diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
index 4dac86b6..b24fc809 100644
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -4,7 +4,7 @@
 
 # Actions to run on pull requests
 
-name: Camkes VM Examples
+name: Camkes-VM-Examples-HW
 
 on:
   pull_request_target:
@@ -14,6 +14,19 @@ on:
 permissions:
   contents: read
 
+# To reduce the load (especiually on the machine queue) we cancel any older runs
+# of this workflow for the current PR. Such runs exist, if there were new pushes
+# to the PR's branch without waiting for the workflow to finish. As a side
+# effect, pushing new commits now becomes a convenient way to cancel all the
+# older runs, e.g. if they are stuck and would only be stopped by the timeout
+# eventually.
+# Note that we could do the concurrency handling at a finer level, and only wrap
+# the actual run on the hardware. But there seems not much gain in letting the
+# older builds run, as these are usually obsolete with new pushes also.
+concurrency:
+  group: ${{ github.workflow }}-pr-${{ github.event.number }}
+  cancel-in-progress: true
+
 jobs:
   code:
     name: Freeze Code