diff --git a/.jenkins/Jenkinsfile b/.jenkins/Jenkinsfile
index 44b40c4442..36a6c3945b 100644
--- a/.jenkins/Jenkinsfile
+++ b/.jenkins/Jenkinsfile
@@ -57,7 +57,7 @@ def hipBuildTest(String backendLabel) {
                     set -x
                     # Check if backend label contains string "amd" or backend host is a server with amd gpu
                     if [[ $backendLabel =~ amd ]]; then
-                        LLVM_PATH=/opt/rocm/llvm ctest
+                        LLVM_PATH=/opt/rocm/llvm ctest -E 'cooperative_streams_least_capacity.tst|cooperative_streams_half_capacity.tst|cooperative_streams_full_capacity.tst|grid_group_data_sharing.tst'
                     else
                         make test
                     fi
@@ -97,7 +97,7 @@ def hipBuildTest(String backendLabel) {
                     set -x
                     # Check if backend label contains string "amd" or backend host is a server with amd gpu
                     if [[ $backendLabel =~ amd ]]; then
-                        LLVM_PATH=/opt/rocm/llvm ctest -E 'Unit_hipGraphChildGraphNodeGetGraph_Functional'
+                        LLVM_PATH=/opt/rocm/llvm ctest -E 'Unit_hipGraphChildGraphNodeGetGraph_Functional|Unit_hipGraphExecMemcpyNodeSetParamsFromSymbol_Negative'
                     else
                         make test
                     fi
diff --git a/tests/src/runtimeApi/cooperativeGrps/cooperative_streams.cpp b/tests/src/runtimeApi/cooperativeGrps/cooperative_streams.cpp
index cc93e825ba..07a4f7711c 100644
--- a/tests/src/runtimeApi/cooperativeGrps/cooperative_streams.cpp
+++ b/tests/src/runtimeApi/cooperativeGrps/cooperative_streams.cpp
@@ -86,6 +86,7 @@ static inline void hipCheckAndFail(hipError_t errval,
   do { hipCheckAndFail((errval), __FILE__, __LINE__); } while (0)
 
 __global__ void test_kernel(uint32_t loops, unsigned long long *array, long long totalTicks) {
+  cooperative_groups::thread_block tb = cooperative_groups::this_thread_block();
   unsigned int rank = blockIdx.x * blockDim.x + threadIdx.x;
 
   for (int i = 0; i < loops; i++) {
@@ -100,6 +101,7 @@ __global__ void test_kernel(uint32_t loops, unsigned long long *array, long long
       // So just ignore those slipped cycles.
       last_clock = cur_clock;
     } while(time_diff < totalTicks);
+    tb.sync();
     array[rank] += clock64();
   }
 }