diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
index 196106f5..56ef87ff 100644
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
@@ -49,6 +49,7 @@ jobs:
         env:
           VCPKG_DEFAULT_TRIPLET: ${{ matrix.triplet }}
           VCPKG_DEFAULT_HOST_TRIPLET: ${{ matrix.triplet }}
+          VCPKG_BUILD_TYPE: release
         with:
           runVcpkgInstall: true
           runVcpkgFormatString: '["install", "--clean-after-build"]'
@@ -83,9 +84,6 @@ jobs:
       - run: opam exec -- opam install . --deps-only --with-test
       - run: opam exec -- dune build --verbose
 
-      - name: Give binary system-specific name
-        run: cp "_build/default/bin/ODiffBin.exe" "odiff-${{ matrix.artifact }}.exe"
-
       - run: opam exec -- dune exec ODiffBin -- --version 
       - run: opam exec -- dune runtest
 
@@ -106,11 +104,21 @@ jobs:
         run: npm ci
       - name: e2e test
         run: npm test
+        
+      - name: Build release binary
+        # this is needed because now ocaml's internal cygwin will be used on windows
+        # which breaks normal line endings
+        env:
+          SHELLOPTS: igncr
+        run: |
+          opam exec -- dune clean && \
+          opam exec -- dune build --release && \
+          cp "_build/default/bin/ODiffBin.exe" "odiff-${{ matrix.artifact }}.exe"
 
       - if: always()
         uses: actions/upload-artifact@v4
         with:
-          if-no-files-found: ignore
+          if-no-files-found: error
           name: odiff-${{ matrix.artifact }}.exe
           path: odiff-${{ matrix.artifact }}.exe
           retention-days: 14
diff --git a/bin/Main.ml b/bin/Main.ml
index 2fc64e26..4894b509 100644
--- a/bin/Main.ml
+++ b/bin/Main.ml
@@ -20,19 +20,22 @@ let main img1Path img2Path diffPath threshold outputDiffMask failOnLayoutChange
     diffColorHex toEmitStdoutParsableString antialiasing ignoreRegions diffLines
     disableGcOptimizations =
   (*
-      We do not need to actually maintain memory size of the allocated RAM by odiff, so we are
-      increasing the minor memory size to avoid most of the possible deallocations. For sure it is 
-      not possible be sure that it won't be run in OCaml because we allocate the Stack and Queue
+    Increase amount of allowed overhead to reduce amount of GC work and cycles.
+    we target 1-2 minor collections per run which is the best tradeoff between
+    amount of memory allocated and time spend on GC.
 
-      By default set the minor heap size to 256mb on 64bit machine
+    For sure it depends on the image size and architecture. Primary target x86_64
   *)
   if not disableGcOptimizations then
     Gc.set
       {
         (Gc.get ()) with
-        Gc.minor_heap_size = 64_000_000;
-        Gc.stack_limit = 2_048_000;
-        Gc.window_size = 25;
+        (* 128 MB *)
+        major_heap_increment = 128 * 1024 * 1024;
+        (* Reasonable high value to reduce major GC frequency *)
+        space_overhead = 500;
+        (* Disable compaction *)
+        max_overhead = 1_000_000;
       };
 
   let module IO1 = (val getIOModule img1Path) in
@@ -64,5 +67,5 @@ let main img1Path img2Path diffPath threshold outputDiffMask failOnLayoutChange
   | Some output when outputDiffMask -> IO1.freeImage output
   | _ -> ());
 
-  (*Gc.print_stat stdout;*)
+  (* Gc.print_stat stdout; *)
   exit exitCode
diff --git a/dune-project b/dune-project
index 880abdf0..1ac0d9dd 100644
--- a/dune-project
+++ b/dune-project
@@ -12,6 +12,7 @@
 
 (generate_opam_files true)
 
+(version 0.3.1-rc.3)
 (source (github dmtrKovalenko/odiff))
 (license MIT)
 (authors "Dmitriy Kovalenko")
@@ -19,7 +20,6 @@
 
 (package
  (name odiff)
- (version 0.3.1-rc.3)
  (synopsis "CLI for comparing images pixel-by-pixel")
  (depends
   odiff-core
diff --git a/images/water-diff.png b/images/water-diff.png
index 998be444..3fd60200 100644
Binary files a/images/water-diff.png and b/images/water-diff.png differ
diff --git a/io/jpg/dune b/io/jpg/dune
index 5517a6f3..0cac8cb1 100644
--- a/io/jpg/dune
+++ b/io/jpg/dune
@@ -8,7 +8,7 @@
   (language c)
   (names ReadJpg)
   (flags
-   (:include jpg_c_flags.sexp)))
+   (:include jpg_c_flags.sexp) -O3))
  (c_library_flags
   (:include jpg_c_library_flags.sexp))
  (libraries odiff-core WritePng))
diff --git a/io/png/ReadPng.c b/io/png/ReadPng.c
index fb9f6713..6526232c 100644
--- a/io/png/ReadPng.c
+++ b/io/png/ReadPng.c
@@ -24,7 +24,7 @@ CAMLprim value read_png_file(value file) {
 
   ctx = spng_ctx_new(0);
   if (ctx == NULL) {
-    fclose(png);    
+    fclose(png);
     caml_failwith("spng_ctx_new() failed");
   }
 
