diff --git a/src/gpu/jit/conv/kernel_builder.cpp b/src/gpu/jit/conv/kernel_builder.cpp
index 0dc4f19a016..26903482c11 100644
--- a/src/gpu/jit/conv/kernel_builder.cpp
+++ b/src/gpu/jit/conv/kernel_builder.cpp
@@ -6137,7 +6137,6 @@ class load_multiply_builder_t {
             // TODO: for now, only b-blocking (per G) of the MAD loop is ready;
             //       please implement n-blocking (per OC) as well!
             ir_assert(a_view.tlayout().size() % a_stride == 0);
-            ir_assert(cfg_.ic == 1);
             ir_assert(masks.is_simd());
 
             std::vector<stmt_t> loop(std::max(1, 32 / m_blk));