diff --git a/rpcs3/Emu/Cell/lv2/sys_rsx.cpp b/rpcs3/Emu/Cell/lv2/sys_rsx.cpp
index 596d8c1399..09c2f3196b 100644
--- a/rpcs3/Emu/Cell/lv2/sys_rsx.cpp
+++ b/rpcs3/Emu/Cell/lv2/sys_rsx.cpp
@@ -198,17 +198,18 @@ error_code sys_rsx_context_allocate(vm::ptr<u32> context_id, vm::ptr<u64> lpar_d
 	auto &reports = vm::_ref<RsxReports>(vm::cast(*lpar_reports, HERE));
 	std::memset(&reports, 0, sizeof(RsxReports));
 
-	for (int i = 0; i < 64; ++i)
+	for (size_t i = 0; i < std::size(reports.notify); ++i)
 		reports.notify[i].timestamp = -1;
 
-	for (int i = 0; i < 256; ++i)
+	for (size_t i = 0; i < std::size(reports.semaphore); i += 4)
 	{
-		reports.semaphore[i].val = 0x1337C0D3;
-		reports.semaphore[i].zero = 0x1337BABE;
-		reports.semaphore[i].zero2 = 0x1337BEEF1337F001;
+		reports.semaphore[i + 0].val.raw() = 0x1337C0D3;
+		reports.semaphore[i + 1].val.raw() = 0x1337BABE;
+		reports.semaphore[i + 2].val.raw() = 0x1337BEEF;
+		reports.semaphore[i + 3].val.raw() = 0x1337F001;
 	}
 
-	for (int i = 0; i < 2048; ++i)
+	for (size_t i = 0; i < std::size(reports.report); ++i)
 	{
 		reports.report[i].val = 0;
 		reports.report[i].timestamp = -1;
diff --git a/rpcs3/Emu/Cell/lv2/sys_rsx.h b/rpcs3/Emu/Cell/lv2/sys_rsx.h
index 308a71cb05..65b6eca25c 100644
--- a/rpcs3/Emu/Cell/lv2/sys_rsx.h
+++ b/rpcs3/Emu/Cell/lv2/sys_rsx.h
@@ -84,11 +84,9 @@ struct RsxDmaControl
 	be_t<u32> unk1;
 };
 
-struct alignas(16) RsxSemaphore
+struct RsxSemaphore
 {
-	be_t<u32> val;
-	be_t<u32> zero;
-	be_t<u64> zero2;
+	atomic_be_t<u32> val;
 };
 
 struct alignas(16) RsxNotify
@@ -106,7 +104,7 @@ struct alignas(16) RsxReport
 
 struct RsxReports
 {
-	RsxSemaphore semaphore[0x100];
+	RsxSemaphore semaphore[0x400];
 	RsxNotify notify[64];
 	RsxReport report[2048];
 };
diff --git a/rpcs3/Emu/RSX/rsx_methods.cpp b/rpcs3/Emu/RSX/rsx_methods.cpp
index 27e44b40cc..036ec7d54d 100644
--- a/rpcs3/Emu/RSX/rsx_methods.cpp
+++ b/rpcs3/Emu/RSX/rsx_methods.cpp
@@ -55,7 +55,7 @@ namespace rsx
 			rsx->sync_point_request.release(true);
 			const u32 addr = get_address(method_registers.semaphore_offset_406e(), method_registers.semaphore_context_dma_406e(), HERE);
 
-			const auto& sema = vm::_ref<atomic_be_t<u32>>(addr);
+			const auto& sema = vm::_ref<RsxSemaphore>(addr).val;
 
 			// TODO: Remove vblank semaphore hack
 			if (addr == rsx->device_addr + 0x30) return;
@@ -126,7 +126,14 @@ namespace rsx
 		{
 			rsx->sync();
 
-			const u32 offset = method_registers.semaphore_offset_406e() & -16;
+			const u32 offset = method_registers.semaphore_offset_406e();
+
+			if (offset % 4)
+			{
+				rsx_log.warning("NV406E semaphore release is using unaligned semaphore, ignoring. (offset=0x%x)", offset);
+				return;
+			}
+
 			const u32 ctxt = method_registers.semaphore_context_dma_406e();
 
 			// By avoiding doing this on flip's semaphore release
@@ -147,7 +154,7 @@ namespace rsx
 				res = &vm::reservation_lock(addr, 4);
 			}
 
-			vm::_ref<atomic_t<RsxSemaphore>>(addr).store({arg, 0, 0});
+			vm::_ref<RsxSemaphore>(addr).val = arg;
 
 			if (res)
 			{
@@ -225,13 +232,16 @@ namespace rsx
 
 			// lle-gcm likes to inject system reserved semaphores, presumably for system/vsh usage
 			// Avoid calling render to avoid any havoc(flickering) they may cause from invalid flush/write
-			const u32 offset = method_registers.semaphore_offset_4097() & -16;
-			vm::_ref<atomic_t<RsxSemaphore>>(get_address(offset, method_registers.semaphore_context_dma_4097(), HERE)).store(
+			const u32 offset = method_registers.semaphore_offset_4097();
+
+			if (offset % 16)
 			{
-				arg,
-				0,
-				0
-			});
+				rsx_log.error("NV4097 semaphore using unaligned offset, recovering. (offset=0x%x)", offset);
+				rsx->recover_fifo();
+				return;
+			}
+
+			vm::_ref<RsxSemaphore>(get_address(offset, method_registers.semaphore_context_dma_4097(), HERE)).val = arg;
 		}
 
 		void back_end_write_semaphore_release(thread* rsx, u32 _reg, u32 arg)
@@ -240,14 +250,17 @@ namespace rsx
 			g_fxo->get<rsx::dma_manager>()->sync();
 			rsx->sync();
 
-			const u32 offset = method_registers.semaphore_offset_4097() & -16;
-			const u32 val = (arg & 0xff00ff00) | ((arg & 0xff) << 16) | ((arg >> 16) & 0xff);
-			vm::_ref<atomic_t<RsxSemaphore>>(get_address(offset, method_registers.semaphore_context_dma_4097(), HERE)).store(
+			const u32 offset = method_registers.semaphore_offset_4097();
+
+			if (offset % 16)
 			{
-				val,
-				0,
-				0
-			});
+				rsx_log.error("NV4097 semaphore using unaligned offset, recovering. (offset=0x%x)", offset);
+				rsx->recover_fifo();
+				return;
+			}
+
+			const u32 val = (arg & 0xff00ff00) | ((arg & 0xff) << 16) | ((arg >> 16) & 0xff);
+			vm::_ref<RsxSemaphore>(get_address(offset, method_registers.semaphore_context_dma_4097(), HERE)).val = val;
 		}
 
 		/**