diff --git a/server/TracyWorker.cpp b/server/TracyWorker.cpp
index eb490c18..dc3ac5c2 100644
--- a/server/TracyWorker.cpp
+++ b/server/TracyWorker.cpp
@@ -1495,6 +1495,9 @@ void Worker::Process( const QueueItem& ev )
     case QueueType::ZoneBegin:
         ProcessZoneBegin( ev.zoneBegin );
         break;
+    case QueueType::ZoneBeginCallstack:
+        ProcessZoneBeginCallstack( ev.zoneBegin );
+        break;
     case QueueType::ZoneBeginAllocSrcLoc:
         ProcessZoneBeginAllocSrcLoc( ev.zoneBegin );
         break;
@@ -1585,10 +1588,8 @@ void Worker::Process( const QueueItem& ev )
     }
 }
 
-void Worker::ProcessZoneBegin( const QueueZoneBegin& ev )
+void Worker::ProcessZoneBeginImpl( ZoneEvent* zone, const QueueZoneBegin& ev )
 {
-    auto zone = m_slab.AllocInit<ZoneEvent>();
-
     CheckSourceLocation( ev.srcloc );
 
     zone->start = TscTime( ev.time );
@@ -1602,6 +1603,22 @@ void Worker::ProcessZoneBegin( const QueueZoneBegin& ev )
     NewZone( zone, ev.thread );
 }
 
+void Worker::ProcessZoneBegin( const QueueZoneBegin& ev )
+{
+    auto zone = m_slab.AllocInit<ZoneEvent>();
+    ProcessZoneBeginImpl( zone, ev );
+}
+
+void Worker::ProcessZoneBeginCallstack( const QueueZoneBegin& ev )
+{
+    auto zone = m_slab.AllocInit<ZoneEvent>();
+    ProcessZoneBeginImpl( zone, ev );
+
+    auto& next = m_nextCallstack[ev.thread];
+    next.type = NextCallstackType::Zone;
+    next.zone = zone;
+}
+
 void Worker::ProcessZoneBeginAllocSrcLoc( const QueueZoneBegin& ev )
 {
     auto it = m_pendingSourceLocationPayload.find( ev.srcloc );
diff --git a/server/TracyWorker.hpp b/server/TracyWorker.hpp
index d8f19e82..bc64f4f1 100644
--- a/server/TracyWorker.hpp
+++ b/server/TracyWorker.hpp
@@ -119,6 +119,20 @@ class Worker
         float compRatio;
     };
 
+    enum class NextCallstackType
+    {
+        Zone
+    };
+
+    struct NextCallstack
+    {
+        NextCallstackType type;
+        union
+        {
+            ZoneEvent* zone;
+        };
+    };
+
 public:
     Worker( const char* addr );
     Worker( FileRead& f, EventType::Type eventMask = EventType::All );
@@ -197,6 +211,7 @@ private:
     tracy_force_inline void DispatchProcess( const QueueItem& ev, char*& ptr );
     tracy_force_inline void Process( const QueueItem& ev );
     tracy_force_inline void ProcessZoneBegin( const QueueZoneBegin& ev );
+    tracy_force_inline void ProcessZoneBeginCallstack( const QueueZoneBegin& ev );
     tracy_force_inline void ProcessZoneBeginAllocSrcLoc( const QueueZoneBegin& ev );
     tracy_force_inline void ProcessZoneEnd( const QueueZoneEnd& ev );
     tracy_force_inline void ProcessFrameMark( const QueueFrameMark& ev );
@@ -224,6 +239,8 @@ private:
     tracy_force_inline void ProcessCallstackMemory( const QueueCallstackMemory& ev );
     tracy_force_inline void ProcessCallstackFrame( const QueueCallstackFrame& ev );
 
+    tracy_force_inline void ProcessZoneBeginImpl( ZoneEvent* zone, const QueueZoneBegin& ev );
+
     tracy_force_inline void CheckSourceLocation( uint64_t ptr );
     void NewSourceLocation( uint64_t ptr );
     tracy_force_inline uint32_t ShrinkSourceLocation( uint64_t srcloc );
@@ -306,6 +323,7 @@ private:
     Vector<uint64_t> m_sourceLocationQueue;
     flat_hash_map<uint64_t, uint32_t, nohash<uint64_t>> m_sourceLocationShrink;
     flat_hash_map<uint64_t, ThreadData*, nohash<uint64_t>> m_threadMap;
+    flat_hash_map<uint64_t, NextCallstack, nohash<uint64_t>> m_nextCallstack;
 
     uint32_t m_pendingStrings;
     uint32_t m_pendingThreads;