diff --git a/Source/Core/VideoCommon/CPUCullImpl.h b/Source/Core/VideoCommon/CPUCullImpl.h
diff --git a/Source/Core/VideoCommon/ConstantManager.h b/Source/Core/VideoCommon/ConstantManager.h
@@ -17,7 +17,7 @@ enum class SrcBlendFactor : u32;
 enum class ZTexOp : u32;
 enum class LogicOp : u32;
 
-struct PixelShaderConstants
+struct alignas(16) PixelShaderConstants
 {
   std::array<int4, 4> colors;
   std::array<int4, 4> kcolors;
@@ -60,7 +60,7 @@ struct PixelShaderConstants
   LogicOp logic_op_mode;
 };
 
-struct VertexShaderConstants
+struct alignas(16) VertexShaderConstants
 {
   u32 components;           // .x
   u32 xfmem_dualTexInfo;    // .y
@@ -109,7 +109,7 @@ enum class VSExpand : u32
   Line,
 };
 
-struct GeometryShaderConstants
+struct alignas(16) GeometryShaderConstants
 {
   float4 stereoparams;
   float4 lineptparams;

diff --git a/Source/Core/VideoCommon/VertexLoaderManager.cpp b/Source/Core/VideoCommon/VertexLoaderManager.cpp
@@ -30,6 +30,7 @@
 #include "VideoCommon/VertexLoaderBase.h"
 #include "VideoCommon/VertexManagerBase.h"
 #include "VideoCommon/VertexShaderManager.h"
+#include "VideoCommon/VideoConfig.h"
 #include "VideoCommon/XFMemory.h"
 
 namespace VertexLoaderManager
@@ -366,17 +367,33 @@ int RunVertices(int vtx_attr_group, OpcodeDecoder::Primitive primitive, int coun
     vertex_shader_manager.SetVertexFormat(loader->m_native_components,
                                           loader->m_native_vertex_format->GetVertexDeclaration());
 
+    // CPUCull's performance increase comes from encoding fewer GPU commands, not sending less data
+    // Therefore it's only useful to check if culling could remove a flush
+    const bool can_cpu_cull = g_ActiveConfig.bCPUCull &&
+                              primitive < OpcodeDecoder::Primitive::GX_DRAW_LINES &&
+                              !g_vertex_manager->HasSendableVertices();
+
     // if cull mode is CULL_ALL, tell VertexManager to skip triangles and quads.
-    // They still need to go through vertex loading, because we need to calculate a zfreeze refrence
-    // slope.
-    bool cullall = (bpmem.genMode.cullmode == CullMode::All &&
-                    primitive < OpcodeDecoder::Primitive::GX_DRAW_LINES);
+    // They still need to go through vertex loading, because we need to calculate a zfreeze
+    // reference slope.
+    const bool cullall = (bpmem.genMode.cullmode == CullMode::All &&
+                          primitive < OpcodeDecoder::Primitive::GX_DRAW_LINES);
 
-    DataReader dst = g_vertex_manager->PrepareForAdditionalData(
-        primitive, count, loader->m_native_vtx_decl.stride, cullall);
+    const int stride = loader->m_native_vtx_decl.stride;
+    DataReader dst = g_vertex_manager->PrepareForAdditionalData(primitive, count, stride,
+                                                                cullall || can_cpu_cull);
 
     count = loader->RunVertices(src, dst.GetPointer(), count);
 
+    if (can_cpu_cull && !cullall)
+    {
+      if (!g_vertex_manager->AreAllVerticesCulled(loader, primitive, dst.GetPointer(), count))
+      {
+        DataReader new_dst = g_vertex_manager->DisableCullAll(stride);
+        memmove(new_dst.GetPointer(), dst.GetPointer(), count * stride);
+      }
+    }
+
     g_vertex_manager->AddIndices(primitive, count);
     g_vertex_manager->FlushData(count, loader->m_native_vtx_decl.stride);
 

diff --git a/Source/Core/VideoCommon/VertexManagerBase.cpp b/Source/Core/VideoCommon/VertexManagerBase.cpp
@@ -104,6 +104,7 @@ VertexManagerBase::~VertexManagerBase() = default;
 bool VertexManagerBase::Initialize()
 {
   m_index_generator.Init();
+  m_cpu_cull.Init();
   return true;
 }
 
@@ -117,6 +118,13 @@ void VertexManagerBase::AddIndices(OpcodeDecoder::Primitive primitive, u32 num_v
   m_index_generator.AddIndices(primitive, num_vertices);
 }
 
