diff --git a/Source/Core/Core/HW/EXI.cpp b/Source/Core/Core/HW/EXI.cpp
index ab11dcfdc6..a6bdcc392d 100644
--- a/Source/Core/Core/HW/EXI.cpp
+++ b/Source/Core/Core/HW/EXI.cpp
@@ -97,6 +97,11 @@ void ChangeDevice(const u8 channel, const TEXIDevices device_type, const u8 devi
 	CoreTiming::ScheduleEvent_Threadsafe(500000000, changeDevice, ((u64)channel << 32) | ((u64)device_type << 16) | device_num);
 }
 
+CEXIChannel* GetChannel(u32 index)
+{
+	return g_Channels[index];
+}
+
 IEXIDevice* FindDevice(TEXIDevices device_type, int customIndex)
 {
 	for (auto& channel : g_Channels)
diff --git a/Source/Core/Core/HW/EXI.h b/Source/Core/Core/HW/EXI.h
index 23cda914d3..c54118716e 100644
--- a/Source/Core/Core/HW/EXI.h
+++ b/Source/Core/Core/HW/EXI.h
@@ -31,6 +31,9 @@ void UpdateInterrupts();
 
 void ChangeDeviceCallback(u64 userdata, int cyclesLate);
 void ChangeDevice(const u8 channel, const TEXIDevices device_type, const u8 device_num);
+
+CEXIChannel* GetChannel(u32 index);
+
 IEXIDevice* FindDevice(TEXIDevices device_type, int customIndex=-1);
 
 } // end of namespace ExpansionInterface
diff --git a/Source/Core/Core/HW/EXI_Channel.cpp b/Source/Core/Core/HW/EXI_Channel.cpp
index 9cb72c7db0..4822aedf37 100644
--- a/Source/Core/Core/HW/EXI_Channel.cpp
+++ b/Source/Core/Core/HW/EXI_Channel.cpp
@@ -31,7 +31,7 @@ CEXIChannel::CEXIChannel(u32 ChannelId) :
 		m_Status.CHIP_SELECT = 1;
 
 	for (auto& device : m_pDevices)
-		device.reset(EXIDevice_Create(EXIDEVICE_NONE, this));
+		device.reset(EXIDevice_Create(EXIDEVICE_NONE, m_ChannelId));
 
 	updateInterrupts = CoreTiming::RegisterEvent("EXIInterrupt", UpdateInterrupts);
 }
@@ -137,7 +137,8 @@ void CEXIChannel::RegisterMMIO(MMIO::Mapping* mmio, u32 base)
 
 				m_Control.TSTART = 0;
 
-				if (pDevice->m_deviceType != EXIDEVICE_MEMORYCARD)
+				// Check if device needs specific timing, otherwise just complete transfer immediately
+				if (!pDevice->UseDelayedTransferCompletion())
 					SendTransferComplete();
 			}
 		})
@@ -163,7 +164,7 @@ void CEXIChannel::RemoveDevices()
 
 void CEXIChannel::AddDevice(const TEXIDevices device_type, const int device_num)
 {
-	IEXIDevice* pNewDevice = EXIDevice_Create(device_type, this);
+	IEXIDevice* pNewDevice = EXIDevice_Create(device_type, m_ChannelId);
 	AddDevice(pNewDevice, device_num);
 }
 
@@ -235,7 +236,7 @@ void CEXIChannel::DoState(PointerWrap &p)
 		IEXIDevice* pDevice = m_pDevices[d].get();
 		TEXIDevices type = pDevice->m_deviceType;
 		p.Do(type);
-		IEXIDevice* pSaveDevice = (type == pDevice->m_deviceType) ? pDevice : EXIDevice_Create(type, this);
+		IEXIDevice* pSaveDevice = (type == pDevice->m_deviceType) ? pDevice : EXIDevice_Create(type, m_ChannelId);
 		pSaveDevice->DoState(p);
 		if (pSaveDevice != pDevice)
 		{
diff --git a/Source/Core/Core/HW/EXI_Channel.h b/Source/Core/Core/HW/EXI_Channel.h
index 76cabf2d47..c106a5759d 100644
--- a/Source/Core/Core/HW/EXI_Channel.h
+++ b/Source/Core/Core/HW/EXI_Channel.h
@@ -80,9 +80,13 @@ private:
 
 	std::unique_ptr<IEXIDevice> m_pDevices[NUM_DEVICES];
 
+	// Since channels operate a bit differently from each other
+	u32 m_ChannelId;
+
 	int updateInterrupts;
 
 	static void UpdateInterrupts(u64 userdata, int cyclesLate);
+
 public:
 	// get device
 	IEXIDevice* GetDevice(const u8 _CHIP_SELECT);
@@ -107,7 +111,4 @@ public:
 