@@ -56,7 +56,8 @@ CAMLprim value read_png_file(value file) {
     caml_failwith(spng_strerror(result));
   };
 
-  ba = caml_ba_alloc(CAML_BA_UINT8 | CAML_BA_C_LAYOUT | CAML_BA_MANAGED, 1, NULL, &out_size);
+  ba = caml_ba_alloc(CAML_BA_UINT8 | CAML_BA_C_LAYOUT | CAML_BA_MANAGED, 1,
+                     NULL, &out_size);
   unsigned char *out = (unsigned char *)Caml_ba_data_val(ba);
 
   result =
diff --git a/io/png/dune b/io/png/dune
index a1746aad..39b282a1 100644
--- a/io/png/dune
+++ b/io/png/dune
@@ -8,7 +8,7 @@
   (language c)
   (names ReadPng)
   (flags
-   (:include png_c_flags.sexp)))
+   (:include png_c_flags.sexp) -O3))
  (c_library_flags
   (:include png_c_library_flags.sexp))
  (libraries odiff-core WritePng))
diff --git a/io/png_write/dune b/io/png_write/dune
index cff8c632..b967aac3 100644
--- a/io/png_write/dune
+++ b/io/png_write/dune
@@ -8,7 +8,7 @@
   (language c)
   (names WritePng)
   (flags
-   (:include png_write_c_flags.sexp)))
+   (:include png_write_c_flags.sexp) -O3))
  (c_library_flags
   (:include png_write_c_library_flags.sexp)))
 
diff --git a/io/tiff/dune b/io/tiff/dune
index 45d5e506..271b4cd0 100644
--- a/io/tiff/dune
+++ b/io/tiff/dune
@@ -8,7 +8,7 @@
   (language c)
   (names ReadTiff)
   (flags
-   (:include tiff_c_flags.sexp)))
+   (:include tiff_c_flags.sexp) -O3))
  (c_library_flags
   (:include tiff_c_library_flags.sexp))
  (libraries odiff-core WritePng))
diff --git a/odiff-core.opam b/odiff-core.opam
index f5cd5069..c997bd0e 100644
--- a/odiff-core.opam
+++ b/odiff-core.opam
@@ -1,5 +1,6 @@
 # This file is generated by dune, edit dune-project instead
 opam-version: "2.0"
+version: "0.3.1-rc.3"
 synopsis: "Pixel-by-pixel image difference algorithm"
 maintainer: ["https://dmtrkovalenko.dev" "dmtr.kovalenko@outlook.com"]
 authors: ["Dmitriy Kovalenko"]
diff --git a/odiff-io.opam b/odiff-io.opam
index 477b0ad0..6c178dac 100644
--- a/odiff-io.opam
+++ b/odiff-io.opam
@@ -1,5 +1,6 @@
 # This file is generated by dune, edit dune-project instead
 opam-version: "2.0"
+version: "0.3.1-rc.3"
 synopsis: "Ready to use io for odiff-core"
 maintainer: ["https://dmtrkovalenko.dev" "dmtr.kovalenko@outlook.com"]
 authors: ["Dmitriy Kovalenko"]
diff --git a/odiff-tests.opam b/odiff-tests.opam
index 693b2b65..c514dcd5 100644
--- a/odiff-tests.opam
+++ b/odiff-tests.opam
@@ -1,5 +1,6 @@
 # This file is generated by dune, edit dune-project instead
 opam-version: "2.0"
+version: "0.3.1-rc.3"
 synopsis: "Internal package for integration tests of odiff"
 maintainer: ["https://dmtrkovalenko.dev" "dmtr.kovalenko@outlook.com"]
 authors: ["Dmitriy Kovalenko"]
diff --git a/src/dune b/src/dune
index 190fb24b..1573174e 100644
--- a/src/dune
+++ b/src/dune
@@ -3,3 +3,12 @@
  (public_name odiff-core)
  (flags
   (-w -40 -w +26)))
+
+(env
+ (dev
+  (flags (:standard -w +42))
+  (ocamlopt_flags (:standard -S)))
+ (release
+  (ocamlopt_flags (:standard -O3 -rounds 5 -unbox-closures -inline 200 -inline-max-depth 7 -unbox-closures-factor 50))))
+
+