diff --git a/src/core/imgui_overlays.cpp b/src/core/imgui_overlays.cpp
index ced6b4edb..15dd71717 100644
--- a/src/core/imgui_overlays.cpp
+++ b/src/core/imgui_overlays.cpp
@@ -405,10 +405,16 @@ void ImGuiManager::DrawPerformanceOverlay()
 #endif
     }
 
-    if (g_settings.display_show_gpu && g_gpu_device->IsGPUTimingEnabled())
+    if (g_settings.display_show_gpu)
     {
-      text.assign("GPU: ");
-      FormatProcessorStat(text, System::GetGPUUsage(), System::GetGPUAverageTime());
+      if (g_gpu_device->IsGPUTimingEnabled())
+      {
+        text.assign("GPU: ");
+        FormatProcessorStat(text, System::GetGPUUsage(), System::GetGPUAverageTime());
+        DRAW_LINE(fixed_font, text, IM_COL32(255, 255, 255, 255));
+      }
+
+      text.format("VRAM: {} MB", (g_gpu_device->GetVRAMUsage() + (1048576 - 1)) / 1048576);
       DRAW_LINE(fixed_font, text, IM_COL32(255, 255, 255, 255));
     }
 
diff --git a/src/util/d3d11_device.cpp b/src/util/d3d11_device.cpp
index b567f888a..2f6ae29ae 100644
--- a/src/util/d3d11_device.cpp
+++ b/src/util/d3d11_device.cpp
@@ -182,6 +182,7 @@ void D3D11Device::SetFeatures(FeatureMask disabled_features)
   m_features.gpu_timing = true;
   m_features.shader_cache = true;
   m_features.pipeline_cache = false;
+  m_features.prefer_unused_textures = false;
 }
 
 bool D3D11Device::CreateSwapChain()
diff --git a/src/util/d3d11_texture.cpp b/src/util/d3d11_texture.cpp
index a67a49342..a929bf61a 100644
--- a/src/util/d3d11_texture.cpp
+++ b/src/util/d3d11_texture.cpp
@@ -23,11 +23,7 @@ std::unique_ptr<GPUTexture> D3D11Device::CreateTexture(u32 width, u32 height, u3
                                                        GPUTexture::Type type, GPUTexture::Format format,
                                                        const void* data, u32 data_stride)
 {
-  std::unique_ptr<D3D11Texture> tex = std::make_unique<D3D11Texture>();
-  if (!tex->Create(m_device.Get(), width, height, layers, levels, samples, type, format, data, data_stride))
-    tex.reset();
-
-  return tex;
+  return D3D11Texture::Create(m_device.Get(), width, height, layers, levels, samples, type, format, data, data_stride);
 }
 
 bool D3D11Device::DownloadTexture(GPUTexture* texture, u32 x, u32 y, u32 width, u32 height, void* out_data,
@@ -154,11 +150,21 @@ std::unique_ptr<GPUSampler> D3D11Device::CreateSampler(const GPUSampler::Config&
   return std::unique_ptr<GPUSampler>(new D3D11Sampler(std::move(ss)));
 }
 
-D3D11Texture::D3D11Texture() = default;
+D3D11Texture::D3D11Texture(u32 width, u32 height, u32 layers, u32 levels, u32 samples, Type type, Format format,
+                           ComPtr<ID3D11Texture2D> texture, ComPtr<ID3D11ShaderResourceView> srv,
+                           ComPtr<ID3D11View> rtv_dsv)
+  : GPUTexture(static_cast<u16>(width), static_cast<u16>(height), static_cast<u8>(layers), static_cast<u8>(levels),
+               static_cast<u8>(samples), type, format),
+    m_texture(std::move(texture)), m_srv(std::move(srv)), m_rtv_dsv(std::move(rtv_dsv))
+{
+}
 
 D3D11Texture::~D3D11Texture()
 {
-  Destroy();
+  D3D11Device::GetInstance().UnbindTexture(this);
+  m_rtv_dsv.Reset();
+  m_srv.Reset();
+  m_texture.Reset();
 }
 