 	// This should only be used to transition interrupts from SP1 to Channel 2
 	void SetEXIINT(bool exiint) { m_Status.EXIINT = !!exiint; }
-
-	// Since channels operate a bit differently from each other
-	u32 m_ChannelId;
 };
diff --git a/Source/Core/Core/HW/EXI_Device.cpp b/Source/Core/Core/HW/EXI_Device.cpp
index 60bddbb80b..dbd2807112 100644
--- a/Source/Core/Core/HW/EXI_Device.cpp
+++ b/Source/Core/Core/HW/EXI_Device.cpp
@@ -5,7 +5,6 @@
 
 #include "Core/ConfigManager.h"
 #include "Core/Core.h"
-#include "Core/HW/EXI_Channel.h"
 #include "Core/HW/EXI_Device.h"
 #include "Core/HW/EXI_DeviceAD16.h"
 #include "Core/HW/EXI_DeviceAMBaseboard.h"
@@ -89,7 +88,7 @@ public:
 
 
 // F A C T O R Y
-IEXIDevice* EXIDevice_Create(TEXIDevices device_type, CEXIChannel* channel)
+IEXIDevice* EXIDevice_Create(TEXIDevices device_type, const int channel_num)
 {
 	IEXIDevice* result = nullptr;
 
@@ -103,8 +102,7 @@ IEXIDevice* EXIDevice_Create(TEXIDevices device_type, CEXIChannel* channel)
 	case EXIDEVICE_MEMORYCARDFOLDER:
 	{
 		bool gci_folder = (device_type == EXIDEVICE_MEMORYCARDFOLDER);
-		device_type = EXIDEVICE_MEMORYCARD;
-		result = new CEXIMemoryCard(channel->m_ChannelId, gci_folder);
+		result = new CEXIMemoryCard(channel_num, gci_folder);
 		break;
 	}
 	case EXIDEVICE_MASKROM:
@@ -116,7 +114,7 @@ IEXIDevice* EXIDevice_Create(TEXIDevices device_type, CEXIChannel* channel)
 		break;
 
 	case EXIDEVICE_MIC:
-		result = new CEXIMic(channel->m_ChannelId);
+		result = new CEXIMic(channel_num);
 		break;
 
 	case EXIDEVICE_ETH:
@@ -140,7 +138,5 @@ IEXIDevice* EXIDevice_Create(TEXIDevices device_type, CEXIChannel* channel)
 	if (result != nullptr)
 		result->m_deviceType = device_type;
 
-	result->m_channel = channel;
-
 	return result;
 }
diff --git a/Source/Core/Core/HW/EXI_Device.h b/Source/Core/Core/HW/EXI_Device.h
index d9e388b8fa..b09732eb06 100644
--- a/Source/Core/Core/HW/EXI_Device.h
+++ b/Source/Core/Core/HW/EXI_Device.h
@@ -7,8 +7,6 @@
 #include "Common/ChunkFile.h"
 #include "Common/CommonTypes.h"
 
-class CEXIChannel;
-
 enum TEXIDevices
 {
 	EXIDEVICE_DUMMY,
@@ -40,6 +38,8 @@ public:
 	virtual void DMAWrite(u32 _uAddr, u32 _uSize);
 	virtual void DMARead (u32 _uAddr, u32 _uSize);
 
+	virtual bool UseDelayedTransferCompletion() {return false;}
+
 	virtual bool IsPresent() {return false;}
 	virtual void SetCS(int) {}
 	virtual void DoState(PointerWrap&) {}
@@ -50,12 +50,9 @@ public:
 	virtual bool IsInterruptSet() {return false;}
 	virtual ~IEXIDevice() {}
 
-	// Pointer to channel attached to the device
-	CEXIChannel* m_channel;
-
 	// for savestates. storing it here seemed cleaner than requiring each implementation to report its type.
 	// I know this class is set up like an interface, but no code requires it to be strictly such.
 	TEXIDevices m_deviceType;
 };
 
-IEXIDevice* EXIDevice_Create(const TEXIDevices device_type, CEXIChannel* channel);
+IEXIDevice* EXIDevice_Create(const TEXIDevices device_type, const int channel_num);
diff --git a/Source/Core/Core/HW/EXI_DeviceMemoryCard.cpp b/Source/Core/Core/HW/EXI_DeviceMemoryCard.cpp
index 618c4e441e..a49fa79c4f 100644
--- a/Source/Core/Core/HW/EXI_DeviceMemoryCard.cpp
+++ b/Source/Core/Core/HW/EXI_DeviceMemoryCard.cpp
@@ -71,7 +71,7 @@ CEXIMemoryCard::CEXIMemoryCard(const int index, bool gciFolder)
 	// we're potentially leaking events here, since there's no UnregisterEvent until emu shutdown, but I guess it's inconsequential
 	et_this_card = CoreTiming::RegisterEvent((index == 0) ? "memcardFlushA" : "memcardFlushB", FlushCallback);
 	et_cmd_done = CoreTiming::RegisterEvent((index == 0) ? "memcardDoneA" : "memcardDoneB", CmdDoneCallback);
-	et_transfer_complete = CoreTiming::RegisterEvent((index == 0) ? "memcardTransferCompleteB" : "memcardTransferCompleteB", TransferCompleteCallback);
+	et_transfer_complete = CoreTiming::RegisterEvent((index == 0) ? "memcardTransferCompleteA" : "memcardTransferCompleteB", TransferCompleteCallback);
 
