diff --git a/magyarsort.h b/magyarsort.h
index d7413b3..5c6fab1 100644
--- a/magyarsort.h
+++ b/magyarsort.h
@@ -216,17 +216,38 @@ namespace MagyarSort {
 
 	};
 
-	/**
+	/*
 	 * Sort the given array (in-place sorting) with the given size.
 	 *
 	 * Rem.: If you use the VectorGiverWithReuse please remind yourself to Gc() it time-to-time!
 	 *
+	 * Beware: GC needs to happen on all threads that use us!
+	 *
 	 * @param arr The array to sort. Result will be in the same array - as sorted.
 	 * @param size The lenght of the array.
-	 * @param VectorGiver is either VectorGiverHeap or VectorGiverWithReuse. Have Give(size_t size, ...) returning value or ref.
+	 * @param REUSE OPTIONAL: When true, we reuse the array instead of always gettin' and releasin' from da heap.
+	 * @param GC OPTIONAL: When true, we garbage collect memory from previous sorts if REUSE is true.
+	 * @param GC_WITHOUT_SORT OPTIONAL: When true, we "just GC" but do not sort in case of GC is true.
 	 */
-	//template<typename VectorGiver = VectorGiverWithReuse>
-	inline void sort(uint32_t arr[], size_t size) noexcept {
+	template<bool REUSE = false, bool GC = false, bool GC_WITHOUT_SORT = false>
+	inline void __attribute__((always_inline)) sort_impl(uint32_t arr[], size_t size) noexcept {
+		// Most funny optimization is this multiply here :-)
+		//
+		// Literally.. come on.. this makes it nearly a compile-time, macro-like
+		// ifdef-like thing as we avoid memory allocations of size BUT also we
+		// optimize the first call for sort when we REUSE the array so size is fine!
+		static thread_local std::vector<uint32_t> arc(size * REUSE);
+
+		// "Garbage-collection"
+		if(GC) {
+			arc = std::vector<uint32_t>();
+			// This must be implemented, because we can only access
+			// the static in our function body so this is the "way".
+			if(GC_WITHOUT_SORT) {
+				return;
+			}
+		}
+
 		// Holds "digit" occurences, prefix sums, whatevers
 		// First "DIGIT_RANGE" elem is for MSB "DIGITS", last is for LSB
 		static thread_local size_t radics[DIGITS * DIGIT_RANGE];
@@ -250,10 +271,19 @@ namespace MagyarSort {
 		// above already anyways...
 
 		// Regular radix sort needs a copy, see: https://www.youtube.com/watch?v=ujb2CIWE8zY
+		// But instead of the below, we do a trickery...
+		//
 		//std::vector<uint32_t> arc(size);
 		//auto arc = VectorGiver::Give(size); // "auto" is needed for this to perform well with some givers!
-		static thread_local std::vector<uint32_t> arc(size);
-		arc.resize(size); // JHP
+		//
+		// Rem.: The branch is optimized out in compile time!
+		if(REUSE) {
+			arc.resize(size);
+		} else {
+			// Must not be .clean() !!!
+			// We must regain memory of previous!
+			arc = std::move(std::vector<uint32_t>(size));
+		}
 
 		uint32_t *from = arr;
 		uint32_t *to = &arc[0];
@@ -262,11 +292,90 @@ namespace MagyarSort {
 
