diff --git a/src/pse/bus.cpp b/src/pse/bus.cpp
index e07b1172d..7a9e1b6c0 100644
--- a/src/pse/bus.cpp
+++ b/src/pse/bus.cpp
@@ -3,6 +3,7 @@
 #include "YBaseLib/Log.h"
 #include "YBaseLib/String.h"
 #include "dma.h"
+#include "gpu.h"
 #include <cstdio>
 Log_SetChannel(Bus);
 
@@ -38,7 +39,7 @@ bool Bus::ReadByte(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_
   return result;
 }
 
-bool Bus::ReadWord(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u16* value)
+bool Bus::ReadHalfWord(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u16* value)
 {
   u32 temp = 0;
   const bool result =
@@ -47,7 +48,7 @@ bool Bus::ReadWord(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_
   return result;
 }
 
-bool Bus::ReadDWord(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u32* value)
+bool Bus::ReadWord(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u32* value)
 {
   return DispatchAccess<MemoryAccessType::Read, MemoryAccessSize::Word>(cpu_address, bus_address, *value);
 }
@@ -58,13 +59,13 @@ bool Bus::WriteByte(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus
   return DispatchAccess<MemoryAccessType::Read, MemoryAccessSize::Byte>(cpu_address, bus_address, temp);
 }
 
-bool Bus::WriteWord(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u16 value)
+bool Bus::WriteHalfWord(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u16 value)
 {
   u32 temp = ZeroExtend32(value);
   return DispatchAccess<MemoryAccessType::Read, MemoryAccessSize::HalfWord>(cpu_address, bus_address, temp);
 }
 
-bool Bus::WriteDWord(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u32 value)
+bool Bus::WriteWord(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u32 value)
 {
   return DispatchAccess<MemoryAccessType::Write, MemoryAccessSize::Word>(cpu_address, bus_address, value);
 }
@@ -178,6 +179,20 @@ bool Bus::WriteExpansionRegion2(MemoryAccessSize size, u32 offset, u32 value)
   return DoInvalidAccess(MemoryAccessType::Write, size, EXP2_BASE | offset, EXP2_BASE | offset, value);
 }
 
+bool Bus::DoReadGPU(MemoryAccessSize size, u32 offset, u32& value)
+{
+  Assert(size == MemoryAccessSize::Word);
+  value = m_gpu->ReadRegister(offset);
+  return true;
+}
+
+bool Bus::DoWriteGPU(MemoryAccessSize size, u32 offset, u32 value)
+{
+  Assert(size == MemoryAccessSize::Word);
+  m_gpu->WriteRegister(offset, value);
+  return true;
+}
+
 bool Bus::ReadSPU(MemoryAccessSize size, u32 offset, u32& value)
 {
   if (offset == 0x1AE)
@@ -186,7 +201,7 @@ bool Bus::ReadSPU(MemoryAccessSize size, u32 offset, u32& value)
     return true;
   }
 
-  //return DoInvalidAccess(MemoryAccessType::Write, size, SPU_BASE | offset, SPU_BASE | offset, value);
+  // return DoInvalidAccess(MemoryAccessType::Write, size, SPU_BASE | offset, SPU_BASE | offset, value);
   value = 0;
   return true;
 }
@@ -201,7 +216,7 @@ bool Bus::WriteSPU(MemoryAccessSize size, u32 offset, u32 value)
   if (offset == 0x1AA)
     return true;
 
-  //return DoInvalidAccess(MemoryAccessType::Write, size, SPU_BASE | offset, SPU_BASE | offset, value);
+  // return DoInvalidAccess(MemoryAccessType::Write, size, SPU_BASE | offset, SPU_BASE | offset, value);
   return true;
 }
 
@@ -212,7 +227,7 @@ bool Bus::DoReadDMA(MemoryAccessSize size, u32 offset, u32& value)
   return true;
 }
 