 D3D11_TEXTURE2D_DESC D3D11Texture::GetDesc() const
@@ -191,11 +197,6 @@ void D3D11Texture::CommitClear(ID3D11DeviceContext1* context)
   m_state = GPUTexture::State::Dirty;
 }
 
-bool D3D11Texture::IsValid() const
-{
-  return static_cast<bool>(m_texture);
-}
-
 bool D3D11Texture::Update(u32 x, u32 y, u32 width, u32 height, const void* data, u32 pitch, u32 layer /*= 0*/,
                           u32 level /*= 0*/)
 {
@@ -268,11 +269,13 @@ DXGI_FORMAT D3D11Texture::GetDXGIFormat() const
   return D3DCommon::GetFormatMapping(m_format).resource_format;
 }
 
-bool D3D11Texture::Create(ID3D11Device* device, u32 width, u32 height, u32 layers, u32 levels, u32 samples, Type type,
-                          Format format, const void* initial_data /* = nullptr */, u32 initial_data_stride /* = 0 */)
+std::unique_ptr<D3D11Texture> D3D11Texture::Create(ID3D11Device* device, u32 width, u32 height, u32 layers, u32 levels,
+                                                   u32 samples, Type type, Format format,
+                                                   const void* initial_data /* = nullptr */,
+                                                   u32 initial_data_stride /* = 0 */)
 {
   if (!ValidateConfig(width, height, layers, layers, samples, type, format))
-    return false;
+    return nullptr;
 
   u32 bind_flags = 0;
   D3D11_USAGE usage = D3D11_USAGE_DEFAULT;
@@ -317,7 +320,7 @@ bool D3D11Texture::Create(ID3D11Device* device, u32 width, u32 height, u32 layer
     Log_ErrorPrintf(
       "Create texture failed: 0x%08X (%ux%u levels:%u samples:%u format:%u bind_flags:%X initial_data:%p)", tex_hr,
       width, height, levels, samples, static_cast<unsigned>(format), bind_flags, initial_data);
-    return false;
+    return nullptr;
   }
 
   ComPtr<ID3D11ShaderResourceView> srv;
@@ -332,7 +335,7 @@ bool D3D11Texture::Create(ID3D11Device* device, u32 width, u32 height, u32 layer
     if (FAILED(hr))
     {
       Log_ErrorPrintf("Create SRV for texture failed: 0x%08X", hr);
-      return false;
+      return nullptr;
     }
   }
 
@@ -347,7 +350,7 @@ bool D3D11Texture::Create(ID3D11Device* device, u32 width, u32 height, u32 layer
     if (FAILED(hr))
     {
       Log_ErrorPrintf("Create RTV for texture failed: 0x%08X", hr);
-      return false;
+      return nullptr;
     }
 
     rtv_dsv = std::move(rtv);
@@ -362,32 +365,14 @@ bool D3D11Texture::Create(ID3D11Device* device, u32 width, u32 height, u32 layer
     if (FAILED(hr))
     {
       Log_ErrorPrintf("Create DSV for texture failed: 0x%08X", hr);
-      return false;
+      return nullptr;
     }
 
     rtv_dsv = std::move(dsv);
   }
 
-  m_texture = std::move(texture);
-  m_srv = std::move(srv);
-  m_rtv_dsv = std::move(rtv_dsv);
-  m_width = static_cast<u16>(width);
-  m_height = static_cast<u16>(height);
-  m_layers = static_cast<u8>(layers);
-  m_levels = static_cast<u8>(levels);
-  m_samples = static_cast<u8>(samples);
-  m_type = type;
-  m_format = format;
-  return true;
-}
-
-void D3D11Texture::Destroy()
-{
-  D3D11Device::GetInstance().UnbindTexture(this);
-  m_rtv_dsv.Reset();
-  m_srv.Reset();
-  m_texture.Reset();
-  ClearBaseProperties();
+  return std::unique_ptr<D3D11Texture>(new D3D11Texture(width, height, layers, levels, samples, type, format,
+                                                        std::move(texture), std::move(srv), std::move(rtv_dsv)));
 }
 
