diff --git a/taskpools/taskpools.nim b/taskpools/taskpools.nim
index 097f58c..b7f3e1d 100644
--- a/taskpools/taskpools.nim
+++ b/taskpools/taskpools.nim
@@ -345,9 +345,23 @@ proc syncAll*(pool: Taskpool) {.raises: [Exception].} =
 # Runtime
 # ---------------------------------------------
 
-proc new*(T: type Taskpool, numThreads = countProcessors()): T {.raises: [Exception].} =
+proc new*(T: type Taskpool, numThreads = countProcessors(), pinToCPU = false): T {.raises: [Exception].} =
   ## Initialize a threadpool that manages `numThreads` threads.
   ## Default to the number of logical processors available.
+  ##
+  ## If pinToCPU is set, threads spawned will be pinned to the core that spawned them.
+  ## This improves performance of memory-intensive workloads by avoiding
+  ## thrashing and reloading core caches when a thread moves around.
+  ##
+  ## pinToCPU option is ignored in:
+  ## - In C++ compilation with Microsoft Visual Studio Compiler
+  ## - MacOS
+  ## - Android
+  #
+  # pinToCPU Status:
+  # - C++ MSVC: implementation missing (need to wrap reinterpret_cast)
+  # - Android: API missing and unclear benefits due to Big.Little architecture
+  # - MacOS: API missing
 
   type TpObj = typeof(default(Taskpool)[])
   # Event notifier requires an extra 64 bytes for alignment
@@ -363,22 +377,26 @@ proc new*(T: type Taskpool, numThreads = countProcessors()): T {.raises: [Except
   # Setup master thread
   workerContext.id = 0
   workerContext.taskpool = tp
-  when not(defined(cpp) and defined(vcc)):
-    # TODO: Nim casts between Windows Handles but that requires reinterpret cast for C++
-    pinToCpu(0)
+
+  if pinToCPU:
+    when not(defined(cpp) and defined(vcc)):
+      # TODO: Nim casts between Windows Handles but that requires reinterpret cast for C++
+      pinToCpu(0)
 
   # Start worker threads
   for i in 1 ..< numThreads:
     createThread(tp.workers[i], worker_entry_fn, (tp, WorkerID(i)))
-    # TODO: we might want to take into account Hyper-Threading (HT)
-    #       and allow spawning tasks and pinning to cores that are not HT-siblings.
-    #       This is important for memory-bound workloads (like copy, addition, ...)
-    #       where both sibling cores will compete for L1 and L2 cache, effectively
-    #       halving the memory bandwidth or worse, flushing what the other put in cache.
-    #       Note that while 2x siblings is common, Xeon Phi has 4x Hyper-Threading.
-    when not(defined(cpp) and defined(vcc)):
-      # TODO: Nim casts between Windows Handles but that requires reinterpret cast for C++
-      pinToCpu(tp.workers[i], i)
+
+    if pinToCPU:
+      # TODO: we might want to take into account Hyper-Threading (HT)
+      #       and allow spawning tasks and pinning to cores that are not HT-siblings.
+      #       This is important for memory-bound workloads (like copy, addition, ...)
+      #       where both sibling cores will compete for L1 and L2 cache, effectively
+      #       halving the memory bandwidth or worse, flushing what the other put in cache.
+      #       Note that while 2x siblings is common, Xeon Phi has 4x Hyper-Threading.
+      when not(defined(cpp) and defined(vcc)):
+        # TODO: Nim casts between Windows Handles but that requires reinterpret cast for C++
+        pinToCpu(tp.workers[i], i)
 
   # Root worker
   setupWorker()