+bool VertexManagerBase::AreAllVerticesCulled(VertexLoaderBase* loader,
+                                             OpcodeDecoder::Primitive primitive, const u8* src,
+                                             u32 count)
+{
+  return m_cpu_cull.AreAllVerticesCulled(loader, primitive, src, count);
+}
+
 DataReader VertexManagerBase::PrepareForAdditionalData(OpcodeDecoder::Primitive primitive,
                                                        u32 count, u32 stride, bool cullall)
 {
@@ -187,6 +195,16 @@ DataReader VertexManagerBase::PrepareForAdditionalData(OpcodeDecoder::Primitive
   return DataReader(m_cur_buffer_pointer, m_end_buffer_pointer);
 }
 
+DataReader VertexManagerBase::DisableCullAll(u32 stride)
+{
+  if (m_cull_all)
+  {
+    m_cull_all = false;
+    ResetBuffer(stride);
+  }
+  return DataReader(m_cur_buffer_pointer, m_end_buffer_pointer);
+}
+
 void VertexManagerBase::FlushData(u32 count, u32 stride)
 {
   m_cur_buffer_pointer += count * stride;
@@ -548,6 +566,8 @@ void VertexManagerBase::Flush()
     // Now the vertices can be flushed to the GPU. Everything following the CommitBuffer() call
     // must be careful to not upload any utility vertices, as the binding will be lost otherwise.
     const u32 num_indices = m_index_generator.GetIndexLen();
+    if (num_indices == 0)
+      return;
     u32 base_vertex, base_index;
     CommitBuffer(m_index_generator.GetNumVerts(),
                  VertexLoaderManager::GetCurrentVertexFormat()->GetVertexStride(), num_indices,

diff --git a/Source/Core/VideoCommon/VertexManagerBase.h b/Source/Core/VideoCommon/VertexManagerBase.h
@@ -9,6 +9,7 @@
 #include "Common/BitSet.h"
 #include "Common/CommonTypes.h"
 #include "Common/MathUtil.h"
+#include "VideoCommon/CPUCull.h"
 #include "VideoCommon/IndexGenerator.h"
 #include "VideoCommon/RenderState.h"
 #include "VideoCommon/ShaderCache.h"
@@ -100,11 +101,18 @@ class VertexManagerBase
 
   PrimitiveType GetCurrentPrimitiveType() const { return m_current_primitive_type; }
   void AddIndices(OpcodeDecoder::Primitive primitive, u32 num_vertices);
+  bool AreAllVerticesCulled(VertexLoaderBase* loader, OpcodeDecoder::Primitive primitive,
+                            const u8* src, u32 count);
   virtual DataReader PrepareForAdditionalData(OpcodeDecoder::Primitive primitive, u32 count,
                                               u32 stride, bool cullall);
+  /// Switch cullall off after a call to PrepareForAdditionalData with cullall true
+  /// Expects that you will add a nonzero number of primitives before the next flush
+  /// Returns whether cullall was changed (false if cullall was already off)
+  DataReader DisableCullAll(u32 stride);
   void FlushData(u32 count, u32 stride);
 
   void Flush();
+  bool HasSendableVertices() const { return !m_is_flushed && !m_cull_all; }
 
   void DoState(PointerWrap& p);
 
@@ -201,6 +209,7 @@ class VertexManagerBase
   bool m_cull_all = false;
 
   IndexGenerator m_index_generator;
+  CPUCull m_cpu_cull;
 
 private:
   // Minimum number of draws per command buffer when attempting to preempt a readback operation.

diff --git a/Source/Core/VideoCommon/VertexShaderManager.cpp b/Source/Core/VideoCommon/VertexShaderManager.cpp
@@ -65,6 +65,97 @@ void VertexShaderManager::Dirty()
   dirty = true;
 }
 
