diff --git a/ggml/src/ggml-metal/ggml-metal-impl.h b/ggml/src/ggml-metal/ggml-metal-impl.h
index 8e8428d381..ca5a00b032 100644
--- a/ggml/src/ggml-metal/ggml-metal-impl.h
+++ b/ggml/src/ggml-metal/ggml-metal-impl.h
@@ -3,7 +3,7 @@
 
 // kernel parameters for mat-vec threadgroups
 //
-// N_R0: number of src0 rows to process per threadgroup
+// N_R0: number of src0 rows to process per simdgroup
 // N_SG: number of simdgroups per threadgroup
 //
 // TODO: for optimal performance, become function of the device and work size
diff --git a/ggml/src/ggml-metal/ggml-metal.m b/ggml/src/ggml-metal/ggml-metal.m
index 49b6314395..846f822ea1 100644
--- a/ggml/src/ggml-metal/ggml-metal.m
+++ b/ggml/src/ggml-metal/ggml-metal.m
@@ -2883,7 +2883,7 @@ static void ggml_metal_encode_node(
                     id<MTLComputePipelineState> pipeline = nil;
 
                     int nsg = 0; // number of simdgroups
-                    int nr0 = 0; // number of src0 rows per threadgroup
+                    int nr0 = 0; // number of src0 rows per simdgroup
                     int nr1 = 1; // number of src1 rows per threadgroup
 
                     size_t smem = 0; // shared memory