diff --git a/server/TracyEvent.hpp b/server/TracyEvent.hpp
index 0b65bd6d..2799565c 100644
--- a/server/TracyEvent.hpp
+++ b/server/TracyEvent.hpp
@@ -132,6 +132,7 @@ struct GpuEvent
     int64_t gpuStart;
     int64_t gpuEnd;
     int32_t srcloc;
+    int32_t callstack;
     // All above is read/saved as-is.
 
     uint16_t thread;
diff --git a/server/TracyWorker.cpp b/server/TracyWorker.cpp
index ec6f4b94..f3a7fe8e 100644
--- a/server/TracyWorker.cpp
+++ b/server/TracyWorker.cpp
@@ -1559,6 +1559,9 @@ void Worker::Process( const QueueItem& ev )
     case QueueType::GpuZoneBegin:
         ProcessGpuZoneBegin( ev.gpuZoneBegin );
         break;
+    case QueueType::GpuZoneBeginCallstack:
+        ProcessGpuZoneBeginCallstack( ev.gpuZoneBegin );
+        break;
     case QueueType::GpuZoneEnd:
         ProcessGpuZoneEnd( ev.gpuZoneEnd );
         break;
@@ -1947,7 +1950,7 @@ void Worker::ProcessGpuNewContext( const QueueGpuNewContext& ev )
     m_gpuCtxMap.emplace( ev.context, gpu );
 }
 
-void Worker::ProcessGpuZoneBegin( const QueueGpuZoneBegin& ev )
+void Worker::ProcessGpuZoneBeginImpl( GpuEvent* zone, const QueueGpuZoneBegin& ev )
 {
     auto it = m_gpuCtxMap.find( ev.context );
     assert( it != m_gpuCtxMap.end() );
@@ -1955,13 +1958,12 @@ void Worker::ProcessGpuZoneBegin( const QueueGpuZoneBegin& ev )
 
     CheckSourceLocation( ev.srcloc );
 
-    auto zone = m_slab.AllocInit<GpuEvent>();
-
     zone->cpuStart = TscTime( ev.cpuTime );
     zone->cpuEnd = -1;
     zone->gpuStart = std::numeric_limits<int64_t>::max();
     zone->gpuEnd = -1;
     zone->srcloc = ShrinkSourceLocation( ev.srcloc );
+    zone->callstack = 0;
     zone->thread = CompressThread( ev.thread );
 
     m_data.lastTime = std::max( m_data.lastTime, zone->cpuStart );
@@ -1978,6 +1980,22 @@ void Worker::ProcessGpuZoneBegin( const QueueGpuZoneBegin& ev )
     ctx->queue.push_back( zone );
 }
 
+void Worker::ProcessGpuZoneBegin( const QueueGpuZoneBegin& ev )
+{
+    auto zone = m_slab.AllocInit<GpuEvent>();
+    ProcessGpuZoneBeginImpl( zone, ev );
+}
+
+void Worker::ProcessGpuZoneBeginCallstack( const QueueGpuZoneBegin& ev )
+{
+    auto zone = m_slab.AllocInit<GpuEvent>();
+    ProcessGpuZoneBeginImpl( zone, ev );
+
+    auto& next = m_nextCallstack[ev.thread];
+    next.type = NextCallstackType::Gpu;
+    next.gpu = zone;
+}
+
 void Worker::ProcessGpuZoneEnd( const QueueGpuZoneEnd& ev )
 {
     auto it = m_gpuCtxMap.find( ev.context );
@@ -2171,6 +2189,9 @@ void Worker::ProcessCallstack( const QueueCallstack& ev )
     case NextCallstackType::Zone:
         next.zone->callstack = it->second;
         break;
+    case NextCallstackType::Gpu:
+        next.gpu->callstack = it->second;
+        break;
     default:
         assert( false );
         break;
@@ -2452,7 +2473,7 @@ void Worker::ReadTimeline( FileRead& f, Vector<GpuEvent*>& vec, uint64_t size )
         auto zone = m_slab.AllocInit<GpuEvent>();
         vec.push_back_no_space_check( zone );
 
