diff --git a/CHANGELOG.md b/CHANGELOG.md
index 2a412ef..1cc87ab 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -14,15 +14,21 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - `Transformation::rotate`. Creates a transformation representing a rotation. [#28]
 - `Batch::clear`. Clears the batch contents, useful to reuse batches in different
   frames.
+- Implementation of `Extend` for `Batch`. [#37]
+- Implementation of `ParallelExtend` for `Batch`. A `Batch` can now be populated
+  using multiple threads, useful to improve performance when dealing with many
+  thousands of quads. [#37]
 - Multiple gravity centers based on mouse clicks in the particles example. [#30]
 
 ### Changed
-- The performance of the particles example has been improved on Windows.
+- The performance of the particles example has been improved considerably on all
+  platforms. [#37]
 
 [#25]: https://github.com/hecrj/coffee/pull/25
 [#26]: https://github.com/hecrj/coffee/pull/26
 [#28]: https://github.com/hecrj/coffee/pull/28
 [#30]: https://github.com/hecrj/coffee/pull/30
+[#37]: https://github.com/hecrj/coffee/pull/37
 
 
 ## [0.2.0] - 2019-04-28
diff --git a/Cargo.toml b/Cargo.toml
index 046476a..92732ac 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -30,6 +30,7 @@ debug = []
 [dependencies]
 image = "0.21"
 nalgebra = "0.18"
+rayon = "1.0"
 
 # gfx (OpenGL)
 gfx = { version = "0.18", optional = true }
@@ -47,4 +48,3 @@ wgpu_glyph = { version = "0.3", optional = true }
 [dev-dependencies]
 # Example dependencies
 rand = "0.6"
-rayon = "1.0"
diff --git a/examples/particles.rs b/examples/particles.rs
index ec0790f..3efb8ae 100644
--- a/examples/particles.rs
+++ b/examples/particles.rs
@@ -28,8 +28,8 @@ struct Particles {
 }
 
 impl Particles {
-    // Try increasing this value! I (@hecrj) can render 350k particles at 100fps
-    // on my system. I have not tried going above that, yet...
+    // Try increasing this value! I (@hecrj) can render 1 MILLION particles at
+    // 90 fps on my system (4790k, GTX 980, Windows 7) using Vulkan.
     const AMOUNT: u32 = 50_000;
 
     // Play with these values to alter the way gravity works.
@@ -148,20 +148,19 @@ impl Game for Particles {
     fn draw(&self, view: &mut View, frame: &mut Frame, timer: &Timer) {
         frame.clear(Color::BLACK);
 
-        // Draw particles all at once!
-        view.batch.clear();
-
+        // When interpolating, we need to know how close the next tick is
         let delta_factor = if view.interpolate {
             timer.next_tick_proximity()
         } else {
             0.0
         };
 
-        for particle in &self.particles {
+        // Generate sprites in parallel! <3 rayon
+        let sprites = self.particles.par_iter().map(|particle| {
             let velocity =
                 particle.velocity + particle.acceleration * delta_factor;
 
-            view.batch.add(Sprite {
+            Sprite {
                 source: Rectangle {
                     x: View::particle_color(velocity),
                     y: 0,
@@ -169,9 +168,16 @@ impl Game for Particles {
                     height: 1,
                 },
                 position: particle.position + velocity * delta_factor,
-            });
-        }
+            }
+        });
+
+        // Clear batch contents from previous frame
+        view.batch.clear();
 
+        // Use the parallel iterator to populate the batch efficiently
+        view.batch.par_extend(sprites);
+
+        // Draw particles all at once!
         view.batch
             .draw(Point::new(0.0, 0.0), &mut frame.as_target());
 
diff --git a/src/graphics/batch.rs b/src/graphics/batch.rs
index 66ae550..02f1f94 100644
--- a/src/graphics/batch.rs
+++ b/src/graphics/batch.rs
@@ -1,3 +1,5 @@
+use rayon::prelude::*;
+
 use crate::graphics::gpu;
 use crate::graphics::{Image, IntoQuad, Point, Target, Transformation, Vector};
 
@@ -66,3 +68,44 @@ impl std::fmt::Debug for Batch {
         write!(f, "Batch {{ image: {:?} }}", self.image,)
     }
 }
+
+impl<Q: IntoQuad> Extend<Q> for Batch {
+    fn extend<I>(&mut self, iter: I)
+    where
+        I: IntoIterator<Item = Q>,
+    {
+        let iter = iter.into_iter();
+        let x_unit = self.x_unit;
+        let y_unit = self.y_unit;
+
+        self.instances.extend(
+            iter.map(|quad| {
+                gpu::Instance::from(quad.into_quad(x_unit, y_unit))
+            }),
+        );
+    }
+}
+
+/// Extend the [`Batch`] using a parallel iterator from [`rayon`].
+///
+/// If you are dealing with many thousands of quads, `par_extend` can help you
+/// speed up your drawing by using multiple threads to populate a [`Batch`].
+///
+/// [`Batch`]: struct.Batch.html
+/// [`rayon`]: https://docs.rs/rayon/1.0/rayon/
+impl<Q: IntoQuad + Send> ParallelExtend<Q> for Batch {
+    fn par_extend<I>(&mut self, par_iter: I)
+    where
+        I: IntoParallelIterator<Item = Q>,
+    {
+        let par_iter = par_iter.into_par_iter();
+        let x_unit = self.x_unit;
+        let y_unit = self.y_unit;
+
+        self.instances.par_extend(
+            par_iter.map(|quad| {
+                gpu::Instance::from(quad.into_quad(x_unit, y_unit))
+            }),
+        );
+    }
+}