+workflows=(
+	.github/workflows/test.yml
+)
+
+# shellcheck disable=SC2016
+echo "=> compare expected content size, actual cached size, and cache-keys used before and after the change in https://github.com/lima-vm/lima/pull/2508"
+# iterate over before and after
+for cache_method in before after; do
+	echo "==> ${cache_method}"
+	echo "content-size actual-size cache-key"
+	output_yaml=$(
+		for workflow in "${workflows[@]}"; do
+			print_runs_on_template_from_workflow "${workflow}"
+		done | while IFS=$'\t' read -r runner template; do
+			runner_os=$(runner_os_from_runner "${runner}")
+			location_digest_size_hash=$(print_location_digest_size_hash_from_template "${template}") || continue
+			read -r location digest size hash containerd containerd_location containerd_digest containerd_size <<<"${location_digest_size_hash}"
+			if [[ ${cache_method} != after ]]; then
+				key=${runner_os}-${hash}
+			elif [[ ${digest} == null ]]; then
+				key=image:$(basename "${location}")-url-sha256:$(echo -n "${location}" | sha256sum | cut -d' ' -f1)
+			else
+				key=image:$(basename "${location}")-${digest}
+			fi
+			if [[ ${containerd} == true ]]; then
+				if [[ ${cache_method} != after ]]; then
+					# previous caching method packages the containerd archive with the image
+					size=$((size + containerd_size))
+				else
+					# new caching method packages the containerd archive separately
+					containerd_key=containerd:$(basename "${containerd_location}")-${containerd_digest}
+					printf -- "- key: %s\n  template: %s\n  location: %s\n  digest: %s\n  size: %s\n" \
+						"${containerd_key}" "${template}" "${containerd_location}" "${containerd_digest}" "${containerd_size}"
+				fi
+			fi
+			printf -- "- key: %s\n  template: %s\n  location: %s\n  digest: %s\n  size: %s\n" \
+				"${key}" "${template}" "${location}" "${digest}" "${size}"
+		done
+	)
+	output_json=$(yq -o=j . <<<"${output_yaml}")
+
+	# print size key
+	jq --argjson actual_size "${actual_cache_sizes}" -r 'unique_by(.key)|sort_by(.key)|.[]|[.size, $actual_size[.key] // 0, .key]|@tsv' <<<"${output_json}" | size_to_mib
+	# total
+	echo "------------"
+	jq '[unique_by(.key)|.[]|.size]|add' <<<"${output_json}" | size_to_mib
+	# save the collected information as yaml if DEBUG is set
+	if [[ -n ${DEBUG:+1} ]]; then
+		cat <<<"${output_yaml}" >".calculate-cache-collected-info-${cache_method}.yaml"
+		echo "Saved the collected information in .calculate-cache-collected-info-${cache_method}.yaml"
+	fi
+	echo ""
+done