-bool Bus::DoWriteDMA(MemoryAccessSize size, u32 offset, u32& value)
+bool Bus::DoWriteDMA(MemoryAccessSize size, u32 offset, u32 value)
 {
   Assert(size == MemoryAccessSize::Word);
   m_dma->WriteRegister(offset, value);
diff --git a/src/pse/bus.h b/src/pse/bus.h
index ac3b8044a..084ea7a72 100644
--- a/src/pse/bus.h
+++ b/src/pse/bus.h
@@ -20,11 +20,11 @@ public:
   bool DoState(StateWrapper& sw);
 
   bool ReadByte(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u8* value);
-  bool ReadWord(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u16* value);
-  bool ReadDWord(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u32* value);
+  bool ReadHalfWord(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u16* value);
+  bool ReadWord(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u32* value);
   bool WriteByte(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u8 value);
-  bool WriteWord(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u16 value);
-  bool WriteDWord(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u32 value);
+  bool WriteHalfWord(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u16 value);
+  bool WriteWord(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u32 value);
 
   template<MemoryAccessType type, MemoryAccessSize size>
   bool DispatchAccess(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddress bus_address, u32& value);
@@ -33,6 +33,9 @@ private:
   static constexpr u32 DMA_BASE = 0x1F801080;
   static constexpr u32 DMA_SIZE = 0x80;
   static constexpr u32 DMA_MASK = DMA_SIZE - 1;
+  static constexpr u32 GPU_BASE = 0x1F801810;
+  static constexpr u32 GPU_SIZE = 0x10;
+  static constexpr u32 GPU_MASK = GPU_SIZE - 1;
   static constexpr u32 SPU_BASE = 0x1F801C00;
   static constexpr u32 SPU_SIZE = 0x300;
   static constexpr u32 SPU_MASK = 0x3FF;
@@ -54,11 +57,14 @@ private:
   bool ReadExpansionRegion2(MemoryAccessSize size, u32 offset, u32& value);
   bool WriteExpansionRegion2(MemoryAccessSize size, u32 offset, u32 value);
 
-  bool ReadSPU(MemoryAccessSize size, u32 offset, u32& value);
-  bool WriteSPU(MemoryAccessSize size, u32 offset, u32 value);
+  bool DoReadGPU(MemoryAccessSize size, u32 offset, u32& value);
+  bool DoWriteGPU(MemoryAccessSize size, u32 offset, u32 value);
 
   bool DoReadDMA(MemoryAccessSize size, u32 offset, u32& value);
-  bool DoWriteDMA(MemoryAccessSize size, u32 offset, u32& value);
+  bool DoWriteDMA(MemoryAccessSize size, u32 offset, u32 value);
+
+  bool ReadSPU(MemoryAccessSize size, u32 offset, u32& value);
+  bool WriteSPU(MemoryAccessSize size, u32 offset, u32 value);
 
   DMA* m_dma = nullptr;
   GPU* m_gpu = nullptr;
diff --git a/src/pse/bus.inl b/src/pse/bus.inl
index a2bbaf0db..649290b0b 100644
--- a/src/pse/bus.inl
+++ b/src/pse/bus.inl
@@ -89,6 +89,15 @@ bool Bus::DispatchAccess(PhysicalMemoryAddress cpu_address, PhysicalMemoryAddres
     return (type == MemoryAccessType::Read) ? DoReadDMA(size, bus_address & DMA_MASK, value) :
                                               DoWriteDMA(size, bus_address & DMA_MASK, value);
   }
+  else if (bus_address < GPU_BASE)
+  {
+    return DoInvalidAccess(type, size, cpu_address, bus_address, value);
+  }
+  else if (bus_address < (GPU_BASE + GPU_SIZE))
+  {
+    return (type == MemoryAccessType::Read) ? DoReadGPU(size, bus_address & GPU_MASK, value) :
+                                              DoWriteGPU(size, bus_address & GPU_MASK, value);
+  }
   else if (bus_address < SPU_BASE)
   {
     return DoInvalidAccess(type, size, cpu_address, bus_address, value);
diff --git a/src/pse/cpu_core.cpp b/src/pse/cpu_core.cpp
index 550ed116d..d9717cc68 100644
--- a/src/pse/cpu_core.cpp
+++ b/src/pse/cpu_core.cpp
@@ -526,6 +526,32 @@ void Core::ExecuteInstruction(Instruction inst, u32 inst_pc)
     }
     break;
 
