diff --git a/bench/abg_minibench.py b/bench/abg_minibench.py
new file mode 100644
index 0000000..b109a67
--- /dev/null
+++ b/bench/abg_minibench.py
@@ -0,0 +1,37 @@
+import numpy as np
+from mpi4py import MPI
+import time
+
+comm_rank = MPI.COMM_WORLD.Get_rank()
+comm_size = MPI.COMM_WORLD.Get_size()
+
+
+if __name__ == "__main__":
+    # abg of all reduce
+    b_sizes = [256, 512, 1024]
+    n_iterations = 100
+
+    timings = np.zeros((len(b_sizes), n_iterations))
+
+    print(f"Bench Allreduce for P={comm_size}")
+    for b_i in b_sizes:
+        b_init = np.random.rand(b_i)
+        for i in range(n_iterations):
+            b = b_init.copy()
+
+            tic = time.perf_counter()
+            MPI.COMM_WORLD.Allreduce(
+                MPI.IN_PLACE,
+                b,
+                op=MPI.SUM,
+            )
+            toc = time.perf_counter()
+
+            timings[b_i, i] = toc - tic
+
+        if comm_rank == 0:
+            print(f"    b_size: {b_i}")
+            print(f"        mean: {np.mean(timings[b_i])}")
+            print(f"        std: {np.std(timings[b_i])}")
+
+    np.save(f"abg_allreduce_timings_{comm_size}.npy", timings)
diff --git a/bench/slurm_abg_minibench.sh b/bench/slurm_abg_minibench.sh
new file mode 100644
index 0000000..1f93df9
--- /dev/null
+++ b/bench/slurm_abg_minibench.sh
@@ -0,0 +1,10 @@
+#!/bin/bash
+
+#SBATCH --job-name=serinv_generate_dataset
+#SBATCH --nodes=1
+#SBATCH --gres=gpu:a40:2
+#SBATCH --time=00:05:00
+#SBATCH --output=%x.%j.out
+#SBATCH --error=%x.%j.err
+
+srun python abg_minibench.py
\ No newline at end of file