+Common::Matrix44 VertexShaderManager::LoadProjectionMatrix()
+{
+  const auto& rawProjection = xfmem.projection.rawProjection;
+
+  switch (xfmem.projection.type)
+  {
+  case ProjectionType::Perspective:
+  {
+    const Common::Vec2 fov_multiplier = g_freelook_camera.IsActive() ?
+                                            g_freelook_camera.GetFieldOfViewMultiplier() :
+                                            Common::Vec2{1, 1};
+    m_projection_matrix[0] = rawProjection[0] * g_ActiveConfig.fAspectRatioHackW * fov_multiplier.x;
+    m_projection_matrix[1] = 0.0f;
+    m_projection_matrix[2] = rawProjection[1] * g_ActiveConfig.fAspectRatioHackW * fov_multiplier.x;
+    m_projection_matrix[3] = 0.0f;
+
+    m_projection_matrix[4] = 0.0f;
+    m_projection_matrix[5] = rawProjection[2] * g_ActiveConfig.fAspectRatioHackH * fov_multiplier.y;
+    m_projection_matrix[6] = rawProjection[3] * g_ActiveConfig.fAspectRatioHackH * fov_multiplier.y;
+    m_projection_matrix[7] = 0.0f;
+
+    m_projection_matrix[8] = 0.0f;
+    m_projection_matrix[9] = 0.0f;
+    m_projection_matrix[10] = rawProjection[4];
+    m_projection_matrix[11] = rawProjection[5];
+
+    m_projection_matrix[12] = 0.0f;
+    m_projection_matrix[13] = 0.0f;
+
+    m_projection_matrix[14] = -1.0f;
+    m_projection_matrix[15] = 0.0f;
+
+    g_stats.gproj = m_projection_matrix;
+  }
+  break;
+
+  case ProjectionType::Orthographic:
+  {
+    m_projection_matrix[0] = rawProjection[0];
+    m_projection_matrix[1] = 0.0f;
+    m_projection_matrix[2] = 0.0f;
+    m_projection_matrix[3] = rawProjection[1];
+
+    m_projection_matrix[4] = 0.0f;
+    m_projection_matrix[5] = rawProjection[2];
+    m_projection_matrix[6] = 0.0f;
+    m_projection_matrix[7] = rawProjection[3];
+
+    m_projection_matrix[8] = 0.0f;
+    m_projection_matrix[9] = 0.0f;
+    m_projection_matrix[10] = rawProjection[4];
+    m_projection_matrix[11] = rawProjection[5];
+
+    m_projection_matrix[12] = 0.0f;
+    m_projection_matrix[13] = 0.0f;
+
+    m_projection_matrix[14] = 0.0f;
+    m_projection_matrix[15] = 1.0f;
+
+    g_stats.g2proj = m_projection_matrix;
+    g_stats.proj = rawProjection;
+  }
+  break;
+
+  default:
+    ERROR_LOG_FMT(VIDEO, "Unknown projection type: {}", xfmem.projection.type);
+  }
+
+  PRIM_LOG("Projection: {} {} {} {} {} {}", rawProjection[0], rawProjection[1], rawProjection[2],
+           rawProjection[3], rawProjection[4], rawProjection[5]);
+
+  auto corrected_matrix = m_viewport_correction * Common::Matrix44::FromArray(m_projection_matrix);
+
+  if (g_freelook_camera.IsActive() && xfmem.projection.type == ProjectionType::Perspective)
+    corrected_matrix *= g_freelook_camera.GetView();
+
+  g_freelook_camera.GetController()->SetClean();
+
+  return corrected_matrix;
+}
+
+void VertexShaderManager::SetProjectionMatrix()
+{
+  if (m_projection_changed || g_freelook_camera.GetController()->IsDirty())
+  {
+    m_projection_changed = false;
+    auto corrected_matrix = LoadProjectionMatrix();
+    memcpy(constants.projection.data(), corrected_matrix.data.data(), 4 * sizeof(float4));
+  }
+}
+
 // Syncs the shader constant buffers with xfmem
 // TODO: A cleaner way to control the matrices without making a mess in the parameters field
 void VertexShaderManager::SetConstants(const std::vector<std::string>& textures)
@@ -317,84 +408,7 @@ void VertexShaderManager::SetConstants(const std::vector<std::string>& textures)
     m_projection_changed = false;
     m_projection_graphics_mod_change = !projection_actions.empty();
 