+    case InstructionOp::lwl:
+    case InstructionOp::lwr:
+    {
+      const VirtualMemoryAddress addr = ReadReg(inst.i.rs) + inst.i.imm_sext32();
+      const VirtualMemoryAddress aligned_addr = addr & ~UINT32_C(3);
+      const u32 aligned_value = ReadMemoryWord(aligned_addr);
+
+      // note: bypasses load delay on the read
+      const u32 existing_value = m_regs.r[static_cast<u8>(inst.i.rt.GetValue())];
+      const u8 shift = (Truncate8(addr) & u8(3)) * u8(8);
+      u32 new_value;
+      if (inst.op == InstructionOp::lwl)
+      {
+        const u32 mask = UINT32_C(0x00FFFFFF) >> shift;
+        new_value = (existing_value & mask) | (aligned_value << (24 - shift));
+      }
+      else
+      {
+        const u32 mask = UINT32_C(0xFFFFFF00) << (24 - shift);
+        new_value = (existing_value & mask) | (aligned_value >> shift);
+      }
+
+      WriteRegDelayed(inst.i.rt, new_value);
+    }
+    break;
+
     case InstructionOp::sb:
     {
       const VirtualMemoryAddress addr = ReadReg(inst.i.rs) + inst.i.imm_sext32();
@@ -550,6 +576,31 @@ void Core::ExecuteInstruction(Instruction inst, u32 inst_pc)
     }
     break;
 
+    case InstructionOp::swl:
+    case InstructionOp::swr:
+    {
+      const VirtualMemoryAddress addr = ReadReg(inst.i.rs) + inst.i.imm_sext32();
+      const VirtualMemoryAddress aligned_addr = addr & ~UINT32_C(3);
+      const u32 mem_value = ReadMemoryWord(aligned_addr);
+      const u32 reg_value = ReadReg(inst.i.rt);
+      const u8 shift = (Truncate8(addr) & u8(3)) * u8(8);
+
+      u32 new_value;
+      if (inst.op == InstructionOp::swl)
+      {
+        const u32 mem_mask = UINT32_C(0xFFFFFF00) << shift;
+        new_value = (mem_value & mem_mask) | (reg_value >> (24 - shift));
+      }
+      else
+      {
+        const u32 mem_mask = UINT32_C(0x00FFFFFF) >> (24 - shift);
+        new_value = (mem_value & mem_mask) | (reg_value << shift);
+      }
+
+      WriteMemoryWord(aligned_addr, new_value);
+    }
+    break;
+
     case InstructionOp::j:
     {
       Branch((m_regs.pc & UINT32_C(0xF0000000)) | (inst.j.target << 2));
diff --git a/src/pse/cpu_disasm.cpp b/src/pse/cpu_disasm.cpp
index 66a542a6c..ba545b4d1 100644
--- a/src/pse/cpu_disasm.cpp
+++ b/src/pse/cpu_disasm.cpp
@@ -61,19 +61,19 @@ static const std::array<const char*, 64> s_base_table = {{
   "UNKNOWN",               // 31
   "lb $rt, $offsetrs",     // 32
   "lh $rt, $offsetrs",     // 33
-  "UNKNOWN",               // 34
+  "lwl $rt, $offsetrs",    // 34
   "lw $rt, $offsetrs",     // 35
   "lbu $rt, $offsetrs",    // 36
   "lhu $rt, $offsetrs",    // 37
-  "UNKNOWN",               // 38
+  "lwr $rt, $offsetrs",    // 38
   "UNKNOWN",               // 39
   "sb $rt, $offsetrs",     // 40
   "sh $rt, $offsetrs",     // 41
-  "UNKNOWN",               // 42
+  "swl $rt, $offsetrs",    // 42
   "sw $rt, $offsetrs",     // 43
   "UNKNOWN",               // 44
   "UNKNOWN",               // 45
-  "UNKNOWN",               // 46
+  "swr $rt, $offsetrs",    // 46
   "UNKNOWN",               // 47
   "UNKNOWN",               // 48
   "UNKNOWN",               // 49
diff --git a/src/pse/dma.cpp b/src/pse/dma.cpp
index e08dd138d..346a9b3b7 100644
--- a/src/pse/dma.cpp
+++ b/src/pse/dma.cpp
@@ -17,7 +17,7 @@ bool DMA::Initialize(Bus* bus, GPU* gpu)
 void DMA::Reset()
 {
   m_state = {};
-  m_DPCR.bits = 0;
+  m_DPCR.bits = 0x07654321;
   m_DCIR = 0;
 }
 