 	interruptSwitch = 0;
 	m_bInterruptSet = 0;
@@ -201,6 +201,11 @@ CEXIMemoryCard::~CEXIMemoryCard()
 	memorycard.reset();
 }
 
+bool CEXIMemoryCard::UseDelayedTransferCompletion()
+{
+	return true;
+}
+
 bool CEXIMemoryCard::IsPresent()
 {
 	return true;
@@ -218,12 +223,7 @@ void CEXIMemoryCard::CmdDone()
 void CEXIMemoryCard::TransferComplete()
 {
 	// Transfer complete, send interrupt
-	m_channel->SendTransferComplete();
-
-	// Page written to memory card, not just to buffer - let's schedule a flush 0.5b cycles into the future (1 sec)
-	// But first we unschedule already scheduled flushes - no point in flushing once per page for a large write.
-	CoreTiming::RemoveEvent(et_this_card);
-	CoreTiming::ScheduleEvent(500000000, et_this_card, (u64)card_index);
+	ExpansionInterface::GetChannel(card_index)->SendTransferComplete();
 }
 
 void CEXIMemoryCard::CmdDoneLater(u64 cycles)
@@ -500,6 +500,13 @@ IEXIDevice* CEXIMemoryCard::FindDevice(TEXIDevices device_type, int customIndex)
 void CEXIMemoryCard::DMARead(u32 _uAddr, u32 _uSize)
 {
 	memorycard->Read(address, _uSize, Memory::GetPointer(_uAddr));
+
+#ifdef _DEBUG
+	if ((address + _uSize) % BLOCK_SIZE == 0)
+		INFO_LOG(EXPANSIONINTERFACE, "reading from block: %x", address / BLOCK_SIZE);
+#endif
+
+	// Schedule transfer complete later based on read speed
 	CoreTiming::ScheduleEvent(_uSize * (SystemTimers::GetTicksPerSecond() / MC_TRANSFER_RATE_READ), et_transfer_complete, (u64)card_index);
 }
 
@@ -508,5 +515,22 @@ void CEXIMemoryCard::DMARead(u32 _uAddr, u32 _uSize)
 void CEXIMemoryCard::DMAWrite(u32 _uAddr, u32 _uSize)
 {
 	memorycard->Write(address, _uSize, Memory::GetPointer(_uAddr));
+
+	// At the end of writing to a block flush to disk
+	// memory card blocks are always(?) written as a whole,
+	// but the dma calls are by page size (0x200) at a time
+	// just in case this is the last block that the game will be writing for a while
+	if (((address + _uSize) % BLOCK_SIZE) == 0)
+	{
+		INFO_LOG(EXPANSIONINTERFACE, "writing to block: %x", address / BLOCK_SIZE);
+		// Page written to memory card, not just to buffer - let's schedule a flush 0.5b cycles into the future (1 sec)
+		// But first we unschedule already scheduled flushes - no point in flushing once per page for a large write
+		// Scheduling event is mainly for raw memory cards as the flush the whole 16MB to disk
+		// Flushing the gci folder is free in comparison
+		CoreTiming::RemoveEvent(et_this_card);
+		CoreTiming::ScheduleEvent(500000000, et_this_card, (u64)card_index);
+	}
+
+	// Schedule transfer complete later based on write speed
 	CoreTiming::ScheduleEvent(_uSize * (SystemTimers::GetTicksPerSecond() / MC_TRANSFER_RATE_WRITE), et_transfer_complete, (u64)card_index);
 }
diff --git a/Source/Core/Core/HW/EXI_DeviceMemoryCard.h b/Source/Core/Core/HW/EXI_DeviceMemoryCard.h
index c1bb1af792..02e5d4d7a1 100644
--- a/Source/Core/Core/HW/EXI_DeviceMemoryCard.h
+++ b/Source/Core/Core/HW/EXI_DeviceMemoryCard.h
@@ -13,6 +13,7 @@ public:
 	virtual ~CEXIMemoryCard();
 	void SetCS(int cs) override;
 	bool IsInterruptSet() override;
+	bool UseDelayedTransferCompletion() override;
 	bool IsPresent() override;
 	void DoState(PointerWrap &p) override;
 	void PauseAndLock(bool doLock, bool unpauseOnUnlock=true) override;