-        f.Read( zone, sizeof( GpuEvent::cpuStart ) + sizeof( GpuEvent::cpuEnd ) + sizeof( GpuEvent::gpuStart ) + sizeof( GpuEvent::gpuEnd ) + sizeof( GpuEvent::srcloc ) );
+        f.Read( zone, sizeof( GpuEvent::cpuStart ) + sizeof( GpuEvent::cpuEnd ) + sizeof( GpuEvent::gpuStart ) + sizeof( GpuEvent::gpuEnd ) + sizeof( GpuEvent::srcloc ) + sizeof( GpuEvent::callstack ) );
         uint64_t thread;
         f.Read( thread );
         zone->thread = CompressThread( thread );
@@ -2472,6 +2493,7 @@ void Worker::ReadTimelinePre032( FileRead& f, Vector<GpuEvent*>& vec, uint64_t s
 
         f.Read( zone, 36 );
         zone->thread = 0;
+        zone->callstack = 0;
         ReadTimelinePre032( f, zone->child );
     }
 }
@@ -2668,7 +2690,7 @@ void Worker::WriteTimeline( FileWrite& f, const Vector<GpuEvent*>& vec )
 
     for( auto& v : vec )
     {
-        f.Write( v, sizeof( GpuEvent::cpuStart ) + sizeof( GpuEvent::cpuEnd ) + sizeof( GpuEvent::gpuStart ) + sizeof( GpuEvent::gpuEnd ) + sizeof( GpuEvent::srcloc ) );
+        f.Write( v, sizeof( GpuEvent::cpuStart ) + sizeof( GpuEvent::cpuEnd ) + sizeof( GpuEvent::gpuStart ) + sizeof( GpuEvent::gpuEnd ) + sizeof( GpuEvent::srcloc ) + sizeof( GpuEvent::callstack ) );
         uint64_t thread = DecompressThread( v->thread );
         f.Write( &thread, sizeof( thread ) );
         WriteTimeline( f, v->child );
diff --git a/server/TracyWorker.hpp b/server/TracyWorker.hpp
index 5ead1e94..368b2e70 100644
--- a/server/TracyWorker.hpp
+++ b/server/TracyWorker.hpp
@@ -121,7 +121,8 @@ class Worker
 
     enum class NextCallstackType
     {
-        Zone
+        Zone,
+        Gpu
     };
 
     struct NextCallstack
@@ -130,6 +131,7 @@ class Worker
         union
         {
             ZoneEvent* zone;
+            GpuEvent* gpu;
         };
     };
 
@@ -229,6 +231,7 @@ private:
     tracy_force_inline void ProcessMessageLiteral( const QueueMessage& ev );
     tracy_force_inline void ProcessGpuNewContext( const QueueGpuNewContext& ev );
     tracy_force_inline void ProcessGpuZoneBegin( const QueueGpuZoneBegin& ev );
+    tracy_force_inline void ProcessGpuZoneBeginCallstack( const QueueGpuZoneBegin& ev );
     tracy_force_inline void ProcessGpuZoneEnd( const QueueGpuZoneEnd& ev );
     tracy_force_inline void ProcessGpuTime( const QueueGpuTime& ev );
     tracy_force_inline void ProcessGpuResync( const QueueGpuResync& ev );
@@ -241,6 +244,7 @@ private:
     tracy_force_inline void ProcessCallstackFrame( const QueueCallstackFrame& ev );
 
     tracy_force_inline void ProcessZoneBeginImpl( ZoneEvent* zone, const QueueZoneBegin& ev );
+    tracy_force_inline void ProcessGpuZoneBeginImpl( GpuEvent* zone, const QueueGpuZoneBegin& ev );
 
     tracy_force_inline void CheckSourceLocation( uint64_t ptr );
     void NewSourceLocation( uint64_t ptr );