diff --git a/render/internal_test.go b/render/internal_test.go
index 11808b4..69f1fd0 100644
--- a/render/internal_test.go
+++ b/render/internal_test.go
@@ -4,7 +4,7 @@ import (
 	"bytes"
 	"errors"
 	"io"
-	"os"
+	"runtime"
 	"testing"
 
 	"github.com/soypat/sdf/form3/must3"
@@ -76,20 +76,44 @@ func TestSTLWriteReadback(t *testing.T) {
 }
 
 func TestOctreeMultithread(t *testing.T) {
-	oct := NewOctreeRenderer(must3.Sphere(1), 8)
+	oct := NewOctreeRenderer(must3.Sphere(20), 100)
 	oct.concurrent = 2
-	buf := make([]Triangle3, oct.concurrent*10)
+	buf := make([]Triangle3, oct.concurrent*100)
 	var err error
 	var nt int
 	var model []Triangle3
 	for err == nil {
 		nt, err = oct.ReadTriangles(buf)
-		model = append(model, buf[nt:]...)
+		model = append(model, buf[:nt]...)
 	}
 	if err != io.EOF {
 		t.Fatal(err)
 	}
-	fp, _ := os.Create("mt.stl")
-	defer fp.Close()
-	WriteSTL(fp, model)
+	if len(model) != oct.triangles {
+		t.Errorf("triangles lost. got %d. octree read %d", len(model), oct.triangles)
+	}
+	if oct.cubes != oct.cubesP {
+		t.Errorf("number of non empty cubes found %d must match number of cubes processed %d", oct.cubes, oct.cubesP)
+	}
+	t.Log(oct.triangles)
+
+}
+
+func BenchmarkBoltThreaded(b *testing.B) {
+	const output = "threaded_bolt.stl"
+	npt := thread.NPT{}
+	npt.SetFromNominal(1.0 / 2.0)
+	object, _ := thread.Bolt(thread.BoltParms{
+		Thread:      npt, // M16x2
+		Style:       thread.NutHex,
+		Tolerance:   0.1,
+		TotalLength: 20,
+		ShankLength: 10,
+	})
+
+	for i := 0; i < b.N; i++ {
+		oct := NewOctreeRenderer(object, 300)
+		oct.concurrent = runtime.NumCPU()
+		CreateSTL(output, oct)
+	}
 }
diff --git a/render/octree_renderer.go b/render/octree_renderer.go
index 8eb5454..c136c17 100644
--- a/render/octree_renderer.go
+++ b/render/octree_renderer.go
@@ -18,6 +18,12 @@ type octree struct {
 	unwritten triangle3Buffer
 	// concurrent goroutine processing.
 	concurrent int
+	// Number of triangles generated.
+	triangles int
+	// number of non empty cubes found.
+	cubes int
+	// number of cubes processed.
+	cubesP int
 }
 
 type cube struct {
@@ -56,8 +62,8 @@ func NewOctreeRenderer(s sdf.SDF3, meshCells int) *octree {
 	return &octree{
 		dc:        *newDc3(s, bb.Min, resolution, levels),
 		unwritten: triangle3Buffer{buf: make([]Triangle3, 0, 1024)},
-		todo:      cubes, //[]cube{{sdf.V3i{0, 0, 0}, levels - 1}}, // process the octree, start at the top level
-		// concurrent: 2,
+		todo:      cubes,
+		cubes:     1,
 	}
 }
 
@@ -79,7 +85,7 @@ func (oc *octree) ReadTriangles(dst []Triangle3) (n int, err error) {
 	}
 	// Number of additional triangles proccessed.
 	var nt int
-	if oc.concurrent <= 1 || len(oc.todo) < oc.concurrent || n < oc.concurrent {
+	if oc.concurrent < 1 || len(oc.todo) < oc.concurrent || len(dst) < oc.concurrent {
 		tproc, nc, newCubes := oc.readTriangles(dst[n:], oc.todo)
 		oc.todo = append(oc.todo, newCubes...)
 		oc.todo = oc.todo[nc:] // this leaks, luckily this is a short lived function?
@@ -124,6 +130,7 @@ func (oc *octree) readTriangles(dst []Triangle3, todo []cube) (n, cubesProcessed
 }
 
 // readTrianglesThreaded is a multithreaded triangle reader implementation for octree.
+// It writes nt triangles into dst.
 func (oc *octree) readTrianglesThreaded(dst []Triangle3) (nt int) {
 	var wg sync.WaitGroup
 	div := len(dst) / oc.concurrent
@@ -136,8 +143,13 @@ func (oc *octree) readTrianglesThreaded(dst []Triangle3) (nt int) {
 		wg.Add(1)
 		go func() {
 			start := div * i
-			work[i] = dst[start : start+div]
-			cubeWork[i] = oc.todo[i*divC : (i+1)*divC]
+			if i == oc.concurrent-1 {
+				work[i] = dst[start:]
+				cubeWork[i] = oc.todo[i*divC:]
+			} else {
+				work[i] = dst[start : start+div]
+				cubeWork[i] = oc.todo[i*divC : (i+1)*divC]
+			}
 			ntc, nc, newC := oc.readTriangles(work[i], cubeWork[i])
 			newCubesC[i] = newC
 			work[i] = work[i][:ntc]
@@ -153,8 +165,8 @@ func (oc *octree) readTrianglesThreaded(dst []Triangle3) (nt int) {
 		// Triangles written.
 		start := div*i - offset
 		if i != oc.concurrent-1 && len(work[i]) != div {
-			offset += div - len(work[i])
 			copy(dst[start+len(work[i]):], dst[start+div:])
+			offset += div - len(work[i])
 		}
 		nt += len(work[i])
 		// Cubes unprocessed.
@@ -207,6 +219,11 @@ func (oc *octree) processCube(dst []Triangle3, c cube) (writtenTriangles int, ne
 			}
 		}
 	}
+	oc.mu.Lock()
+	oc.triangles += writtenTriangles
+	oc.cubes += len(newCubes)
+	oc.cubesP++
+	oc.mu.Unlock()
 	return writtenTriangles, newCubes
 }
 
diff --git a/render/stl.go b/render/stl.go
index 50f90d9..da4eefb 100644
--- a/render/stl.go
+++ b/render/stl.go
@@ -76,7 +76,7 @@ func (h stlHeader) put(b []byte) {
 	binary.LittleEndian.PutUint32(b[80:], h.Count)
 }
 
-const trianglesInBuffer = 1 << 10
+const trianglesInBuffer = 1 << 14
 
 type stlReader struct {
 	r   Renderer