diff --git a/config.toml.example b/config.toml.example
index 299bfd779e57a..b0dcb0738ab20 100644
--- a/config.toml.example
+++ b/config.toml.example
@@ -427,6 +427,12 @@ changelog-seen = 2
 # Uses the rustc defaults: https://doc.rust-lang.org/rustc/codegen-options/index.html#codegen-units
 #codegen-units = if incremental { 256 } else { 16 }
 
+# Tweaks the numbers of codegen units to improve compile time, typically by
+# using a single codegen unit for crates. The `codegen-units` configuration is
+# preferred over this field. This is recommended to set for release builds and CI,
+# but not for development as it can slow down partial rebuilds.
+#codegen-units-fast = false
+
 # Sets the number of codegen units to build the standard library with,
 # regardless of what the codegen-unit setting for the rest of the compiler is.
 # NOTE: building with anything other than 1 is known to occasionally have bugs.
diff --git a/src/bootstrap/bin/rustc.rs b/src/bootstrap/bin/rustc.rs
index 9611c866df599..79a115598994b 100644
--- a/src/bootstrap/bin/rustc.rs
+++ b/src/bootstrap/bin/rustc.rs
@@ -70,6 +70,29 @@ fn main() {
                 cmd.arg("-Ztime-passes");
             }
         }
+
+        if let Some(profile) = env::var_os("RUSTC_TUNE_COMPILER_CODEGEN_UNITS") {
+            let cgus = match crate_name {
+                // This crate is large and at the tail of the compiler crates, give it extra CGUs.
+                //"rustc_query_impl" => Some(96),
+                // This compiles after all other crates so give it default CGUs to speed it up.
+                "rustc_driver" => None,
+                _ => {
+                    if profile == "fast" {
+                        Some(1)
+                    } else {
+                        if crate_name.starts_with("rustc_") {
+                            None
+                        } else {
+                            // Compile crates.io crates with a single CGU for faster compile times
+                            Some(1)
+                        }
+                    }
+                }
+            };
+
+            cgus.map(|cgus| cmd.arg(&format!("-Ccodegen-units={}", cgus)));
+        }
     }
 
     // Print backtrace in case of ICE
diff --git a/src/bootstrap/builder.rs b/src/bootstrap/builder.rs
index b4fc1d4f28da7..16ca1966823c3 100644
--- a/src/bootstrap/builder.rs
+++ b/src/bootstrap/builder.rs
@@ -1884,6 +1884,9 @@ impl<'a> Builder<'a> {
             (Mode::Std, Some(n), _) | (_, _, Some(n)) => {
                 cargo.env(profile_var("CODEGEN_UNITS"), n.to_string());
             }
+            (Mode::Rustc, _, _) => {
+                cargo.env("RUSTC_TUNE_COMPILER_CODEGEN_UNITS", "fast");
+            }
             _ => {
                 // Don't set anything
             }
diff --git a/src/bootstrap/config.rs b/src/bootstrap/config.rs
index fdd659c60ca7f..a45ff0e798252 100644
--- a/src/bootstrap/config.rs
+++ b/src/bootstrap/config.rs
@@ -145,6 +145,7 @@ pub struct Config {
     pub rust_optimize: bool,
     pub rust_codegen_units: Option<u32>,
     pub rust_codegen_units_std: Option<u32>,
+    pub rust_codegen_units_fast: bool,
     pub rust_debug_assertions: bool,
     pub rust_debug_assertions_std: bool,
     pub rust_overflow_checks: bool,
@@ -717,6 +718,7 @@ define_config! {
         debug: Option<bool> = "debug",
         codegen_units: Option<u32> = "codegen-units",
         codegen_units_std: Option<u32> = "codegen-units-std",
+        codegen_units_fast: Option<bool> = "codegen-units-fast",
         debug_assertions: Option<bool> = "debug-assertions",
         debug_assertions_std: Option<bool> = "debug-assertions-std",
         overflow_checks: Option<bool> = "overflow-checks",
@@ -1130,6 +1132,7 @@ impl Config {
 
             config.rust_codegen_units = rust.codegen_units.map(threads_from_config);
             config.rust_codegen_units_std = rust.codegen_units_std.map(threads_from_config);
+            set(&mut config.rust_codegen_units_fast, rust.codegen_units_fast);
             config.rust_profile_use = flags.rust_profile_use.or(rust.profile_use);
             config.rust_profile_generate = flags.rust_profile_generate.or(rust.profile_generate);
             config.download_rustc_commit = config.download_ci_rustc_commit(rust.download_rustc);
diff --git a/src/ci/run.sh b/src/ci/run.sh
index 0db9c993eecb4..f6c3b2b3eb902 100755
--- a/src/ci/run.sh
+++ b/src/ci/run.sh
@@ -56,6 +56,7 @@ RUST_CONFIGURE_ARGS="$RUST_CONFIGURE_ARGS --disable-manage-submodules"
 RUST_CONFIGURE_ARGS="$RUST_CONFIGURE_ARGS --enable-locked-deps"
 RUST_CONFIGURE_ARGS="$RUST_CONFIGURE_ARGS --enable-cargo-native-static"
 RUST_CONFIGURE_ARGS="$RUST_CONFIGURE_ARGS --set rust.codegen-units-std=1"
+RUST_CONFIGURE_ARGS="$RUST_CONFIGURE_ARGS --set rust.codegen-units-fast=true"
 
 # Only produce xz tarballs on CI. gz tarballs will be generated by the release
 # process by recompressing the existing xz ones. This decreases the storage