diff --git a/cscs-checks/microbenchmarks/alloc_speed/alloc_speed.py b/cscs-checks/microbenchmarks/alloc_speed/alloc_speed.py
new file mode 100644
index 0000000000..768aabf46a
--- /dev/null
+++ b/cscs-checks/microbenchmarks/alloc_speed/alloc_speed.py
@@ -0,0 +1,65 @@
+import reframe as rfm
+import reframe.utility.sanity as sn
+
+
+@rfm.parameterized_test(['no'], ['2M'])
+class AllocSpeedTest(rfm.RegressionTest):
+    def __init__(self, hugepages):
+        super().__init__()
+
+        self.descr = 'Time to allocate 4096 MB using %s hugepages' % hugepages
+        self.sourcepath = 'alloc_speed.cpp'
+        self.build_system = 'SingleSource'
+        self.build_system.cxxflags = ['-O3', '-std=c++11']
+        self.valid_systems = ['daint:gpu', 'daint:mc', 'dom:gpu', 'dom:mc']
+        self.valid_prog_environs = ['PrgEnv-gnu']
+        if hugepages == 'no':
+            self.valid_systems += ['kesch:cn', 'kesch:pn']
+        else:
+            self.modules = ['craype-hugepages%s' % hugepages]
+
+        self.sanity_patterns = sn.assert_found('4096 MB', self.stdout)
+        self.perf_patterns = {
+            'time': sn.extractsingle(r'4096 MB, allocation time (?P<time>\S+)',
+                                     self.stdout, 'time', float)
+        }
+        self.sys_reference = {
+            'no': {
+                'dom:gpu': {
+                    'time': (1.80, None, 0.10, 's')
+                },
+                'dom:mc': {
+                    'time': (2.00, None, 0.10, 's')
+                },
+                'daint:gpu': {
+                    'time': (1.80, None, 0.10, 's')
+                },
+                'daint:mc': {
+                    'time': (2.00, None, 0.10, 's')
+                },
+                'kesch:cn': {
+                    'time': (1.25, None, 0.10, 's')
+                },
+                'kesch:pn': {
+                    'time': (0.55, None, 0.10, 's')
+                }
+            },
+            '2M': {
+                'dom:gpu': {
+                    'time': (0.11, None, 0.10, 's')
+                },
+                'dom:mc': {
+                    'time': (0.20, None, 0.10, 's')
+                },
+                'daint:gpu': {
+                    'time': (0.11, None, 0.10, 's')
+                },
+                'daint:mc': {
+                    'time': (0.20, None, 0.10, 's')
+                },
+            },
+        }
+        self.reference = self.sys_reference[hugepages]
+
+        self.maintainers = ['AK', 'VH']
+        self.tags = {'production'}
diff --git a/cscs-checks/microbenchmarks/alloc_speed/src/alloc_speed.cpp b/cscs-checks/microbenchmarks/alloc_speed/src/alloc_speed.cpp
new file mode 100644
index 0000000000..3627013633
--- /dev/null
+++ b/cscs-checks/microbenchmarks/alloc_speed/src/alloc_speed.cpp
@@ -0,0 +1,30 @@
+#include <chrono>
+#include <iostream>
+#include <algorithm>
+#include <memory>
+
+double test_alloc(size_t n)
+{
+    auto t_start = std::chrono::system_clock::now();
+    std::unique_ptr<char[]> ptr(new char[n]);
+
+    /* time to allocate + fill */
+    std::fill(ptr.get(), ptr.get() + n, 0);
+    auto t_alloc_fill = std::chrono::system_clock::now();
+
+    /* time too fill */
+    std::fill(ptr.get(), ptr.get() + n, 0);
+    auto t_alloc_two_fills = std::chrono::system_clock::now();
+
+    return std::chrono::duration_cast<std::chrono::duration<double>>((t_alloc_fill - t_start) - (t_alloc_two_fills - t_alloc_fill)).count();
+}
+
+int main(int argc, char** argv)
+{
+    for (size_t i = 20; i < 33; ++i)
+    {
+        std::cout << (1L << i) / static_cast<double>(1024.0*1024.0) << " MB, allocation time " << test_alloc(1L << i) << " sec.\n";
+    }
+
+    return 0;
+}