diff --git a/ggml-sycl.cpp b/ggml-sycl.cpp
index 390652b3e..855015978 100644
--- a/ggml-sycl.cpp
+++ b/ggml-sycl.cpp
@@ -15258,23 +15258,20 @@ static void ggml_sycl_mul_mat(const ggml_tensor * src0, const ggml_tensor * src1
     bool              use_mul_mat_q =  ggml_sycl_supports_mmq(src0->type)
         && src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32;
 
-    // fp16 performance always better on gen12+
-    const bool fp16_performance_good = true;
-
     // mmvq and mmq need the __dp4a instruction which is available for gen12+
     // Workaround in https://github.com/ggerganov/llama.cpp/commit/95f84d5ce8b449a9b16009434aca800df504a02e
     use_mul_mat_q     = use_mul_mat_q && (src0->type != GGML_TYPE_IQ2_XXS);
 #ifdef SYCL_USE_XMX
-    use_mul_mat_q     = use_mul_mat_q     && (!fp16_performance_good || src1->ne[1] <= MMQ_MAX_BATCH_SIZE);
+    use_mul_mat_q     = use_mul_mat_q     && (src1->ne[1] <= MMQ_MAX_BATCH_SIZE);
 #endif // SYCL_USE_XMX
 
-    if (!split && !fp16_performance_good && src0->type == GGML_TYPE_F16 && ggml_is_permuted(src0) && ggml_is_permuted(src1) && src1->ne[1] == 1) {
+    if (!split && src0->type == GGML_TYPE_F16 && ggml_is_permuted(src0) && ggml_is_permuted(src1) && src1->ne[1] == 1) {
         // KQ single-batch
         ggml_sycl_mul_mat_vec_p021(src0, src1, dst);
-    } else if (!split && !fp16_performance_good && src0->type == GGML_TYPE_F16 && !ggml_is_contiguous(src0) && !ggml_is_transposed(src1) && src1->ne[1] == 1) {
+    } else if (!split && src0->type == GGML_TYPE_F16 && !ggml_is_contiguous(src0) && !ggml_is_transposed(src1) && src1->ne[1] == 1) {
         // KQV single-batch
         ggml_sycl_mul_mat_vec_nc(src0, src1, dst);
-    } else if (!split && src0->type == GGML_TYPE_F16 && (src1->type == GGML_TYPE_F16 || fp16_performance_good) && !ggml_is_transposed(src0) && !ggml_is_transposed(src1) && src1->ne[2]*src1->ne[3] > 1) {
+    } else if (!split && src0->type == GGML_TYPE_F16 && (src1->type == GGML_TYPE_F16) && !ggml_is_transposed(src0) && !ggml_is_transposed(src1) && src1->ne[2]*src1->ne[3] > 1) {
         // KQ + KQV multi-batch
         ggml_sycl_mul_mat_batched_sycl(src0, src1, dst);
     } else if (use_dequantize_mul_mat_vec) {