 D3D11TextureBuffer::D3D11TextureBuffer(Format format, u32 size_in_elements) : GPUTextureBuffer(format, size_in_elements)
diff --git a/src/util/d3d11_texture.h b/src/util/d3d11_texture.h
index 4d6815d98..5175af611 100644
--- a/src/util/d3d11_texture.h
+++ b/src/util/d3d11_texture.h
@@ -42,7 +42,6 @@ public:
   template<typename T>
   using ComPtr = Microsoft::WRL::ComPtr<T>;
 
-  D3D11Texture();
   ~D3D11Texture();
 
   ALWAYS_INLINE ID3D11Texture2D* GetD3DTexture() const { return m_texture.Get(); }
@@ -75,16 +74,13 @@ public:
   }
   ALWAYS_INLINE operator bool() const { return static_cast<bool>(m_texture); }
 
-  bool Create(ID3D11Device* device, u32 width, u32 height, u32 layers, u32 levels, u32 samples, Type type,
-              Format format, const void* initial_data = nullptr, u32 initial_data_stride = 0);
-
-  void Destroy();
+  static std::unique_ptr<D3D11Texture> Create(ID3D11Device* device, u32 width, u32 height, u32 layers, u32 levels,
+                                              u32 samples, Type type, Format format, const void* initial_data = nullptr,
+                                              u32 initial_data_stride = 0);
 
   D3D11_TEXTURE2D_DESC GetDesc() const;
   void CommitClear(ID3D11DeviceContext1* context);
 
-  bool IsValid() const override;
-
   bool Update(u32 x, u32 y, u32 width, u32 height, const void* data, u32 pitch, u32 layer = 0, u32 level = 0) override;
   bool Map(void** map, u32* map_stride, u32 x, u32 y, u32 width, u32 height, u32 layer = 0, u32 level = 0) override;
   void Unmap() override;
@@ -92,6 +88,9 @@ public:
   void SetDebugName(const std::string_view& name) override;
 
 private:
+  D3D11Texture(u32 width, u32 height, u32 layers, u32 levels, u32 samples, Type type, Format format,
+               ComPtr<ID3D11Texture2D> texture, ComPtr<ID3D11ShaderResourceView> srv, ComPtr<ID3D11View> rtv_dsv);
+
   ComPtr<ID3D11Texture2D> m_texture;
   ComPtr<ID3D11ShaderResourceView> m_srv;
   ComPtr<ID3D11View> m_rtv_dsv;
diff --git a/src/util/d3d12_device.cpp b/src/util/d3d12_device.cpp
index fbbfe2698..2b64175b2 100644
--- a/src/util/d3d12_device.cpp
+++ b/src/util/d3d12_device.cpp
@@ -1187,6 +1187,7 @@ void D3D12Device::SetFeatures(FeatureMask disabled_features)
   m_features.gpu_timing = true;
   m_features.shader_cache = true;
   m_features.pipeline_cache = true;
+  m_features.prefer_unused_textures = true;
 
   BOOL allow_tearing_supported = false;
   HRESULT hr = m_dxgi_factory->CheckFeatureSupport(DXGI_FEATURE_PRESENT_ALLOW_TEARING, &allow_tearing_supported,
diff --git a/src/util/d3d12_texture.h b/src/util/d3d12_texture.h
index b170d126c..277e22c73 100644
--- a/src/util/d3d12_texture.h
+++ b/src/util/d3d12_texture.h
@@ -37,7 +37,6 @@ public:
   ALWAYS_INLINE DXGI_FORMAT GetDXGIFormat() const { return m_dxgi_format; }
   ALWAYS_INLINE ID3D12Resource* GetResource() const { return m_resource.Get(); }
 
-  bool IsValid() const override { return static_cast<bool>(m_resource); }
   bool Update(u32 x, u32 y, u32 width, u32 height, const void* data, u32 pitch, u32 layer = 0, u32 level = 0) override;
   bool Map(void** map, u32* map_stride, u32 x, u32 y, u32 width, u32 height, u32 layer = 0, u32 level = 0) override;
   void Unmap() override;
diff --git a/src/util/gpu_device.cpp b/src/util/gpu_device.cpp
index 0b649e06c..285c88ce3 100644
--- a/src/util/gpu_device.cpp
+++ b/src/util/gpu_device.cpp
@@ -38,6 +38,7 @@ Log_SetChannel(GPUDevice);
 std::unique_ptr<GPUDevice> g_gpu_device;
 
 static std::string s_pipeline_cache_path;
+size_t GPUDevice::s_total_vram_usage = 0;
 
