diff --git a/Source/Core/Common/CPUDetect.h b/Source/Core/Common/CPUDetect.h
index 4e619c5b8a..f789ba41a2 100644
--- a/Source/Core/Common/CPUDetect.h
+++ b/Source/Core/Common/CPUDetect.h
@@ -42,6 +42,9 @@ struct CPUInfo
   bool bAVX2 = false;
   bool bBMI1 = false;
   bool bBMI2 = false;
+  // PDEP and PEXT are ridiculously slow on AMD Zen, so we have this flag to avoid using them there
+  // Zen 2 is also affected by this issue
+  bool bFastBMI2 = false;
   bool bFMA = false;
   bool bFMA4 = false;
   bool bAES = false;
@@ -54,6 +57,7 @@ struct CPUInfo
   bool bLAHFSAHF64 = false;
   bool bLongMode = false;
   bool bAtom = false;
+  bool bZen = false;
 
   // ARMv8 specific
   bool bFP = false;
diff --git a/Source/Core/Common/x64CPUDetect.cpp b/Source/Core/Common/x64CPUDetect.cpp
index b66ea45cfa..338c74b808 100644
--- a/Source/Core/Common/x64CPUDetect.cpp
+++ b/Source/Core/Common/x64CPUDetect.cpp
@@ -118,6 +118,9 @@ void CPUInfo::Detect()
         (model == 0x1C || model == 0x26 || model == 0x27 || model == 0x35 || model == 0x36 ||
          model == 0x37 || model == 0x4A || model == 0x4D || model == 0x5A || model == 0x5D))
       bAtom = true;
+    // Detect AMD Zen (all models)
+    if (family == 23)
+      bZen = true;
     logical_cpu_count = (cpu_id[1] >> 16) & 0xFF;
     ht = (cpu_id[3] >> 28) & 1;
 
@@ -172,6 +175,7 @@ void CPUInfo::Detect()
   }
 
   bFlushToZero = bSSE;
+  bFastBMI2 = bBMI2 && !bZen;
 
   if (max_ex_fn >= 0x80000004)
   {
diff --git a/Source/Core/Core/PowerPC/Jit64Common/Jit64AsmCommon.cpp b/Source/Core/Core/PowerPC/Jit64Common/Jit64AsmCommon.cpp
index ab06681a88..57fb829688 100644
--- a/Source/Core/Core/PowerPC/Jit64Common/Jit64AsmCommon.cpp
+++ b/Source/Core/Core/PowerPC/Jit64Common/Jit64AsmCommon.cpp
@@ -68,7 +68,7 @@ void CommonAsmRoutines::GenConvertDoubleToSingle()
 
   // Don't Denormalize
 
-  if (cpu_info.bBMI2)
+  if (cpu_info.bFastBMI2)
   {
     // Extract bits 0-1 and 5-34
     MOV(64, R(RSCRATCH), Imm64(0xc7ffffffe0000000));
diff --git a/Source/Core/VideoCommon/VertexLoaderX64.cpp b/Source/Core/VideoCommon/VertexLoaderX64.cpp
index c29981b9ec..96832e3449 100644
--- a/Source/Core/VideoCommon/VertexLoaderX64.cpp
+++ b/Source/Core/VideoCommon/VertexLoaderX64.cpp
@@ -284,7 +284,7 @@ void VertexLoaderX64::ReadColor(OpArg data, u64 attribute, int format)
     //                   RRRRRGGG GGGBBBBB
     // AAAAAAAA BBBBBBBB GGGGGGGG RRRRRRRR
     LoadAndSwap(16, scratch1, data);
-    if (cpu_info.bBMI1 && cpu_info.bBMI2)
+    if (cpu_info.bBMI1 && cpu_info.bFastBMI2)
     {
       MOV(32, R(scratch2), Imm32(0x07C3F7C0));
       PDEP(32, scratch3, scratch1, R(scratch2));
@@ -324,7 +324,7 @@ void VertexLoaderX64::ReadColor(OpArg data, u64 attribute, int format)
     //                   RRRRGGGG BBBBAAAA
     // AAAAAAAA BBBBBBBB GGGGGGGG RRRRRRRR
     LoadAndSwap(16, scratch1, data);
-    if (cpu_info.bBMI2)
+    if (cpu_info.bFastBMI2)
     {
       MOV(32, R(scratch2), Imm32(0x0F0F0F0F));
       PDEP(32, scratch1, scratch1, R(scratch2));
@@ -353,7 +353,7 @@ void VertexLoaderX64::ReadColor(OpArg data, u64 attribute, int format)
     // AAAAAAAA BBBBBBBB GGGGGGGG RRRRRRRR
     data.AddMemOffset(-1);  // subtract one from address so we can use a 32bit load and bswap
     LoadAndSwap(32, scratch1, data);
-    if (cpu_info.bBMI2)
+    if (cpu_info.bFastBMI2)
     {
       MOV(32, R(scratch2), Imm32(0xFCFCFCFC));
       PDEP(32, scratch1, scratch1, R(scratch2));