Hardbootme test intermittently failing on systems with six GPUs

It is normal for GPU memory mapping to continue after the OS
login prompt is available.   Add delay before starting HTX to
ensure GPUs are available.

Resolves openbmc/openbmc-test-automation#1381

Change-Id: I7fadf646392c26161bc521beddec86c39a2a7e24
Signed-off-by: Steven Sombar <ssombar@us.ibm.com>
diff --git a/systest/gpu_stress_test.robot b/systest/gpu_stress_test.robot
index 76ed6b2..a384150 100755
--- a/systest/gpu_stress_test.robot
+++ b/systest/gpu_stress_test.robot
@@ -82,6 +82,7 @@
     Rpvars  loop_count
 
     REST Power On  stack_mode=skip
+    Run Key U  Sleep \ 15s
 
     # Collect data before the test starts.
     Collect NVIDIA Log File  start
@@ -155,6 +156,7 @@
     [Documentation]  Do the initial test setup.
 
     REST Power On  stack_mode=skip
+    Run Key U  Sleep \ 15s
     Delete All Error Logs
     Tool Exist  lspci
     Tool Exist  htxcmdline