 GPUSampler::GPUSampler() = default;
 
@@ -763,6 +764,11 @@ Common::Rectangle<s32> GPUDevice::FlipToLowerLeft(const Common::Rectangle<s32>&
   return Common::Rectangle<s32>(rc.left, flipped_y, rc.right, flipped_y + height);
 }
 
+bool GPUDevice::IsTexturePoolType(GPUTexture::Type type)
+{
+  return (type == GPUTexture::Type::Texture || type == GPUTexture::Type::DynamicTexture);
+}
+
 std::unique_ptr<GPUTexture> GPUDevice::FetchTexture(u32 width, u32 height, u32 layers, u32 levels, u32 samples,
                                                     GPUTexture::Type type, GPUTexture::Format format,
                                                     const void* data /*= nullptr*/, u32 data_stride /*= 0*/)
@@ -778,21 +784,56 @@ std::unique_ptr<GPUTexture> GPUDevice::FetchTexture(u32 width, u32 height, u32 l
                               format,
                               0u};
 
-  for (auto it = m_texture_pool.begin(); it != m_texture_pool.end(); ++it)
+  const bool is_texture = IsTexturePoolType(type);
+  TexturePool& pool = is_texture ? m_texture_pool : m_target_pool;
+  const u32 pool_size = (is_texture ? MAX_TEXTURE_POOL_SIZE : MAX_TARGET_POOL_SIZE);
+
+  TexturePool::iterator it;
+
+  if (is_texture && m_features.prefer_unused_textures)
   {
-    if (it->key == key)
+    // Try to find a texture that wasn't used this frame first.
+    // Start at the back of the pool.
+    it = m_texture_pool.end();
+    for (auto rit = m_texture_pool.rbegin(); rit != m_texture_pool.rend(); ++rit)
     {
-      if (data && !it->texture->Update(0, 0, width, height, data, data_stride, 0, 0))
+      if (rit->use_counter == m_texture_pool_counter)
       {
-        // This shouldn't happen...
-        Log_ErrorFmt("Failed to upload {}x{} to pooled texture", width, height);
-        break;
+        // We're into textures recycled this frame, not going to find anything newer.
+        // But prefer reuse over creating a new texture.
+        if (m_texture_pool.size() < pool_size)
+          break;
       }
 
+      if (rit->key == key)
+      {
+        it = std::prev(rit.base());
+        break;
+      }
+    }
+  }
+  else
+  {
+    for (it = pool.begin(); it != pool.end(); ++it)
+    {
+      if (it->key == key)
+        break;
+    }
+  }
+
+  if (it != pool.end())
+  {
+    if (!data || it->texture->Update(0, 0, width, height, data, data_stride, 0, 0))
+    {
       ret = std::move(it->texture);
       m_texture_pool.erase(it);
       return ret;
     }
+    else
+    {
+      // This shouldn't happen...
+      Log_ErrorFmt("Failed to upload {}x{} to pooled texture", width, height);
+    }
   }
 
   ret = CreateTexture(width, height, layers, levels, samples, type, format, data, data_stride);
@@ -814,23 +855,6 @@ void GPUDevice::RecycleTexture(std::unique_ptr<GPUTexture> texture)
   if (!texture)
     return;
 
-  u32 remove_count = m_texture_pool_counter + POOL_PURGE_DELAY;
-  if (remove_count < m_texture_pool_counter)
-  {
-    // wrapped around, handle it
-    if (m_texture_pool.empty())
-    {
-      m_texture_pool_counter = 0;
-      remove_count = POOL_PURGE_DELAY;
-    }
-    else
-    {
-      const u32 reduce = m_texture_pool.front().remove_count;
-      m_texture_pool_counter -= reduce;
-      remove_count -= reduce;
-    }
-  }
-
   const TexturePoolKey key = {static_cast<u16>(texture->GetWidth()),
                               static_cast<u16>(texture->GetHeight()),
                               static_cast<u8>(texture->GetLayers()),
@@ -840,28 +864,74 @@ void GPUDevice::RecycleTexture(std::unique_ptr<GPUTexture> texture)
                               texture->GetFormat(),
                               0u};
 
