diff --git a/examples/master/repository/speed_benchmark.py b/examples/master/repository/speed_benchmark.py
new file mode 100644
index 000000000..6c188f544
--- /dev/null
+++ b/examples/master/repository/speed_benchmark.py
@@ -0,0 +1,142 @@
+import time
+
+from artiq import *
+
+
+class _PayloadNOP(EnvExperiment):
+    def build(self):
+        pass
+
+    def run(self):
+        pass
+
+
+class _PayloadCoreNOP(EnvExperiment):
+    def build(self):
+        self.attr_device("core")
+
+    @kernel
+    def run(self):
+        pass
+
+
+class _PayloadCoreSend100Ints(EnvExperiment):
+    def build(self):
+        self.attr_device("core")
+
+    def devnull(self, d):
+        pass
+
+    @kernel
+    def run(self):
+        for i in range(100):
+            self.devnull(42)
+
+
+class _PayloadCoreSend1MB(EnvExperiment):
+    def build(self):
+        self.attr_device("core")
+
+    def devnull(self, d):
+        pass
+
+    @kernel
+    def run(self):
+        # FIXME: make this a single list
+        # affected by issue #82 at the moment
+        data = [0 for _ in range(20000//4)]
+        for i in range(50):
+            self.devnull(data)
+
+
+class _PayloadCorePrimes(EnvExperiment):
+    def build(self):
+        self.attr_device("core")
+
+    def devnull(self, d):
+        pass
+
+    @kernel
+    def run(self):
+        for x in range(1, 1000):
+            d = 2
+            prime = True
+            while d*d <= x:
+                if x % d == 0:
+                    prime = False
+                    break
+                d += 1
+            if prime:
+                self.devnull(x)
+
+
+class SpeedBenchmark(EnvExperiment):
+    """Speed benchmark"""
+    def build(self):
+        self.attr_argument("mode", EnumerationValue(["Single experiment",
+                                              "With pause",
+                                              "With scheduler"]))
+        self.attr_argument("payload", EnumerationValue(["NOP",
+                                                        "CoreNOP",
+                                                        "CoreSend100Ints",
+                                                        "CoreSend1MB",
+                                                        "CorePrimes"]))
+        self.attr_argument("nruns", NumberValue(10, min=1, max=1000))
+        self.attr_device("core")
+        self.attr_device("scheduler")
+
+    def run_with_scheduler(self):
+        nruns = int(self.nruns)
+
+        donop_expid = dict(self.scheduler.expid)
+        donop_expid["class_name"] = "_Payload" + self.payload
+        donop_expid["arguments"] = {}
+        for i in range(nruns):
+            self.scheduler.submit(self.scheduler.pipeline_name, donop_expid,
+                                  self.scheduler.priority, None, False)
+
+        report_expid = dict(self.scheduler.expid)
+        report_expid["class_name"] = "_Report"
+        report_expid["arguments"] = {
+            "start_time": time.monotonic(),
+            "nruns": nruns}
+        self.scheduler.submit(self.scheduler.pipeline_name, report_expid,
+                              self.scheduler.priority, None, False)
+
+    def run_without_scheduler(self, pause):
+        payload = globals()["_Payload" + self.payload](*self.dbs())
+
+        start_time = time.monotonic()
+        for i in range(int(self.nruns)):
+            payload.run()
+            if pause:
+                self.core.comm.close()
+                self.scheduler.pause()
+        end_time = time.monotonic()
+
+        self.set_result("benchmark_run_time",
+                        (end_time-start_time)/self.nruns,
+                        realtime=True)
+
+
+    def run(self):
+        if self.mode == "Single experiment":
+            self.run_without_scheduler(False)
+        elif self.mode == "With pause":
+            self.run_without_scheduler(True)
+        elif self.mode == "With scheduler":
+            self.run_with_scheduler()
+        else:
+            raise ValueError
+
+
+class _Report(EnvExperiment):
+    def build(self):
+        self.attr_argument("start_time")
+        self.attr_argument("nruns")
+
+    def run(self):
+        end_time = time.monotonic()
+        self.set_result("benchmark_run_time",
+                        (end_time-self.start_time)/self.nruns,
+                        realtime=True)