diff --git a/src/avx512-16bit-qsort.hpp b/src/avx512-16bit-qsort.hpp
index 6dbe24d..fbe1856 100644
--- a/src/avx512-16bit-qsort.hpp
+++ b/src/avx512-16bit-qsort.hpp
@@ -552,9 +552,7 @@ avx512_qsort_fp16_helper(uint16_t *arr, arrsize_t arrsize)
     bool use_parallel = arrsize > 100000;
 
     if (use_parallel) {
-        // This thread limit was determined experimentally; it may be better for it to be the number of physical cores on the system
-        constexpr int thread_limit = 8;
-        int thread_count = std::min(thread_limit, omp_get_max_threads());
+        int thread_count = xss_get_num_threads();
         arrsize_t task_threshold = std::max((arrsize_t)100000, arrsize / 100);
 
         // We use omp parallel and then omp single to setup the threads that will run the omp task calls in qsort_
diff --git a/src/xss-common-argsort.h b/src/xss-common-argsort.h
index 4f5c30d..6c071c2 100644
--- a/src/xss-common-argsort.h
+++ b/src/xss-common-argsort.h
@@ -629,9 +629,7 @@ X86_SIMD_SORT_INLINE void xss_argsort(T *arr,
         bool use_parallel = arrsize > 10000;
 
         if (use_parallel) {
-            // This thread limit was determined experimentally; it may be better for it to be the number of physical cores on the system
-            constexpr int thread_limit = 8;
-            int thread_count = std::min(thread_limit, omp_get_max_threads());
+            int thread_count = xss_get_num_threads();
             arrsize_t task_threshold
                     = std::max((arrsize_t)10000, arrsize / 100);
 
diff --git a/src/xss-common-includes.h b/src/xss-common-includes.h
index a7c34c1..c36f7db 100644
--- a/src/xss-common-includes.h
+++ b/src/xss-common-includes.h
@@ -85,6 +85,13 @@
 #if defined(XSS_USE_OPENMP) && defined(_OPENMP)
 #define XSS_COMPILE_OPENMP
 #include <omp.h>
+
+// Limit the number of threads to 16: emperically determined, can be probably
+// better tuned at a later stage
+X86_SIMD_SORT_INLINE int xss_get_num_threads()
+{
+    return std::min(16, (int)omp_get_max_threads());
+}
 #endif
 
 template <class... T>
diff --git a/src/xss-common-keyvaluesort.hpp b/src/xss-common-keyvaluesort.hpp
index 1a15de7..3a07e01 100644
--- a/src/xss-common-keyvaluesort.hpp
+++ b/src/xss-common-keyvaluesort.hpp
@@ -610,9 +610,7 @@ X86_SIMD_SORT_INLINE void xss_qsort_kv(
         bool use_parallel = arrsize > 10000;
 
         if (use_parallel) {
-            // This thread limit was determined experimentally; it may be better for it to be the number of physical cores on the system
-            constexpr int thread_limit = 8;
-            int thread_count = std::min(thread_limit, omp_get_max_threads());
+            int thread_count = xss_get_num_threads();
             arrsize_t task_threshold
                     = std::max((arrsize_t)10000, arrsize / 100);
 
diff --git a/src/xss-common-qsort.h b/src/xss-common-qsort.h
index cf4a34a..73b947a 100644
--- a/src/xss-common-qsort.h
+++ b/src/xss-common-qsort.h
@@ -672,9 +672,7 @@ X86_SIMD_SORT_INLINE void xss_qsort(T *arr, arrsize_t arrsize, bool hasnan)
         bool use_parallel = arrsize > 100000;
 
         if (use_parallel) {
-            // This thread limit was determined experimentally; it may be better for it to be the number of physical cores on the system
-            constexpr int thread_limit = 8;
-            int thread_count = std::min(thread_limit, omp_get_max_threads());
+            int thread_count = xss_get_num_threads();
             arrsize_t task_threshold
                     = std::max((arrsize_t)100000, arrsize / 100);