-  m_texture_pool.push_back({std::move(texture), remove_count, key});
+  const bool is_texture = IsTexturePoolType(texture->GetType());
+  TexturePool& pool = is_texture ? m_texture_pool : m_target_pool;
+  pool.push_back({std::move(texture), m_texture_pool_counter, key});
+
+  const u32 max_size = is_texture ? MAX_TEXTURE_POOL_SIZE : MAX_TARGET_POOL_SIZE;
+  while (pool.size() >= max_size)
+  {
+    Log_ProfileFmt("Trim {}x{} texture from pool", pool.front().texture->GetWidth(), pool.front().texture->GetHeight());
+    pool.pop_front();
+  }
 }
 
 void GPUDevice::PurgeTexturePool()
 {
   m_texture_pool_counter = 0;
   m_texture_pool.clear();
+  m_target_pool.clear();
 }
 
 void GPUDevice::TrimTexturePool()
 {
-  if (m_texture_pool.empty())
+  GL_INS_FMT("Texture Pool Size: {}", m_texture_pool.size());
+  GL_INS_FMT("Target Pool Size: {}", m_target_pool.size());
+  GL_INS_FMT("VRAM Usage: {:.2f} MB", s_total_vram_usage / 1048576.0);
+
+  Log_DebugFmt("Texture Pool Size: {} Target Pool Size: {} VRAM: {:.2f} MB", m_texture_pool.size(),
+               m_target_pool.size(), s_total_vram_usage / 1048756.0);
+
+  if (m_texture_pool.empty() && m_target_pool.empty())
     return;
 
-  const u32 counter = m_texture_pool_counter++;
-  for (auto it = m_texture_pool.begin(); it != m_texture_pool.end();)
+  const u32 prev_counter = m_texture_pool_counter++;
+  for (u32 pool_idx = 0; pool_idx < 2; pool_idx++)
   {
-    if (counter < it->remove_count)
-      break;
+    TexturePool& pool = pool_idx ? m_target_pool : m_texture_pool;
+    for (auto it = pool.begin(); it != pool.end();)
+    {
+      const u32 delta = (it->use_counter - prev_counter);
+      if (delta < POOL_PURGE_DELAY)
+        break;
 
-    Log_ProfileFmt("Trim {}x{} texture from pool", it->texture->GetWidth(), it->texture->GetHeight());
-    it = m_texture_pool.erase(it);
+      Log_ProfileFmt("Trim {}x{} texture from pool", it->texture->GetWidth(), it->texture->GetHeight());
+      it = pool.erase(it);
+    }
+  }
+
+  if (m_texture_pool_counter < prev_counter) [[unlikely]]
+  {
+    // wrapped around, handle it
+    if (m_texture_pool.empty() && m_target_pool.empty())
+    {
+      m_texture_pool_counter = 0;
+    }
+    else
+    {
+      const u32 texture_min =
+        m_texture_pool.empty() ? std::numeric_limits<u32>::max() : m_texture_pool.front().use_counter;
+      const u32 target_min =
+        m_target_pool.empty() ? std::numeric_limits<u32>::max() : m_target_pool.front().use_counter;
+      const u32 reduce = std::min(texture_min, target_min);
+      m_texture_pool_counter -= reduce;
+      for (u32 pool_idx = 0; pool_idx < 2; pool_idx++)
+      {
+        TexturePool& pool = pool_idx ? m_target_pool : m_texture_pool;
+        for (TexturePoolEntry& entry : pool)
+          entry.use_counter -= reduce;
+      }
+    }
   }
 }
 
diff --git a/src/util/gpu_device.h b/src/util/gpu_device.h
index c47c88550..76292ecdb 100644
--- a/src/util/gpu_device.h
+++ b/src/util/gpu_device.h
@@ -424,6 +424,8 @@ protected:
 class GPUDevice
 {
 public:
+  friend GPUTexture;
+
   // TODO: drop virtuals
   // TODO: gpu crash handling on present
   using DrawIndex = u16;
@@ -451,6 +453,7 @@ public:
     bool gpu_timing : 1;
     bool shader_cache : 1;
     bool pipeline_cache : 1;
+    bool prefer_unused_textures : 1;
   };
 
   struct AdapterAndModeList
@@ -627,6 +630,7 @@ public:
   virtual void SetVSync(bool enabled) = 0;
 
