diff --git a/ggml.c b/ggml.c
index f8cd58df..2b4d3455 100644
--- a/ggml.c
+++ b/ggml.c
@@ -17283,10 +17283,18 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
         } else {
             // wait for other threads to finish
             const int last = node_n;
-            do {
-                //sched_yield();
+            while (true) {
+                // TODO: this sched_yield can have significant impact on the performance - either positive or negative
+                //       depending on the workload and the operating system.
+                //       since it is not clear what is the best approach, it should potentially become user-configurable
+                //       ref: https://github.com/ggerganov/ggml/issues/291
+#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
+                sched_yield();
+#endif
+
                 node_n = atomic_load(&state->shared->node_n);
-            } while (node_n == last);
+                if (node_n != last) break;
+            };
         }
 
         // check if we should stop
diff --git a/whisper.cpp b/whisper.cpp
index 13937ea8..50ebba74 100644
--- a/whisper.cpp
+++ b/whisper.cpp
@@ -5332,7 +5332,8 @@ WHISPER_API const char * whisper_bench_ggml_mul_mat_str(int n_threads) {
     // b: N*N*sizeof(float)
     // c: N*N*sizeof(float)
     // when F16 is used, there is an extra work buffer of size N*N*sizeof(float)
-    std::vector<char> buf(4llu*N_max*N_max*sizeof(float) + 4*512);
+    std::vector<uint8_t> buf(3llu*N_max*N_max*sizeof(float) + 3*ggml_tensor_overhead());
+    std::vector<uint8_t> work;
 
     // put a bunch of random data in the buffer
     for (size_t i = 0; i < buf.size(); i++) buf[i] = i;
@@ -5387,12 +5388,12 @@ WHISPER_API const char * whisper_bench_ggml_mul_mat_str(int n_threads) {
             double tsum = 0.0;
 
             // heat-up
-            ggml_graph_compute_with_ctx(ctx0, &gf, n_threads);
+            ggml_graph_compute_helper(work, &gf, n_threads);
 
             for (int i = 0; i < n_max; ++i) {
                 const int64_t t0 = ggml_time_us();
 
-                ggml_graph_compute_with_ctx(ctx0, &gf, n_threads);
+                ggml_graph_compute_helper(work, &gf, n_threads);
 
                 const int64_t t1 = ggml_time_us();