diff --git a/rpcs3/Emu/Cell/SPUThread.cpp b/rpcs3/Emu/Cell/SPUThread.cpp
index f0b11d71ff..000eeb4877 100644
--- a/rpcs3/Emu/Cell/SPUThread.cpp
+++ b/rpcs3/Emu/Cell/SPUThread.cpp
@@ -487,9 +487,16 @@ void spu_thread::cpu_stop()
 		if (verify(HERE, group->running--) == 1)
 		{
 			{
-				group->stop_count++;
 				std::lock_guard lock(group->mutex);
+				group->stop_count++;
 				group->run_state = SPU_THREAD_GROUP_STATUS_INITIALIZED;
+
+				if (const auto ppu = std::exchange(group->waiter, nullptr))
+				{
+					// Send exit status directly to the joining thread
+					ppu->gpr[4] = group->join_state;
+					ppu->gpr[5] = group->exit_status;
+				}
 			}
 
 			// Notify on last thread stopped
diff --git a/rpcs3/Emu/Cell/lv2/sys_spu.cpp b/rpcs3/Emu/Cell/lv2/sys_spu.cpp
index 85794fd601..1c84648034 100644
--- a/rpcs3/Emu/Cell/lv2/sys_spu.cpp
+++ b/rpcs3/Emu/Cell/lv2/sys_spu.cpp
@@ -633,9 +633,7 @@ error_code sys_spu_thread_group_join(ppu_thread& ppu, u32 id, vm::ptr<u32> cause
 		return CELL_ESRCH;
 	}
 
-	u32 join_state = 0;
-	s32 exit_value = 0;
-
+	do
 	{
 		std::unique_lock lock(group->mutex);
 
@@ -644,12 +642,25 @@ error_code sys_spu_thread_group_join(ppu_thread& ppu, u32 id, vm::ptr<u32> cause
 			return CELL_ESTAT;
 		}
 
-		if (group->join_state.fetch_or(SPU_TGJSF_IS_JOINING) & SPU_TGJSF_IS_JOINING)
+		if (group->waiter)
 		{
 			// another PPU thread is joining this thread group
 			return CELL_EBUSY;
 		}
 
+		if (group->run_state == SPU_THREAD_GROUP_STATUS_INITIALIZED)
+		{
+			// Already terminated
+			ppu.gpr[4] = group->join_state;
+			ppu.gpr[5] = group->exit_status;
+			break;
+		}
+		else
+		{
+			// Subscribe to receive status in r4-r5
+			group->waiter = &ppu;
+		}
+
 		const u64 last_stop = group->stop_count - !group->running;
 
 		lv2_obj::sleep(ppu);
@@ -663,18 +674,15 @@ error_code sys_spu_thread_group_join(ppu_thread& ppu, u32 id, vm::ptr<u32> cause
 
 			group->cond.wait(lock);
 		}
-
-		join_state = group->join_state;
-		exit_value = group->exit_status;
-		group->join_state &= ~SPU_TGJSF_IS_JOINING;
 	}
+	while (0);
 
 	if (ppu.test_stopped())
 	{
 		return 0;
 	}
 
-	switch (join_state & ~SPU_TGJSF_IS_JOINING)
+	switch (ppu.gpr[4] & (SPU_TGJSF_GROUP_EXIT | SPU_TGJSF_TERMINATED))
 	{
 	case 0:
 	{
@@ -699,7 +707,7 @@ error_code sys_spu_thread_group_join(ppu_thread& ppu, u32 id, vm::ptr<u32> cause
 
 	if (status)
 	{
-		*status = group->exit_status;
+		*status = static_cast<s32>(ppu.gpr[5]);
 	}
 
 	return CELL_OK;
diff --git a/rpcs3/Emu/Cell/lv2/sys_spu.h b/rpcs3/Emu/Cell/lv2/sys_spu.h
index 72b416fe2c..8a90656af0 100644
--- a/rpcs3/Emu/Cell/lv2/sys_spu.h
+++ b/rpcs3/Emu/Cell/lv2/sys_spu.h
@@ -210,7 +210,6 @@ enum : u32
 // SPU Thread Group Join State Flag
 enum : u32
 {
-	SPU_TGJSF_IS_JOINING = (1 << 0),
 	SPU_TGJSF_TERMINATED = (1 << 1), // set if SPU Thread Group is terminated by sys_spu_thread_group_terminate
 	SPU_TGJSF_GROUP_EXIT = (1 << 2), // set if SPU Thread Group is terminated by sys_spu_thread_group_exit
 };
@@ -237,6 +236,7 @@ struct lv2_spu_group
 	atomic_t<u32> running; // Number of running threads
 	cond_variable cond; // used to signal waiting PPU thread
 	atomic_t<u64> stop_count;
+	class ppu_thread* waiter = nullptr;
 
 	std::array<std::shared_ptr<named_thread<spu_thread>>, 256> threads; // SPU Threads
 	std::array<std::pair<sys_spu_image, std::vector<sys_spu_segment>>, 256> imgs; // SPU Images