diff --git a/rtiow/renderer/src/bvh_triangles.rs b/rtiow/renderer/src/bvh_triangles.rs
index 4b15359..4113ab6 100644
--- a/rtiow/renderer/src/bvh_triangles.rs
+++ b/rtiow/renderer/src/bvh_triangles.rs
@@ -59,6 +59,12 @@ where
     bvh_nodes: Vec<BVHNode>,
 }
 
+#[derive(Debug, Default)]
+struct Bin {
+    bounds: AABB,
+    tri_count: usize,
+}
+
 impl<M> fmt::Debug for BVHTriangles<M>
 where
     M: Material,
@@ -102,6 +108,7 @@ where
     }
 }
 
+#[derive(Debug)]
 struct SplitCost {
     pos: f32,
     axis: usize,
@@ -175,6 +182,7 @@ where
         info!("BVHTriangles build stats:");
         info!("  Nodes:   {}", stats.nodes);
         info!("  Leaves:  {}", stats.leafs);
+        info!("  Tris:    {}", bvh.triangles.len());
         info!("  Min Tri: {}", stats.min_tris);
         info!("  Max Tri: {}", stats.max_tris);
         info!("  Avg Tri: {}", bvh.triangles.len() / stats.leafs);
@@ -226,20 +234,71 @@ where
         };
 
         for axis in 0..3 {
-            let bounds_min = node.aabb.min()[axis];
-            let bounds_max = node.aabb.max()[axis];
+            let mut bounds_min = f32::MAX;
+            let mut bounds_max = f32::MIN;
+
+            for i in 0..node.tri_count {
+                let triangle = &self.triangles[self.triangle_index[(node.left_first + i) as usize]];
+                bounds_min = bounds_min.min(triangle.centroid[axis]);
+                bounds_max = bounds_max.max(triangle.centroid[axis]);
+            }
+
             if bounds_min == bounds_max {
                 continue;
             }
-            let scale = (bounds_max - bounds_min) / 100.;
 
-            for i in 0..100 {
-                let candidate_pos = bounds_min + (i as f32) * scale;
-                let cost = self.evaluate_sah(node, axis, candidate_pos);
-                if cost <= best.cost {
-                    best.pos = candidate_pos;
+            const NUM_BINS: usize = 8;
+            let mut bins: Vec<_> = (0..NUM_BINS)
+                .map(|_| Bin {
+                    bounds: AABB::infinite(),
+                    tri_count: 0,
+                })
+                .collect();
+            let scale = bins.len() as f32 / (bounds_max - bounds_min);
+            // populate the bins
+            for i in 0..node.tri_count {
+                let triangle = &self.triangles[self.triangle_index[(node.left_first + i) as usize]];
+                let bin_idx = (triangle.centroid[axis] - bounds_min) * scale;
+                let bin_idx = (bins.len() - 1).min(bin_idx as usize);
+
+                bins[bin_idx].tri_count += 1;
+                bins[bin_idx].bounds.grow(triangle.verts[0]);
+                bins[bin_idx].bounds.grow(triangle.verts[1]);
+                bins[bin_idx].bounds.grow(triangle.verts[2]);
+            }
+
+            // gather data for the 7 planes between the 8 bins
+            let mut left_area: Vec<_> = (0..bins.len() - 1).map(|_| 0.).collect();
+            let mut right_area: Vec<_> = (0..bins.len() - 1).map(|_| 0.).collect();
+            let mut left_count: Vec<_> = (0..bins.len() - 1).map(|_| 0).collect();
+            let mut right_count: Vec<_> = (0..bins.len() - 1).map(|_| 0).collect();
+            let mut left_box = AABB::infinite();
+            let mut right_box = AABB::infinite();
+            let mut left_sum = 0;
+            let mut right_sum = 0;
+            for i in 0..(bins.len() - 1) {
+                left_sum += bins[i].tri_count;
+                left_count[i] = left_sum;
+                left_box.grow(bins[i].bounds.min());
+                left_box.grow(bins[i].bounds.max());
+                left_area[i] = left_box.area();
+
+                right_sum += bins[bins.len() - 1 - i].tri_count;
+                right_count[bins.len() - 2 - i] = right_sum;
+                right_box.grow(bins[bins.len() - 1 - i].bounds.min());
+                right_box.grow(bins[bins.len() - 1 - i].bounds.max());
+                right_area[bins.len() - 2 - i] = right_box.area();
+            }
+
+            let scale = (bounds_max - bounds_min) / bins.len() as f32;
+            // calculate SAH cost for the 7 planes
+            for i in 0..(bins.len() - 1) {
+                let plane_cost =
+                    left_count[i] as f32 * left_area[i] + right_count[i] as f32 * right_area[i];
+                if plane_cost < best.cost {
                     best.axis = axis;
-                    best.cost = cost;
+                    best.pos = bounds_min + scale * (i as f32 + 1.);
+                    best.cost = plane_cost;
                 }
             }
         }
@@ -306,34 +365,6 @@ where
         self.subdivide(right_child_idx as usize);
     }
 
-    fn evaluate_sah(&self, node: &BVHNode, axis: usize, pos: f32) -> f32 {
-        // determine triangle counts and bounds for this split candidate
-        let mut left_box = AABB::infinite();
-        let mut right_box = AABB::infinite();
-        let mut left_count = 0;
-        let mut right_count = 0;
-        for i in 0..node.tri_count {
-            let triangle = &self.triangles[self.triangle_index[(node.left_first + i) as usize]];
-            if triangle.centroid[axis] < pos {
-                left_count += 1;
-                left_box.grow(triangle.verts[0]);
-                left_box.grow(triangle.verts[1]);
-                left_box.grow(triangle.verts[2]);
-            } else {
-                right_count += 1;
-                right_box.grow(triangle.verts[0]);
-                right_box.grow(triangle.verts[1]);
-                right_box.grow(triangle.verts[2]);
-            }
-        }
-        let cost = left_count as f32 * left_box.area() + right_count as f32 * right_box.area();
-        if cost > 0. {
-            cost
-        } else {
-            f32::MAX
-        }
-    }
-
     fn intersect_bvh(&self, r: Ray, t_min: f32, t_max: f32) -> Option<HitRecord> {
         let mut node = &self.bvh_nodes[ROOT_NODE_IDX];
         let mut stack = Vec::with_capacity(2);