diff --git a/ggml/src/ggml-metal.m b/ggml/src/ggml-metal.m
index f314ad20a..aee354cdd 100644
--- a/ggml/src/ggml-metal.m
+++ b/ggml/src/ggml-metal.m
@@ -3145,7 +3145,7 @@ static void ggml_metal_encode_node(
                     GGML_ASSERT(nqptg  % 8  == 0);
                     GGML_ASSERT(ncpsg  % 32 == 0);
 
-                    // 16*32*nsgmax
+                    // 16*32*(nsg)
                     // the shared memory needed for the simdgroups to load the KV cache
                     // each thread loads (dequantizes) 16 head elements, there are 32 threads in th SG
                     //