diff --git a/bench/standard/aggregates/many_int_groups_1M.bench b/bench/standard/aggregates/many_int_groups_1M.bench
new file mode 100644
index 000000000..1296054f8
--- /dev/null
+++ b/bench/standard/aggregates/many_int_groups_1M.bench
@@ -0,0 +1,8 @@
+# GROUP BY with many int groups
+
+setup
+CREATE TEMP VIEW v(i1, i2) AS
+  SELECT * FROM generate_series(1, 4) g1, generate_series(1, 1000000) g2
+
+run
+SELECT sum(i1) FROM v GROUP BY i2;
diff --git a/bench/standard/aggregates/many_med_string_groups_1M.bench b/bench/standard/aggregates/many_med_string_groups_1M.bench
new file mode 100644
index 000000000..76adb16f9
--- /dev/null
+++ b/bench/standard/aggregates/many_med_string_groups_1M.bench
@@ -0,0 +1,9 @@
+# GROUP BY with many string groups
+
+setup
+CREATE TEMP VIEW v(ints, strings) AS
+  SELECT a, repeat(b::string, 100)
+    FROM generate_series(1, 4) g1(a), generate_series(1, 1000000) g2(b)
+
+run
+SELECT sum(ints) FROM v GROUP BY strings;
diff --git a/bench/standard/aggregates/many_small_string_groups_1M.bench b/bench/standard/aggregates/many_small_string_groups_1M.bench
new file mode 100644
index 000000000..af13bda28
--- /dev/null
+++ b/bench/standard/aggregates/many_small_string_groups_1M.bench
@@ -0,0 +1,8 @@
+# GROUP BY with many string groups
+
+setup
+CREATE TEMP VIEW v(ints, strings) AS
+  SELECT a, b::string FROM generate_series(1, 4) g1(a), generate_series(1, 1000000) g2(b)
+
+run
+SELECT sum(ints) FROM v GROUP BY strings;
diff --git a/crates/rayexec_bench/src/benchmark.rs b/crates/rayexec_bench/src/benchmark.rs
index f5d3a33a5..03149fe9e 100644
--- a/crates/rayexec_bench/src/benchmark.rs
+++ b/crates/rayexec_bench/src/benchmark.rs
@@ -20,6 +20,27 @@ impl Benchmark {
         Self::from_buf_read(reader)
     }
 
+    /// Construct a benchmark run from some reader.
+    ///
+    /// The content should include setup queries and benchmark queries.
+    ///
+    /// Setup queries are annotated with "setup":
+    ///
+    /// ```text
+    /// setup
+    /// CREATE TABLE ...
+    /// ```
+    ///
+    /// Benchmark queries are annotated with "run":
+    ///
+    /// ```text
+    /// run
+    /// SELECT * FROM ...
+    /// ```
+    ///
+    /// Any number of setup and benchmark queries can be provided, however a
+    /// setup query cannot be defined after a benchmark query has already been
+    /// defined.
     pub fn from_buf_read(reader: impl BufRead) -> Result<Self> {
         let lines = reader.lines();
 
diff --git a/crates/rayexec_bench/src/lib.rs b/crates/rayexec_bench/src/lib.rs
index 1aae34c97..7f89b2cc4 100644
--- a/crates/rayexec_bench/src/lib.rs
+++ b/crates/rayexec_bench/src/lib.rs
@@ -116,13 +116,13 @@ pub fn run(builder: impl EngineBuilder, default_dir: &str) -> Result<()> {
 
     // Print results.
     println!(
-        "{:<40}\t{:>7}\t{:>14}",
+        "{:<60}\t{:>6}\t{:>14}",
         "benchmark_name", "count", "time_ms"
     );
 
     for (name, times) in &all_times {
         for (idx, query_time) in times.query_times_ms.iter().enumerate() {
-            println!("{:<40}\t{:>7}\t{:>14}", name, idx + 1, query_time);
+            println!("{:<60}\t{:>6}\t{:>14}", name, idx + 1, query_time);
         }
     }