@@ -52,5 +52,160 @@ u32 DMA::ReadRegister(u32 offset)
 
 void DMA::WriteRegister(u32 offset, u32 value)
 {
+  const u32 channel_index = offset >> 4;
+  if (channel_index < 7)
+  {
+    ChannelState& state = m_state[channel_index];
+    switch (offset & UINT32_C(0x0F))
+    {
+      case 0x00:
+      {
+        state.base_address = value & BASE_ADDRESS_MASK;
+        Log_DebugPrintf("DMA channel %u base address <- 0x%08X", channel_index, state.base_address);
+        return;
+      }
+      case 0x04:
+      {
+        Log_DebugPrintf("DMA channel %u block control <- 0x%08X", channel_index, value);
+        state.block_control.bits = value;
+        return;
+      }
+
+      case 0x08:
+      {
+        state.channel_control.bits = (state.channel_control.bits & ~ChannelState::ChannelControl::WRITE_MASK) |
+                                     (value & ChannelState::ChannelControl::WRITE_MASK);
+        Log_DebugPrintf("DMA channel %u channel control <- 0x%08X", channel_index, state.channel_control.bits);
+        if (CanRunChannel(static_cast<Channel>(channel_index)))
+          RunDMA(static_cast<Channel>(channel_index));
+
+        return;
+      }
+
+      default:
+        break;
+    }
+  }
+  else
+  {
+    switch (offset)
+    {
+      case 0x70:
+      {
+        Log_DebugPrintf("DPCR <- 0x%08X", value);
+        m_DPCR.bits = value;
+        return;
+      }
+
+      case 0x74:
+      {
+        m_DCIR = (m_DCIR & ~DCIR_WRITE_MASK) | (value & DCIR_WRITE_MASK);
+        Log_DebugPrintf("DCIR <- 0x%08X", m_DCIR);
+        return;
+      }
+
+      default:
+        break;
+    }
+  }
+
   Log_ErrorPrintf("Unhandled register write: %02X <- %08X", offset, value);
 }
+
+void DMA::SetRequest(Channel channel, bool request)
+{
+  ChannelState& cs = m_state[static_cast<u32>(channel)];
+  cs.request = request;
+
+  if (CanRunChannel(channel))
+    RunDMA(channel);
+}
+
+bool DMA::CanRunChannel(Channel channel) const
+{
+  if (!m_DPCR.GetMasterEnable(channel))
+    return false;
+
+  const ChannelState& cs = m_state[static_cast<u32>(channel)];
+  if (cs.channel_control.start_trigger)
+    return true;
+
+  return (cs.channel_control.enable_busy && cs.request);
+}
+
+void DMA::RunDMA(Channel channel)
+{
+  ChannelState& cs = m_state[static_cast<u32>(channel)];
+  const PhysicalMemoryAddress memory_address = cs.base_address;
+  const bool copy_to_device = cs.channel_control.copy_to_device;
+  Log_DebugPrintf("Running DMA for channel %u", static_cast<u32>(channel));
+  Assert(Common::IsAlignedPow2(memory_address, 4));
+
+  // start/trigger bit is cleared on beginning of transfer
+  cs.channel_control.start_trigger = false;
+
+  switch (cs.channel_control.sync_mode)
+  {
+    case SyncMode::Manual:
+    {
+      const u32 word_count = cs.block_control.manual.GetWordCount();
+      Log_DebugPrintf(" ... copying %u words %s 0x%08X", word_count, copy_to_device ? "from" : "to", memory_address);
+      if (copy_to_device)
+      {
+        for (u32 i = 0; i < word_count; i++)
+        {
+          u32 memory_value = 0;
+          m_bus->DispatchAccess<MemoryAccessType::Read, MemoryAccessSize::Word>(memory_address, memory_address,
+                                                                                memory_value);
+          DMAWrite(channel, memory_value);
+        }
+      }
+      else
+      {
+        for (u32 i = 0; i < word_count; i++)
+        {
+          u32 memory_value = DMARead(channel);
+          m_bus->DispatchAccess<MemoryAccessType::Write, MemoryAccessSize::Word>(memory_address, memory_address,
+                                                                                 memory_value);
+        }
+      }
+    }
+    break;
+
+    case SyncMode::Request:
+    case SyncMode::LinkedList:
+    default:
+      Panic("Unimplemented sync mode");
+      break;
+  }
+
+  // start/busy bit is cleared on end of transfer
+  cs.channel_control.enable_busy = false;
+}
+
+u32 DMA::DMARead(Channel channel)
+{
+  switch (channel)
+  {
+    case Channel::OTC:
+    {
+      // we just return zeros here.. guessing it's pulled low?
+      return 0;
+    }
+
+    case Channel::MDECin:
+    case Channel::MDECout:
+    case Channel::GPU:
+    case Channel::CDROM:
+    case Channel::SPU:
+    case Channel::PIO:
+    default:
+      Panic("Unhandled DMA channel write");
+      return UINT32_C(0xFFFFFFFF);
+  }
+}
+
+void DMA::DMAWrite(Channel channel, u32 value)
+{
+  Panic("Unhandled DMA channel write");
+}
diff --git a/src/pse/dma.h b/src/pse/dma.h
index 31cd7556d..661a09466 100644
--- a/src/pse/dma.h
+++ b/src/pse/dma.h
@@ -34,18 +34,33 @@ public:
   u32 ReadRegister(u32 offset);
   void WriteRegister(u32 offset, u32 value);
 