-    const auto& rawProjection = xfmem.projection.rawProjection;
-
-    switch (xfmem.projection.type)
-    {
-    case ProjectionType::Perspective:
-    {
-      const Common::Vec2 fov_multiplier = g_freelook_camera.IsActive() ?
-                                              g_freelook_camera.GetFieldOfViewMultiplier() :
-                                              Common::Vec2{1, 1};
-      m_projection_matrix[0] =
-          rawProjection[0] * g_ActiveConfig.fAspectRatioHackW * fov_multiplier.x;
-      m_projection_matrix[1] = 0.0f;
-      m_projection_matrix[2] =
-          rawProjection[1] * g_ActiveConfig.fAspectRatioHackW * fov_multiplier.x;
-      m_projection_matrix[3] = 0.0f;
-
-      m_projection_matrix[4] = 0.0f;
-      m_projection_matrix[5] =
-          rawProjection[2] * g_ActiveConfig.fAspectRatioHackH * fov_multiplier.y;
-      m_projection_matrix[6] =
-          rawProjection[3] * g_ActiveConfig.fAspectRatioHackH * fov_multiplier.y;
-      m_projection_matrix[7] = 0.0f;
-
-      m_projection_matrix[8] = 0.0f;
-      m_projection_matrix[9] = 0.0f;
-      m_projection_matrix[10] = rawProjection[4];
-      m_projection_matrix[11] = rawProjection[5];
-
-      m_projection_matrix[12] = 0.0f;
-      m_projection_matrix[13] = 0.0f;
-
-      m_projection_matrix[14] = -1.0f;
-      m_projection_matrix[15] = 0.0f;
-
-      g_stats.gproj = m_projection_matrix;
-    }
-    break;
-
-    case ProjectionType::Orthographic:
-    {
-      m_projection_matrix[0] = rawProjection[0];
-      m_projection_matrix[1] = 0.0f;
-      m_projection_matrix[2] = 0.0f;
-      m_projection_matrix[3] = rawProjection[1];
-
-      m_projection_matrix[4] = 0.0f;
-      m_projection_matrix[5] = rawProjection[2];
-      m_projection_matrix[6] = 0.0f;
-      m_projection_matrix[7] = rawProjection[3];
-
-      m_projection_matrix[8] = 0.0f;
-      m_projection_matrix[9] = 0.0f;
-      m_projection_matrix[10] = rawProjection[4];
-      m_projection_matrix[11] = rawProjection[5];
-
-      m_projection_matrix[12] = 0.0f;
-      m_projection_matrix[13] = 0.0f;
-
-      m_projection_matrix[14] = 0.0f;
-      m_projection_matrix[15] = 1.0f;
-
-      g_stats.g2proj = m_projection_matrix;
-      g_stats.proj = rawProjection;
-    }
-    break;
-
-    default:
-      ERROR_LOG_FMT(VIDEO, "Unknown projection type: {}", xfmem.projection.type);
-    }
-
-    PRIM_LOG("Projection: {} {} {} {} {} {}", rawProjection[0], rawProjection[1], rawProjection[2],
-             rawProjection[3], rawProjection[4], rawProjection[5]);
-
-    auto corrected_matrix =
-        m_viewport_correction * Common::Matrix44::FromArray(m_projection_matrix);
-
-    if (g_freelook_camera.IsActive() && xfmem.projection.type == ProjectionType::Perspective)
-      corrected_matrix *= g_freelook_camera.GetView();
+    auto corrected_matrix = LoadProjectionMatrix();
 
     GraphicsModActionData::Projection projection{&corrected_matrix};
     for (auto action : projection_actions)
@@ -404,8 +418,6 @@ void VertexShaderManager::SetConstants(const std::vector<std::string>& textures)
 
     memcpy(constants.projection.data(), corrected_matrix.data.data(), 4 * sizeof(float4));
 
-    g_freelook_camera.GetController()->SetClean();
-
     dirty = true;
   }
 

diff --git a/Source/Core/VideoCommon/VertexShaderManager.h b/Source/Core/VideoCommon/VertexShaderManager.h
@@ -24,6 +24,7 @@ class alignas(16) VertexShaderManager
   void DoState(PointerWrap& p);
 
   // constant management
+  void SetProjectionMatrix();
   void SetConstants(const std::vector<std::string>& textures);
 
   void InvalidateXFRange(int start, int end);
@@ -64,4 +65,6 @@ class alignas(16) VertexShaderManager
   std::array<int, 2> m_minmax_lights_changed{};
 
   Common::Matrix44 m_viewport_correction{};
+
+  Common::Matrix44 LoadProjectionMatrix();
 };
diff --git a/Source/Core/VideoCommon/VideoConfig.cpp b/Source/Core/VideoCommon/VideoConfig.cpp
@@ -113,6 +113,7 @@ void VideoConfig::Refresh()
   iShaderCompilationMode = Config::Get(Config::GFX_SHADER_COMPILATION_MODE);
   iShaderCompilerThreads = Config::Get(Config::GFX_SHADER_COMPILER_THREADS);
   iShaderPrecompilerThreads = Config::Get(Config::GFX_SHADER_PRECOMPILER_THREADS);
+  bCPUCull = Config::Get(Config::GFX_CPU_CULL);
 
   texture_filtering_mode = Config::Get(Config::GFX_ENHANCE_FORCE_TEXTURE_FILTERING);
   iMaxAnisotropy = Config::Get(Config::GFX_ENHANCE_MAX_ANISOTROPY);

diff --git a/Source/Core/VideoCommon/VideoConfig.h b/Source/Core/VideoCommon/VideoConfig.h
@@ -138,6 +138,7 @@ struct VideoConfig final
   bool bPerfQueriesEnable = false;
   bool bBBoxEnable = false;
   bool bForceProgressive = false;
+  bool bCPUCull = false;
 
   bool bEFBEmulateFormatChanges = false;
   bool bSkipEFBCopyToRam = false;

diff --git a/Source/Core/VideoCommon/XFMemory.h b/Source/Core/VideoCommon/XFMemory.h
@@ -423,7 +423,7 @@ struct Projection
   ProjectionType type;
 };
 
-struct XFMemory
+struct alignas(16) XFMemory
 {
   float posMatrices[256];                   // 0x0000 - 0x00ff
   u32 unk0[768];                            // 0x0100 - 0x03ff