diff --git a/tests/playground/Makefile b/tests/playground/Makefile
index e53ba77a39..141a85dcc5 100644
--- a/tests/playground/Makefile
+++ b/tests/playground/Makefile
@@ -4,8 +4,8 @@ BENCH=ulimit -s unlimited && bench
 CROSS_BENCHES = binarytrees deriv expr_const_folding rbmap
 
 # basic version usable without Nix
-#CROSS_CATS = .lean .hs .gc.hs .llvm.hs .ml .gc.ml
-CROSS_CATS = .lean .gcc.lean .hs .gc.hs .llvm.hs .llvm.gc.hs .ml .gc.ml .flambda.ml
+#CROSS_CATS = .lean .perf.lean .hs .gc.hs .perf.hs .llvm.hs .ml .gc.ml .perf.ml
+CROSS_CATS = .lean .lean.perf .gcc.lean .hs .gc.hs .hs.perf .llvm.hs .ml .gc.ml .ml.perf .flambda.ml
 
 GHC_FLAGS = -O3
 OCAML_FLAGS = -O3
@@ -72,6 +72,9 @@ bench/%gc.hs.bench: %hs.out | bench
 bench/%gc.ml.bench: %gc.ml.out | bench
 	ulimit -s unlimited && OCAML_INSTR_FILE=$@ time -ao $@ -f '%e' ./$< $(BENCH_PARAMS)
 
+bench/%.perf.bench: %.out | bench
+	ulimit -s unlimited && perf stat -e cache-references,cache-misses,branches,branch-misses -o $@ ./$< $(BENCH_PARAMS)
+
 # fork() breaks instrumentation
 bench/binarytrees.gc.ml.bench:
 	touch $@
diff --git a/tests/playground/report.py b/tests/playground/report.py
index f00889f2cd..71774e9cce 100755
--- a/tests/playground/report.py
+++ b/tests/playground/report.py
@@ -3,6 +3,7 @@
 import ast
 import json
 import os
+import re
 import sys
 
 def read(bench, cat):
@@ -33,15 +34,24 @@ def pp_gc_ml(bench, cat, norm):
             gc_nanos += int(data[2]) - int(data[1])
     return f"{gc_nanos*1e-9/wall_secs:.0%}"
 
+def pp_perf(bench, cat, norm):
+    stat = open(f"bench/{bench}{cat}.bench", 'r').read()
+    #misses_percent = float(re.search("([0-9.]+) % of all cache refs", stat)[1]) / 100
+    misses = int(re.search("([0-9,]+)\s+cache-misses", stat)[1].replace(',', ''))
+    secs = float(re.search("([0-9.]+) seconds time elapsed", stat)[1])
+    return f"{misses*1e-6/secs:.3f}"
+
 CATBAG = {
     '.lean': ("Lean", pp_bench),
     '.gcc.lean': ("Lean+GCC9", pp_bench),
+    '.lean.perf': ("cache misses (1M/s)", pp_perf),
     '.hs': ("GHC", pp_bench),
     '.gc.hs': ("%GC", pp_gc_hs),
+    '.hs.perf': ("cache misses", pp_perf),
     '.llvm.hs': ("GHC -fllvm", pp_bench),
-    '.llvm.gc.hs': ("%GC", pp_gc_hs),
     '.ml': ("OCaml", pp_bench),
     '.gc.ml': ("%GC", pp_gc_ml),
+    '.ml.perf': ("cache misses", pp_perf),
     '.flambda.ml': ("OCaml+Flambda", pp_bench),
 }