diff --git a/src/ARM.cpp b/src/ARM.cpp
index 7072978d..8c74a248 100644
--- a/src/ARM.cpp
+++ b/src/ARM.cpp
@@ -210,6 +210,7 @@ void ARMv5::Reset()
 
     WBWritePointer = 16;
     WBFillPointer = 0;
+    WBDelay = 0;
 
     ARM::Reset();
 }
diff --git a/src/ARM.h b/src/ARM.h
index 12675023..16f4b72a 100644
--- a/src/ARM.h
+++ b/src/ARM.h
@@ -377,6 +377,7 @@ public:
 
     u8 WBWritePointer;
     u8 WBFillPointer;
+    u64 WBDelay;
     u32 WBAddr; // current working address for the write buffer
     u32 storeaddr[16]; // debugging
     u64 WBCycles[16]; // timestamp each write will complete
diff --git a/src/CP15.cpp b/src/CP15.cpp
index c0da4711..1d082f04 100644
--- a/src/CP15.cpp
+++ b/src/CP15.cpp
@@ -1297,10 +1297,11 @@ bool ARMv5::DataWrite8(u32 addr, u8 val)
     }
     else
     {
-        NDS.ARM9Timestamp = NDS.ARM9Timestamp + ((1<<NDS.ARM9ClockShift)-1) & ~((1<<NDS.ARM9ClockShift)-1);
+        if (WBDelay > NDS.ARM9Timestamp) NDS.ARM9Timestamp = WBDelay;
         DataCycles = 1;
         WriteBufferWrite(addr, 3, 1);
         WriteBufferWrite(val, 0, MemTimings[addr >> 12][1], addr);
+        WBDelay = NDS.ARM9Timestamp + 2;
     }
     return true;
 }
@@ -1357,10 +1358,11 @@ bool ARMv5::DataWrite16(u32 addr, u16 val)
     }
     else
     {
-        NDS.ARM9Timestamp = NDS.ARM9Timestamp + ((1<<NDS.ARM9ClockShift)-1) & ~((1<<NDS.ARM9ClockShift)-1);
+        if (WBDelay > NDS.ARM9Timestamp) NDS.ARM9Timestamp = WBDelay;
         DataCycles = 1;
         WriteBufferWrite(addr, 3, 1);
         WriteBufferWrite(val, 1, MemTimings[addr >> 12][1], addr);
+        WBDelay = NDS.ARM9Timestamp + 2;
     }
     return true;
 }
@@ -1417,10 +1419,11 @@ bool ARMv5::DataWrite32(u32 addr, u32 val)
     }
     else
     {
-        NDS.ARM9Timestamp = NDS.ARM9Timestamp + ((1<<NDS.ARM9ClockShift)-1) & ~((1<<NDS.ARM9ClockShift)-1);
+        if (WBDelay > NDS.ARM9Timestamp) NDS.ARM9Timestamp = WBDelay;
         DataCycles = 1;
         WriteBufferWrite(addr, 3, 1);
         WriteBufferWrite(val, 2, MemTimings[addr >> 12][2], addr);
+        WBDelay = NDS.ARM9Timestamp + 2;
     }
     return true;
 }
@@ -1476,9 +1479,9 @@ bool ARMv5::DataWrite32S(u32 addr, u32 val)
     }
     else
     {
-        DataCycles += (((NDS.ARM9Timestamp + DataCycles) + ((1<<NDS.ARM9ClockShift)-1) & ~((1<<NDS.ARM9ClockShift)-1)) - (NDS.ARM9Timestamp + DataCycles));
         DataCycles += 1;
         WriteBufferWrite(val, 2, MemTimings[addr >> 12][3], addr);
+        WBDelay = NDS.ARM9Timestamp + DataCycles + 1;
     }
     return true;
 }