+  void SetRequest(Channel channel, bool request);
+
 private:
-  Bus* m_bus = nullptr;
-  GPU* m_gpu = nullptr;
+  static constexpr PhysicalMemoryAddress BASE_ADDRESS_MASK = UINT32_C(0x00FFFFFF);
 
   enum class SyncMode : u32
   {
-    Word = 0,
-    Block = 1,
+    Manual = 0,
+    Request = 1,
     LinkedList = 2,
     Reserved = 3
   };
 
+  // is everything enabled for a channel to operate?
+  bool CanRunChannel(Channel channel) const;
+
+  void RunDMA(Channel channel);
+
+  // from device -> memory
+  u32 DMARead(Channel channel);
+
+  // from memory -> device
+  void DMAWrite(Channel channel, u32 value);
+
+  Bus* m_bus = nullptr;
+  GPU* m_gpu = nullptr;
+
   struct ChannelState
   {
     u32 base_address;
@@ -56,25 +71,35 @@ private:
       struct
       {
         BitField<u32, u32, 0, 16> word_count;
-      } word_mode;
+
+        u32 GetWordCount() const { return (word_count == 0) ? 0x10000 : word_count; }
+      } manual;
       struct
       {
         BitField<u32, u32, 0, 16> block_size;
         BitField<u32, u32, 16, 16> block_count;
-      } block_mode;
+
+        u32 GetBlockSize() const { return (block_size == 0) ? 0x10000 : block_size; }
+        u32 GetBlockCount() const { return (block_count == 0) ? 0x10000 : block_count; }
+      } request;
     } block_control;
 
     union ChannelControl
     {
       u32 bits;
-      BitField<u32, bool, 0, 1> direction_to_ram;
+      BitField<u32, bool, 0, 1> copy_to_device;
       BitField<u32, bool, 1, 1> address_step_forward;
       BitField<u32, bool, 8, 1> chopping_enable;
       BitField<u32, SyncMode, 9, 2> sync_mode;
       BitField<u32, u32, 16, 3> chopping_dma_window_size;
       BitField<u32, u32, 20, 3> chopping_cpu_window_size;
+      BitField<u32, bool, 24, 1> enable_busy;
       BitField<u32, bool, 28, 1> start_trigger;
+
+      static constexpr u32 WRITE_MASK = 0b01110001'01110111'00000111'00000011;
     } channel_control;
+
+    bool request = false;
   };
 
   std::array<ChannelState, NUM_CHANNELS> m_state = {};
@@ -83,8 +108,8 @@ private:
   {
     u32 bits;
 
