diff --git a/src/GPU.cpp b/src/GPU.cpp
index 1b272cc9..bb3d57fd 100644
--- a/src/GPU.cpp
+++ b/src/GPU.cpp
@@ -30,15 +30,17 @@ namespace melonDS
 using Platform::Log;
 using Platform::LogLevel;
 
-#define LINE_CYCLES  (355*6)
+#define LINE_CYCLES   (355*6)
 #define HBLANK_CYCLES (48+(256*6))
 #define FRAME_CYCLES  (LINE_CYCLES * 263)
+#define READ_CYCLES   (520) // CHECKME: Probably off by a little bit
 
 enum
 {
     LCD_StartHBlank = 0,
     LCD_StartScanline,
     LCD_FinishFrame,
+    LCD_ReadScanline,
 };
 
 
@@ -73,6 +75,8 @@ GPU::GPU(melonDS::NDS& nds, std::unique_ptr<Renderer3D>&& renderer3d, std::uniqu
     NDS.RegisterEventFunc(Event_LCD, LCD_StartHBlank, MemberEventFunc(GPU, StartHBlank));
     NDS.RegisterEventFunc(Event_LCD, LCD_StartScanline, MemberEventFunc(GPU, StartScanline));
     NDS.RegisterEventFunc(Event_LCD, LCD_FinishFrame, MemberEventFunc(GPU, FinishFrame));
+    NDS.RegisterEventFunc(Event_LCD, LCD_ReadScanline, MemberEventFunc(GPU, ReadScanline));
+    NDS.RegisterEventFunc(Event_DisplayFIFO, 0, MemberEventFunc(GPU, DisplayFIFO));
     NDS.RegisterEventFunc(Event_DisplayFIFO, 0, MemberEventFunc(GPU, DisplayFIFO));
 
     InitFramebuffers();
@@ -85,6 +89,7 @@ GPU::~GPU() noexcept
     NDS.UnregisterEventFunc(Event_LCD, LCD_StartHBlank);
     NDS.UnregisterEventFunc(Event_LCD, LCD_StartScanline);
     NDS.UnregisterEventFunc(Event_LCD, LCD_FinishFrame);
+    NDS.UnregisterEventFunc(Event_LCD, LCD_ReadScanline);
     NDS.UnregisterEventFunc(Event_DisplayFIFO, 0);
 }
 
@@ -910,11 +915,10 @@ void GPU::StartHBlank(u32 line) noexcept
 
     if (DispStat[0] & (1<<4)) NDS.SetIRQ(0, IRQ_HBlank);
     if (DispStat[1] & (1<<4)) NDS.SetIRQ(1, IRQ_HBlank);
-
-    if (VCount < 262)
-        NDS.ScheduleEvent(Event_LCD, true, (LINE_CYCLES - HBLANK_CYCLES), LCD_StartScanline, line+1);
+    if (VCount == 262 || VCount < 191) // this is probably wrong, but i haven't dug deep enough to prove it yet
+        NDS.ScheduleEvent(Event_LCD, true, (LINE_CYCLES - HBLANK_CYCLES - READ_CYCLES), LCD_ReadScanline, line);
     else
-        NDS.ScheduleEvent(Event_LCD, true, (LINE_CYCLES - HBLANK_CYCLES), LCD_FinishFrame, line+1);
+        NDS.ScheduleEvent(Event_LCD, true, (LINE_CYCLES - HBLANK_CYCLES), LCD_StartScanline, line+1);
 }
 
 void GPU::FinishFrame(u32 lines) noexcept
@@ -949,6 +953,19 @@ void GPU::BlankFrame() noexcept
     TotalScanlines = 263;
 }
 
+void GPU::ReadScanline(u32 line) noexcept
+{
+    int scanline;
+    scanline = (VCount == 262 ? 0 : (line+1));
+    GPU3D.ScanlineSync(scanline);
+    if (GPU3D.UnderflowFlagVCount == scanline) GPU3D.DispCnt |= (1<<12);
+    
+    if (VCount != 262)
+        NDS.ScheduleEvent(Event_LCD, true, READ_CYCLES, LCD_StartScanline, line+1);
+    else
+        NDS.ScheduleEvent(Event_LCD, true, READ_CYCLES, LCD_FinishFrame, line+1);
+}
+
 void GPU::StartScanline(u32 line) noexcept
 {
     if (line == 0)
@@ -1002,13 +1019,6 @@ void GPU::StartScanline(u32 line) noexcept
             NDS.ScheduleEvent(Event_DisplayFIFO, false, 32, 0, 0);
     }
 