 		// With an other API we could spare this copy if we can delete original arr and return ptr or something...
 		// I am fine with this... this is not my main idea anyways, just little ILP tweak to regular radix sort
-		//if(to != arr) { // <- logically, but bad they are already swapped here!!! BEWARE
+		//if(to != arr) // <- logically, but bad they are already swapped here!!! BEWARE
 		if(from != arr) { // <- in reality this is what we want because of last swap happened anyways!
 			memcpy(arr, from, size);
 		}
 	}
+
+	/**
+	 * Garbage collect reused data structures from last call.
+	 *
+	 * This is optimized and is a NO-OP if MAGYAR_SORT_DEFAULT_REUSE is not defined!
+	 *  - unless you use the FORCE! May it be with you if you need it.
+	 *
+	 * @param FORCE OPTIONAL: When true, the gc happens even if MAGYAR_SORT_DEFAULT_REUSE is not defined!
+	 */
+	template<bool FORCE = false>
+	inline void gc() noexcept {
+		if(FORCE) {
+			// Only GC-ing
+			MagyarSort::sort_impl<true, true, true>(nullptr, 0);
+		} else {
+#ifdef MAGYAR_SORT_DEFAULT_REUSE
+			// Only GC-ing
+			MagyarSort::sort_impl<true, true, true>(nullptr, 0);
+#endif
+		}
+	}
+
+	/**
+	 * Sort the given array (in-place sorting) with the given size.
+	 *
+	 * Rem.: Please remind yourself to cc() from time-to-time!
+	 * Rem.: Thread-safe to use!
+	 *
+	 * Beware: MagyarSort::gc<true>(); needs to happen on all threads that use this variant otherwise memory leaks away!
+	 *         Please mind the "true" template parameter that forces the GC even when sort by default not reuses...
+	 *
+	 * @param arr The array to sort. Result will be in the same array - as sorted.
+	 * @param size The lenght of the array.
+	 * @param GC OPTIONAL: When true, we garbage collect before this sort - so cached memory size will be "size" elems.
+	 */
+	template<bool GC = false>
+	inline void __attribute__((always_inline)) sort_reuse(uint32_t arr[], size_t size) noexcept {
+		// Reuse the temporary vectors across runs
+		// This results in much less heap allocations and much faster on gcc
+		// and also a bit faster on clang too.
+		MagyarSort::sort_impl<true>(arr, size);
+	}
+
+	/**
+	 * Sort the given array (in-place sorting) with the given size.
+	 *
+	 * Rem.: Thread-safe to use!
+	 *
+	 * Beware: MagyarSort::gc(); needs to happen on all threads that use this variant otherwise memory leaks away!
+	 *
+	 * @param arr The array to sort. Result will be in the same array - as sorted.
+	 * @param size The lenght of the array.
+	 */
+	inline void __attribute__((always_inline)) sort_no_reuse(uint32_t arr[], size_t size) noexcept {
+		// We use the heap once per every call...
+		// This is safer and we do not need garbage collecting
+		MagyarSort::sort_impl(arr, size);
+	}
+
+	/*
+	 * Sort the given array (in-place sorting) with the given size.
+	 *
+	 * Rem.: If you use the VectorGiverWithReuse please remind yourself to Gc() it time-to-time!
+	 *
+	 * Beware: MagyarSort::gc(); should be called after "sort bursts" (consecutive fast sorts of when you need memory
+	 *         on all threads that use this variant otherwise memory leaks away as biggest sorted array keeps being in ram!
+	 *         This depends on the config #define MAGYAR_SORT_DEFAULT_REUSE is defined or not. Define and you get reuse
+	 *         and if you get reuse you can call multiple sorts with reused temporary buffers that you gc() afterwards!
+	 *
+	 * @param arr The array to sort. Result will be in the same array - as sorted.
+	 * @param size The lenght of the array.
+	 */
+	inline void sort(uint32_t arr[], size_t size) noexcept {
+#ifdef MAGYAR_SORT_DEFAULT_REUSE
+		MagyarSort::sort_reuse(arr, size);
+#else
+		MagyarSort::sort_no_reuse(arr, size);
+#endif
+	}
 };
 
 #endif
diff --git a/simd-sort/speed.cpp b/simd-sort/speed.cpp
index aa58089..c60c1b4 100644
--- a/simd-sort/speed.cpp
+++ b/simd-sort/speed.cpp
@@ -14,6 +14,7 @@
 #include "quicksort-all.cpp"
 #include "avx2-altquicksort.h"
 //#include "avx2-nate-quicksort.cpp"
+#define MAGYAR_SORT_DEFAULT_REUSE
 #include "../magyarsort.h" // mine
 #include "avx2-natenodutch-quicksort.h"
 #define USE_RDTSC // undef to get measurments in seconds
diff --git a/simd-sort/speed_avx2 b/simd-sort/speed_avx2
index 6bbb97a..cb8adeb 100755
Binary files a/simd-sort/speed_avx2 and b/simd-sort/speed_avx2 differ
diff --git a/test.cpp b/test.cpp
index 5572222..cd946f1 100644
--- a/test.cpp
+++ b/test.cpp
@@ -32,6 +32,8 @@
 #include <vector>
 #include <chrono>
 #include <algorithm> // std::sort
+
+#define MAGYAR_SORT_DEFAULT_REUSE
 #include "magyarsort.h"
 
 #ifdef SKA_SORT
diff --git a/ypsu.cpp b/ypsu.cpp
index 1d11e51..859887c 100644
--- a/ypsu.cpp
+++ b/ypsu.cpp
@@ -11,7 +11,11 @@
   #include <string>
   #include <vector>
   #include "ska_sort.hpp"
+
+
+  #define MAGYAR_SORT_DEFAULT_REUSE
   #include "magyarsort.h"
+
   std::map<std::string, double> results;
   std::map<std::string, double> worst;
   void measure(const std::string &inputtype, const std::string &name,