-    u8 GetPriority(Channel channel) { return ((bits >> (static_cast<u8>(channel) * 4)) & u32(3)); }
-    bool GetMasterEnable(Channel channel)
+    u8 GetPriority(Channel channel) const { return ((bits >> (static_cast<u8>(channel) * 4)) & u32(3)); }
+    bool GetMasterEnable(Channel channel) const
     {
       return ConvertToBoolUnchecked((bits >> (static_cast<u8>(channel) * 4 + 3)) & u32(1));
     }
diff --git a/src/pse/gpu.cpp b/src/pse/gpu.cpp
new file mode 100644
index 000000000..7c7549b90
--- /dev/null
+++ b/src/pse/gpu.cpp
@@ -0,0 +1,65 @@
+#include "gpu.h"
+#include "YBaseLib/Log.h"
+#include "bus.h"
+Log_SetChannel(GPU);
+
+GPU::GPU() = default;
+
+GPU::~GPU() = default;
+
+bool GPU::Initialize(Bus* bus, DMA* dma)
+{
+  m_bus = bus;
+  m_dma = dma;
+  return true;
+}
+
+void GPU::Reset()
+{
+  SoftReset();
+}
+
+void GPU::SoftReset()
+{
+  m_GPUSTAT.bits = 0x14802000;
+}
+
+u32 GPU::ReadRegister(u32 offset)
+{
+  if (offset == 0x00)
+  {
+    // GPUREAD
+    Log_ErrorPrintf("GPUREAD");
+    return 0;
+  }
+  else if (offset == 0x04)
+  {
+    // GPUSTAT
+    return m_GPUSTAT.bits;
+  }
+
+  Log_ErrorPrintf("Unhandled register read: %02X", offset);
+  return UINT32_C(0xFFFFFFFF);
+}
+
+void GPU::WriteRegister(u32 offset, u32 value)
+{
+  if (offset == 0x00)
+    WriteGP0(value);
+  else if (offset == 0x04)
+    WriteGP1(value);
+  else
+    Log_ErrorPrintf("Unhandled register write: %02X <- %08X", offset, value);
+}
+
+void GPU::WriteGP0(u32 value)
+{
+  const u8 command = Truncate8(value >> 24);
+  Log_ErrorPrintf("Unimplemented GP0 command 0x%02X", command);
+}
+
+void GPU::WriteGP1(u32 value)
+{
+  const u8 command = Truncate8(value >> 24);
+  Log_ErrorPrintf("Unimplemented GP1 command 0x%02X", command);
+}
diff --git a/src/pse/gpu.h b/src/pse/gpu.h
new file mode 100644
index 000000000..ff8d94ecb
--- /dev/null
+++ b/src/pse/gpu.h
@@ -0,0 +1,58 @@
+#pragma once
+#include "common/bitfield.h"
+#include "types.h"
+#include <array>
+
+class Bus;
+class DMA;
+
+class GPU
+{
+public:
+  GPU();
+  ~GPU();
+
+  bool Initialize(Bus* bus, DMA* dma);
+  void Reset();
+
+  u32 ReadRegister(u32 offset);
+  void WriteRegister(u32 offset, u32 value);
+
+private:
+  void SoftReset();
+  void WriteGP0(u32 value);
+  void WriteGP1(u32 value);
+
+  Bus* m_bus = nullptr;
+  DMA* m_dma = nullptr;
+
+  union GPUSTAT
+  {
+    u32 bits;
+    BitField<u32, u8, 0, 4> texture_page_x_base;
+    BitField<u32, u8, 4, 1> texture_page_y_base;
+    BitField<u32, u8, 5, 2> semi_transparency;
+    BitField<u32, u8, 7, 2> texture_page_colors;
+    BitField<u32, bool, 9, 1> dither_enable;
+    BitField<u32, bool, 10, 1> draw_to_display_area;
+    BitField<u32, bool, 11, 1> draw_set_mask_bit;
+    BitField<u32, bool, 12, 1> draw_to_masked_pixels;
+    BitField<u32, bool, 13, 1> interlaced_field;
+    BitField<u32, bool, 14, 1> reverse_flag;
+    BitField<u32, bool, 15, 1> texture_disable;
+    BitField<u32, u8, 16, 1> horizontal_resolution_2;
+    BitField<u32, u8, 17, 2> horizontal_resolution_1;
