diff --git a/TracyLua.hpp b/TracyLua.hpp
index 54e0097d..c9a64313 100644
--- a/TracyLua.hpp
+++ b/TracyLua.hpp
@@ -182,10 +182,10 @@ static tracy_force_inline void SendLuaCallstack( lua_State* L, uint32_t depth )
     }
     assert( dst - ptr == spaceNeeded + 2 );
 
-    TracyLfqPrepare( QueueType::CallstackAlloc );
+    TracyQueuePrepare( QueueType::CallstackAlloc );
     MemWrite( &item->callstackAllocFat.ptr, (uint64_t)ptr );
     MemWrite( &item->callstackAllocFat.nativePtr, (uint64_t)Callstack( depth ) );
-    TracyLfqCommit;
+    TracyQueueCommit( callstackAllocFatThread );
 }
 
 static inline int LuaZoneBeginS( lua_State* L )
diff --git a/client/TracyProfiler.cpp b/client/TracyProfiler.cpp
index 07588d4b..5e164cbf 100644
--- a/client/TracyProfiler.cpp
+++ b/client/TracyProfiler.cpp
@@ -2448,6 +2448,29 @@ Profiler::DequeueStatus Profiler::DequeueSerial()
                     tracy_free_fast( (void*)ptr );
                     break;
                 }
+                case QueueType::Callstack:
+                {
+                    ThreadCtxCheckSerial( callstackFatThread );
+                    ptr = MemRead<uint64_t>( &item->callstackFat.ptr );
+                    SendCallstackPayload( ptr );
+                    tracy_free_fast( (void*)ptr );
+                    break;
+                }
+                case QueueType::CallstackAlloc:
+                {
+                    ThreadCtxCheckSerial( callstackAllocFatThread );
+                    ptr = MemRead<uint64_t>( &item->callstackAllocFat.nativePtr );
+                    if( ptr != 0 )
+                    {
+                        CutCallstack( (void*)ptr, "lua_pcall" );
+                        SendCallstackPayload( ptr );
+                        tracy_free_fast( (void*)ptr );
+                    }
+                    ptr = MemRead<uint64_t>( &item->callstackAllocFat.ptr );
+                    SendCallstackAlloc( ptr );
+                    tracy_free_fast( (void*)ptr );
+                    break;
+                }
 #endif
                 default:
                     assert( false );
@@ -3141,11 +3164,12 @@ void Profiler::ReportTopology()
 void Profiler::SendCallstack( int depth, const char* skipBefore )
 {
 #ifdef TRACY_HAS_CALLSTACK
-    TracyLfqPrepare( QueueType::Callstack );
     auto ptr = Callstack( depth );
     CutCallstack( ptr, skipBefore );
+
+    TracyQueuePrepare( QueueType::Callstack );
     MemWrite( &item->callstackFat.ptr, (uint64_t)ptr );
-    TracyLfqCommit;
+    TracyQueueCommit( callstackFatThread );
 #endif
 }
 
diff --git a/client/TracyProfiler.hpp b/client/TracyProfiler.hpp
index 26f146b3..c2f201b1 100644
--- a/client/TracyProfiler.hpp
+++ b/client/TracyProfiler.hpp
@@ -548,9 +548,9 @@ public:
     {
 #ifdef TRACY_HAS_CALLSTACK
         auto ptr = Callstack( depth );
-        TracyLfqPrepare( QueueType::Callstack );
+        TracyQueuePrepare( QueueType::Callstack );
         MemWrite( &item->callstackFat.ptr, (uint64_t)ptr );
-        TracyLfqCommit;
+        TracyQueueCommit( callstackFatThread );
 #endif
     }
 
diff --git a/common/TracyQueue.hpp b/common/TracyQueue.hpp
index b6f24013..629292e8 100644
--- a/common/TracyQueue.hpp
+++ b/common/TracyQueue.hpp
@@ -453,12 +453,22 @@ struct QueueCallstackFat
     uint64_t ptr;
 };
 
+struct QueueCallstackFatThread : public QueueCallstackFat
+{
+    uint32_t thread;
+};
+
 struct QueueCallstackAllocFat
 {
     uint64_t ptr;
     uint64_t nativePtr;
 };
 
+struct QueueCallstackAllocFatThread : public QueueCallstackAllocFat
+{
+    uint32_t thread;
+};
+
 struct QueueCallstackSample
 {
     int64_t time;
@@ -633,7 +643,9 @@ struct QueueItem
         QueueMemFree memFree;
         QueueMemNamePayload memName;
         QueueCallstackFat callstackFat;
+        QueueCallstackFatThread callstackFatThread;
         QueueCallstackAllocFat callstackAllocFat;
+        QueueCallstackAllocFatThread callstackAllocFatThread;
         QueueCallstackSample callstackSample;
         QueueCallstackSampleFat callstackSampleFat;
         QueueCallstackFrameSize callstackFrameSize;