-    if (VCount == GPU3D.UnderflowFlagVCount)
-    {
-        // appears to get set the vcount before the underflow occured?
-        // probably gets updated the instant the underflow happened, which might be annoying to work out with precision.
-        GPU3D.DispCnt |= (1<<12);
-    }
-
     if (VCount == 262)
     {
         // frame end
@@ -1020,7 +1030,7 @@ void GPU::StartScanline(u32 line) noexcept
     {
         if (VCount == 192)
         {
-            // in reality rendering already finishes at line 144
+            // in reality rendering already finishes at line 144 (can take up to ~191 depending on load)
             // and games might already start to modify texture memory.
             // That doesn't matter for us because we cache the entire
             // texture memory anyway and only update it before the start
diff --git a/src/GPU.h b/src/GPU.h
index 780d5e01..e1f4b89d 100644
--- a/src/GPU.h
+++ b/src/GPU.h
@@ -506,6 +506,7 @@ public:
     void BlankFrame() noexcept;
     void StartScanline(u32 line) noexcept;
     void StartHBlank(u32 line) noexcept;
+    void ReadScanline(u32 line) noexcept;
 
     void DisplayFIFO(u32 x) noexcept;
 
diff --git a/src/GPU3D.cpp b/src/GPU3D.cpp
index 4c177e4c..cbd2721c 100644
--- a/src/GPU3D.cpp
+++ b/src/GPU3D.cpp
@@ -2547,6 +2547,10 @@ void GPU3D::SetRenderXPos(u16 xpos) noexcept
     RenderXPos = xpos & 0x01FF;
 }
 
+void GPU3D::ScanlineSync(int line) noexcept
+{
+    CurrentRenderer->ScanlineSync(line);
+}
 
 u32* GPU3D::GetLine(int line) noexcept
 {
diff --git a/src/GPU3D.h b/src/GPU3D.h
index d35894d6..ada40fb1 100644
--- a/src/GPU3D.h
+++ b/src/GPU3D.h
@@ -114,6 +114,7 @@ public:
 
     void SetRenderXPos(u16 xpos) noexcept;
     [[nodiscard]] u16 GetRenderXPos() const noexcept { return RenderXPos; }
+    void ScanlineSync(int line) noexcept;
     u32* GetLine(int line) noexcept;
 
     void WriteToGXFIFO(u32 val) noexcept;
@@ -454,6 +455,7 @@ public:
     virtual void RenderFrame(GPU& gpu) = 0;
     virtual void RestartFrame(GPU& gpu) {};
     virtual u32* GetLine(int line) = 0;
+    virtual void ScanlineSync(int line) {};
     virtual void Blit(const GPU& gpu) {};
     virtual void PrepareCaptureFrame() {}
 protected:
diff --git a/src/GPU3D_Soft.cpp b/src/GPU3D_Soft.cpp
index c3b162d1..d99bbba6 100644
--- a/src/GPU3D_Soft.cpp
+++ b/src/GPU3D_Soft.cpp
@@ -1959,7 +1959,7 @@ void SoftRenderer::FinishPushScanline(s32 y, s32 pixelsremain)
     RasterTiming += std::clamp(ScanlineTimeout - RasterTiming, 0, 12);\
     \
     /* set the underflow flag if one of the scanlines came within 14 cycles of visible underflow */\
-    if ((ScanlineTimeout <= RasterTiming) && (gpu.GPU3D.UnderflowFlagVCount == (u16)-1)) gpu.GPU3D.UnderflowFlagVCount = y-1;
+    if ((ScanlineTimeout <= RasterTiming) && (gpu.GPU3D.UnderflowFlagVCount == (u16)-1)) gpu.GPU3D.UnderflowFlagVCount = y - (y&1 ? 0 : 1);
 
 void SoftRenderer::RenderPolygonsFast(GPU& gpu, Polygon** polygons, int npolys)
 {
@@ -2156,8 +2156,7 @@ void SoftRenderer::RenderThreadFunc(GPU& gpu)
         RenderThreadRendering = false;
     }
 }
-
-u32* SoftRenderer::GetLine(int line)
+void SoftRenderer::ScanlineSync(int line)
 {
     if (RenderThreadRunning.load(std::memory_order_relaxed))
     {
@@ -2167,7 +2166,10 @@ u32* SoftRenderer::GetLine(int line)
             // so we don't need to wait for a specific row)
             Platform::Semaphore_Wait(Sema_ScanlineCount);
     }
+}
 
+u32* SoftRenderer::GetLine(int line)
+{
     return &FinalBuffer[line * ScanlineWidth];
 }
 
diff --git a/src/GPU3D_Soft.h b/src/GPU3D_Soft.h
index 0c4baf79..4b1a8e52 100644
--- a/src/GPU3D_Soft.h
+++ b/src/GPU3D_Soft.h
@@ -40,6 +40,7 @@ public:
     void RenderFrame(GPU& gpu) override;
     void RestartFrame(GPU& gpu) override;
     u32* GetLine(int line) override;
+    void ScanlineSync(int line) override;
 
     void SetupRenderThread(GPU& gpu);
     void EnableRenderThread();