diff --git a/CHANGELOG.md b/CHANGELOG.md
index 6e13a4b5c..49c6c1dbe 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,6 +7,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+- fix: respect GGML_METAL=OFF on macOS by @provoke210 in #2370
 - feat: update llama.cpp to ggml-org/llama.cpp@v0.4.0
 
 ## [0.3.35]
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 89c12958c..773b56819 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -125,8 +125,8 @@ if (LLAMA_BUILD)
             set(GGML_F16C "OFF" CACHE BOOL "ggml: enable F16C" FORCE)
         endif()
 
-        # Metal settings (enable for both architectures)
-        set(GGML_METAL "ON" CACHE BOOL "ggml: enable Metal" FORCE)
+        # Enable Metal by default for both architectures, respecting user settings.
+        set(GGML_METAL "ON" CACHE BOOL "ggml: enable Metal")
         set(GGML_METAL_EMBED_LIBRARY "ON" CACHE BOOL "ggml: embed metal library" FORCE)
     endif()