diff --git a/Core/gb.h b/Core/gb.h
index 536e6fd..8ec5e1c 100644
--- a/Core/gb.h
+++ b/Core/gb.h
@@ -559,6 +559,8 @@ struct GB_gameboy_internal_s {
         bool lcd_disabled_outside_of_vblank;
         int32_t allowed_pending_cycles;
         uint16_t mode3_batching_length;
+        uint8_t joyp_switching_delay;
+        uint8_t joyp_switch_value;
     )
 
     /* APU */
diff --git a/Core/memory.c b/Core/memory.c
index dd8e88b..6bbc949 100644
--- a/Core/memory.c
+++ b/Core/memory.c
@@ -627,6 +627,10 @@ static uint8_t read_high_memory(GB_gameboy_t *gb, uint16_t addr)
             case GB_IO_JOYP:
                 gb->joyp_accessed = true;
                 GB_timing_sync(gb);
+                if (unlikely(gb->joyp_switching_delay)) {
+                    return (gb->io_registers[addr & 0xFF] & ~0x30) | (gb->joyp_switch_value & 0x30);
+                }
+                return gb->io_registers[addr & 0xFF];
             case GB_IO_TMA:
             case GB_IO_LCDC:
             case GB_IO_SCY:
@@ -1497,6 +1501,14 @@ static void write_high_memory(GB_gameboy_t *gb, uint16_t addr, uint8_t value)
                     GB_update_joyp(gb);
                 }
                 else if ((gb->io_registers[GB_IO_JOYP] & 0x30) != (value & 0x30)) {
+                    if (gb->model < GB_MODEL_SGB) { // DMG only
+                        if (gb->joyp_switching_delay) {
+                            gb->io_registers[GB_IO_JOYP] = (gb->joyp_switch_value & 0xF0) | (gb->io_registers[GB_IO_JOYP] & 0x0F);
+                        }
+                        gb->joyp_switch_value = value;
+                        gb->joyp_switching_delay = 24;
+                        value &= gb->io_registers[GB_IO_JOYP];
+                    }
                     GB_sgb_write(gb, value);
                     gb->io_registers[GB_IO_JOYP] = (value & 0xF0) | (gb->io_registers[GB_IO_JOYP] & 0x0F);
                     GB_update_joyp(gb);
diff --git a/Core/timing.c b/Core/timing.c
index 91191cf..1cd04be 100644
--- a/Core/timing.c
+++ b/Core/timing.c
@@ -449,6 +449,16 @@ void GB_advance_cycles(GB_gameboy_t *gb, uint8_t cycles)
     
     gb->rumble_on_cycles += gb->rumble_strength & 3;
     gb->rumble_off_cycles += (gb->rumble_strength & 3) ^ 3;
+    if (gb->joyp_switching_delay) {
+        if (gb->joyp_switching_delay > cycles) {
+            gb->joyp_switching_delay -= cycles;
+        }
+        else {
+            gb->joyp_switching_delay = 0;
+            gb->io_registers[GB_IO_JOYP] = (gb->joyp_switch_value & 0xF0) | (gb->io_registers[GB_IO_JOYP] & 0x0F);
+            GB_update_joyp(gb);
+        }
+    }
         
     GB_apu_run(gb, false);
     GB_display_run(gb, cycles, false);