diff --git a/rpcs3/Emu/RSX/RSXThread.cpp b/rpcs3/Emu/RSX/RSXThread.cpp
index 3b2606b902..0aeef5487b 100644
--- a/rpcs3/Emu/RSX/RSXThread.cpp
+++ b/rpcs3/Emu/RSX/RSXThread.cpp
@@ -763,11 +763,18 @@ namespace rsx
 				sync_point_request.release(false);
 			}
 
-			// Execute backend-local tasks first
-			do_local_task(performance_counters.state);
+			// Update sub-units every 64 cycles. The local handler is invoked for other functions externally on-demand anyway.
+			// This avoids expensive calls to check timestamps which involves reading some values from TLS storage on windows.
+			// If something is going on in the backend that requires an update, set the interrupt bit explicitly.
+			if ((m_cycles_counter++ & 63) == 0 ||
+				m_graphics_state & (rsx::pipeline_state::backend_interrupt_bits))
+			{
+				// Execute backend-local tasks first
+				do_local_task(performance_counters.state);
 
-			// Update sub-units
-			zcull_ctrl->update(this);
+				// Update other sub-units
+				zcull_ctrl->update(this);
+			}
 
 			// Execute FIFO queue
 			run_FIFO();
@@ -2952,6 +2959,8 @@ namespace rsx
 
 				m_invalidated_memory_range = unmap_range;
 			}
+
+			m_graphics_state |= rsx::pipeline_state::backend_interrupt;
 		}
 	}
 
@@ -3143,6 +3152,7 @@ namespace rsx
 
 			async_flip_buffer = buffer;
 			async_flip_requested |= flip_request::emu_requested;
+			m_graphics_state |= rsx::pipeline_state::backend_interrupt;
 		}
 	}
 
diff --git a/rpcs3/Emu/RSX/RSXThread.h b/rpcs3/Emu/RSX/RSXThread.h
index bfd2c7c961..3c0016acc2 100644
--- a/rpcs3/Emu/RSX/RSXThread.h
+++ b/rpcs3/Emu/RSX/RSXThread.h
@@ -132,11 +132,15 @@ namespace rsx
 
 		push_buffer_arrays_dirty = 0x20000,   // Push buffers have data written to them (immediate mode vertex buffers)
 
+		backend_interrupt = 0x80000000,       // Backend interrupt, must serve immediately
+
 		fragment_program_dirty = fragment_program_ucode_dirty | fragment_program_state_dirty,
 		vertex_program_dirty = vertex_program_ucode_dirty | vertex_program_state_dirty,
 		invalidate_pipeline_bits = fragment_program_dirty | vertex_program_dirty,
 		invalidate_zclip_bits = vertex_state_dirty | zclip_config_state_dirty,
 		memory_barrier_bits = framebuffer_reads_dirty,
+		backend_interrupt_bits = memory_barrier_bits | backend_interrupt,
+
 		all_dirty = ~0u
 	};
 
@@ -424,6 +428,7 @@ namespace rsx
 	{
 		u64 timestamp_ctrl = 0;
 		u64 timestamp_subvalue = 0;
+		u64 m_cycles_counter = 0;
 
 		display_flip_info_t m_queued_flip{};
 
diff --git a/rpcs3/Emu/RSX/RSXZCULL.cpp b/rpcs3/Emu/RSX/RSXZCULL.cpp
index f32f5c0946..0621b17697 100644
--- a/rpcs3/Emu/RSX/RSXZCULL.cpp
+++ b/rpcs3/Emu/RSX/RSXZCULL.cpp
@@ -14,7 +14,22 @@ namespace rsx
 		}
 
 		ZCULL_control::~ZCULL_control()
-		{}
+		{
+			std::scoped_lock lock(m_pages_mutex);
+
+			for (auto& block : m_locked_pages)
+			{
+				for (auto& p : block)
+				{
+					if (p.second.prot != utils::protection::rw)
+					{
+						utils::memory_protect(vm::base(p.first), 4096, utils::protection::rw);
+					}
+				}
+
+				block.clear();
+			}
+		}
 
 		void ZCULL_control::set_active(class ::rsx::thread* ptimer, bool state, bool flush_queue)
 		{
diff --git a/rpcs3/Emu/RSX/VK/VKGSRender.cpp b/rpcs3/Emu/RSX/VK/VKGSRender.cpp
index 3155ea3e86..c552666384 100644
--- a/rpcs3/Emu/RSX/VK/VKGSRender.cpp
+++ b/rpcs3/Emu/RSX/VK/VKGSRender.cpp
@@ -804,6 +804,7 @@ bool VKGSRender::on_access_violation(u32 address, bool is_writing)
 
 			g_fxo->get<rsx::dma_manager>().set_mem_fault_flag();
 			m_queue_status |= flush_queue_state::deadlock;
+			m_graphics_state |= rsx::pipeline_state::backend_interrupt;
 
 			// Wait for deadlock to clear
 			while (m_queue_status & flush_queue_state::deadlock)
@@ -1646,6 +1647,9 @@ void VKGSRender::sync_hint(rsx::FIFO_hint hint, void* args)
 
 void VKGSRender::do_local_task(rsx::FIFO_state state)
 {
+	// Clear interrupt bit if set
+	m_graphics_state &= ~rsx::pipeline_state::backend_interrupt;
+
 	if (m_queue_status & flush_queue_state::deadlock)
 	{
 		// Clear offloader deadlock