diff --git a/README.md b/README.md
index f70e8d4..11fdfcd 100644
--- a/README.md
+++ b/README.md
@@ -21,4 +21,12 @@ the top left tile.
 
     slicerdicer --help
 
-    slicerdicer --filename foo.png --tile-size 256
+    slicerdicer --filename foo.png --tile-size 256 --concurrency 5
+
+## Notes
+
+It's going to eat some memory. 
+
+In my tests on an 32641 x 16471, 8-bit/color RGB PNG, memory usage peaks at
+around 2.7GB.
+
diff --git a/main.go b/main.go
index fd3500a..8814c74 100644
--- a/main.go
+++ b/main.go
@@ -5,15 +5,14 @@ import "image/png"
 import "github.com/disintegration/imaging"
 import "runtime"
 import "flag"
-
-
 import "fmt"
 import "os"
 
 func main() {
 
-  filenamePtr := flag.String("filename", "screenshot.png", "filename to open")
-  tile_size   := flag.Int   ("tile-size", 512, "tile size, in pixels")
+  filenamePtr  := flag.String("filename", "screenshot.png", "filename to open")
+  tileSizePtr := flag.Int   ("tile-size", 512, "tile size, in pixels")
+  concurrencyPtr := flag.Int ("concurrency", 5, "how many tiles to generate concurrently (threads)")
 
   flag.Parse()
 
@@ -26,20 +25,24 @@ func main() {
 
   size := src.Bounds().Max
 
-  tile_size_x := *tile_size
-  tile_size_y := *tile_size
-
-  fmt.Println("starting tiling")
+  tile_size_x := *tileSizePtr
+  tile_size_y := *tileSizePtr
 
   z := 0
 
+  concurrency := *concurrencyPtr
+  sem := make(chan bool, concurrency)
+
+  fmt.Println("starting tiling with concurrency of", concurrency)
+
   // outer loop for zoom
   for {
     if (z == 0) {
       // do nothing
     } else {
       // halve image size
-      src = imaging.Resize(src, size.X/2, 0, imaging.Linear)
+      src = imaging.Resize(src, size.X/2, 0, imaging.NearestNeighbor)
+      runtime.GC()
       // recalculate size
       size = src.Bounds().Max
       // we are done if we are now smaller then the tile
@@ -51,29 +54,37 @@ func main() {
     fmt.Print(fmt.Sprintf("zoom level: %d (%d x %d)\n", z, size.X, size.Y))
 
     for y := 0 ; y <= (size.Y / tile_size_y) ; y++ {
-
       for x := 0 ; x <= (size.X / tile_size_x) ; x++ {
- 
-        output_filename := fmt.Sprintf("tile-%d-%d-%d.png", z, x, y)
-        cropped := imaging.Crop(src, image.Rect(tile_size_x*x, tile_size_y*y, tile_size_x*x+tile_size_x, tile_size_y*y+tile_size_y));
-
-        fmt.Print("writing to: ", output_filename, "        ");
-        fmt.Print("\r")
-
-        writer, _ := os.Create(output_filename)
-        err = png.Encode(writer, cropped)
-        writer.Close()
-        runtime.GC()
-        if err != nil {
-          fmt.Println(err)
-        }
+        sem <- true
+        go tile(src, z, x, y, tile_size_x, tile_size_y, sem)
       }
+
     }
 
-    fmt.Print("\r                                   \r")
     z++
   }
 
-  fmt.Println("done")
+  // drain at the end of each zoom level
+  // since we are about to modify the source image
+  for i := 0; i < cap(sem); i++ {
+    sem <- true
+  }
 
+  fmt.Println("done")
+}
+
+func tile (src image.Image, z, x, y int, tile_size_x, tile_size_y int, sem chan bool) {
+  defer func() { <-sem }()
+  output_filename := fmt.Sprintf("tile-%d-%d-%d.png", z, x, y)
+  cropped := imaging.Crop(src, image.Rect(tile_size_x*x, tile_size_y*y, tile_size_x*x+tile_size_x, tile_size_y*y+tile_size_y));
+
+  writer, _ := os.Create(output_filename)
+  err := png.Encode(writer, cropped)
+  if err != nil {
+    fmt.Println(err)
+  }
+  writer.Close()
+
+  runtime.GC()
+  return;
 }