diff --git a/models/demos/falcon7b/tests/test_perf_falcon.py b/models/demos/falcon7b/tests/test_perf_falcon.py
index 6c776b423d8..7e036c3b239 100644
--- a/models/demos/falcon7b/tests/test_perf_falcon.py
+++ b/models/demos/falcon7b/tests/test_perf_falcon.py
@@ -588,10 +588,10 @@ def test_perf_wh_bare_metal(
     @pytest.mark.parametrize(
         "llm_mode, num_devices, num_layers, batch, seq_len, kv_cache_len, model_config_str, expected_output_pcc, expected_k_cache_pcc, expected_v_cache_pcc, expected_inference_time, async_mode",
         (
-            ("prefill", 4, 32, 1, 256, 0, "BFLOAT16-DRAM", 0.98, 0.99, 0.96, 0.225, False),  # Issue 7816 Inference time
-            ("decode", 4, 32, 32, 1, 1024, "BFLOAT16-L1_SHARDED", 0.87, 0.91, 0.91, 0.27, False),
+            ("prefill", 4, 32, 1, 256, 0, "BFLOAT16-DRAM", 0.98, 0.99, 0.96, 0.18, False),  # Issue 7816 Inference time
+            ("decode", 4, 32, 32, 1, 1024, "BFLOAT16-L1_SHARDED", 0.87, 0.91, 0.91, 0.21, False),
             ("prefill", 4, 32, 1, 256, 0, "BFLOAT16-DRAM", 0.98, 0.99, 0.96, 0.18, True),
-            ("decode", 4, 32, 32, 1, 1024, "BFLOAT16-L1_SHARDED", 0.87, 0.91, 0.91, 0.10, True),
+            ("decode", 4, 32, 32, 1, 1024, "BFLOAT16-L1_SHARDED", 0.87, 0.91, 0.91, 0.09, True),
         ),
         ids=[
             "prefill_seq256",