+    BitField<u32, u8, 19, 1> vetical_resolution;
+    BitField<u32, bool, 20, 1> pal_mode;
+    BitField<u32, bool, 21, 1> display_area_color_depth_24;
+    BitField<u32, bool, 22, 1> vertical_interlace;
+    BitField<u32, bool, 23, 1> display_enable;
+    BitField<u32, bool, 24, 1> interrupt_request;
+    BitField<u32, bool, 25, 1> dma_data_request;
+    BitField<u32, bool, 26, 1> ready_to_recieve_cmd;
+    BitField<u32, bool, 27, 1> ready_to_send_vram;
+    BitField<u32, bool, 28, 1> ready_to_recieve_dma;
+    BitField<u32, u8, 29, 2> dma_direction;
+    BitField<u32, bool, 31, 1> drawing_even_line;
+  } m_GPUSTAT = {};
+};
diff --git a/src/pse/pse.vcxproj b/src/pse/pse.vcxproj
index 37a474a77..ac31a0ee2 100644
--- a/src/pse/pse.vcxproj
+++ b/src/pse/pse.vcxproj
@@ -39,6 +39,7 @@
     <ClCompile Include="cpu_core.cpp" />
     <ClCompile Include="cpu_disasm.cpp" />
     <ClCompile Include="dma.cpp" />
+    <ClCompile Include="gpu.cpp" />
     <ClCompile Include="system.cpp" />
   </ItemGroup>
   <ItemGroup>
@@ -47,6 +48,7 @@
     <ClInclude Include="cpu_disasm.h" />
     <ClInclude Include="cpu_types.h" />
     <ClInclude Include="dma.h" />
+    <ClInclude Include="gpu.h" />
     <ClInclude Include="save_state_version.h" />
     <ClInclude Include="system.h" />
     <ClInclude Include="types.h" />
diff --git a/src/pse/pse.vcxproj.filters b/src/pse/pse.vcxproj.filters
index d3c82f4a3..71fa7c85d 100644
--- a/src/pse/pse.vcxproj.filters
+++ b/src/pse/pse.vcxproj.filters
@@ -6,6 +6,7 @@
     <ClCompile Include="cpu_disasm.cpp" />
     <ClCompile Include="bus.cpp" />
     <ClCompile Include="dma.cpp" />
+    <ClCompile Include="gpu.cpp" />
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="types.h" />
@@ -16,6 +17,7 @@
     <ClInclude Include="cpu_disasm.h" />
     <ClInclude Include="bus.h" />
     <ClInclude Include="dma.h" />
+    <ClInclude Include="gpu.h" />
   </ItemGroup>
   <ItemGroup>
     <None Include="cpu_core.inl" />
diff --git a/src/pse/system.cpp b/src/pse/system.cpp
index 7ae996095..3a7842190 100644
--- a/src/pse/system.cpp
+++ b/src/pse/system.cpp
@@ -9,10 +9,13 @@ bool System::Initialize()
   if (!m_cpu.Initialize(&m_bus))
     return false;
 
-  if (!m_bus.Initialize(this, &m_dma, nullptr))
+  if (!m_bus.Initialize(this, &m_dma, &m_gpu))
     return false;
 
-  if (!m_dma.Initialize(&m_bus, nullptr))
+  if (!m_dma.Initialize(&m_bus, &m_gpu))
+    return false;
+
+  if (!m_gpu.Initialize(&m_bus, &m_dma))
     return false;
 
   return true;
@@ -22,6 +25,8 @@ void System::Reset()
 {
   m_cpu.Reset();
   m_bus.Reset();
+  m_dma.Reset();
+  m_gpu.Reset();
 }
 
 void System::RunFrame()
diff --git a/src/pse/system.h b/src/pse/system.h
index bc94af776..c81e23e08 100644
--- a/src/pse/system.h
+++ b/src/pse/system.h
@@ -1,6 +1,7 @@
 #pragma once
 #include "bus.h"
 #include "dma.h"
+#include "gpu.h"
 #include "cpu_core.h"
 #include "types.h"
 
@@ -19,4 +20,5 @@ private:
   CPU::Core m_cpu;
   Bus m_bus;
   DMA m_dma;
+  GPU m_gpu;
 };