   ALWAYS_INLINE bool IsDebugDevice() const { return m_debug_device; }
+  ALWAYS_INLINE size_t GetVRAMUsage() const { return s_total_vram_usage; }
 
   bool UpdateImGuiFontTexture();
   bool UsesLowerLeftOrigin() const;
@@ -675,7 +679,9 @@ protected:
   std::unique_ptr<GPUSampler> m_linear_sampler;
 
 private:
-  static constexpr u32 POOL_PURGE_DELAY = 60;
+  static constexpr u32 MAX_TEXTURE_POOL_SIZE = 100;
+  static constexpr u32 MAX_TARGET_POOL_SIZE = 50;
+  static constexpr u32 POOL_PURGE_DELAY = 300;
 
   struct TexturePoolKey
   {
@@ -700,19 +706,27 @@ private:
   struct TexturePoolEntry
   {
     std::unique_ptr<GPUTexture> texture;
-    u32 remove_count;
+    u32 use_counter;
     TexturePoolKey key;
   };
 
+  using TexturePool = std::deque<TexturePoolEntry>;
+
   void OpenShaderCache(const std::string_view& base_path, u32 version);
   void CloseShaderCache();
   bool CreateResources();
   void DestroyResources();
 
+  static bool IsTexturePoolType(GPUTexture::Type type);
+
+  static size_t s_total_vram_usage;
+
   std::unique_ptr<GPUPipeline> m_imgui_pipeline;
   std::unique_ptr<GPUTexture> m_imgui_font_texture;
 
-  std::deque<TexturePoolEntry> m_texture_pool;
+  TexturePool m_texture_pool;
+  TexturePool m_target_pool;
+  size_t m_pool_vram_usage = 0;
   u32 m_texture_pool_counter = 0;
 
   // TODO: Move out.
diff --git a/src/util/gpu_texture.cpp b/src/util/gpu_texture.cpp
index 9994524f5..165d4dca3 100644
--- a/src/util/gpu_texture.cpp
+++ b/src/util/gpu_texture.cpp
@@ -10,15 +10,17 @@
 
 Log_SetChannel(GPUTexture);
 
-GPUTexture::GPUTexture() = default;
-
 GPUTexture::GPUTexture(u16 width, u16 height, u8 layers, u8 levels, u8 samples, Type type, Format format)
   : m_width(width), m_height(height), m_layers(layers), m_levels(levels), m_samples(samples), m_type(type),
     m_format(format)
 {
+  GPUDevice::s_total_vram_usage += GetVRAMUsage();
 }
 
-GPUTexture::~GPUTexture() = default;
+GPUTexture::~GPUTexture()
+{
+  GPUDevice::s_total_vram_usage -= GetVRAMUsage();
+}
 
 const char* GPUTexture::GetFormatName(Format format)
 {
@@ -48,23 +50,30 @@ const char* GPUTexture::GetFormatName(Format format)
   return format_names[static_cast<u8>(format)];
 }
 
-void GPUTexture::ClearBaseProperties()
-{
-  m_width = 0;
-  m_height = 0;
-  m_layers = 0;
-  m_levels = 0;
-  m_samples = 0;
-  m_type = GPUTexture::Type::Unknown;
-  m_format = GPUTexture::Format::Unknown;
-  m_state = State::Dirty;
-}
-
 std::array<float, 4> GPUTexture::GetUNormClearColor() const
 {
   return GPUDevice::RGBA8ToFloat(m_clear_value.color);
 }
 
+size_t GPUTexture::GetVRAMUsage() const
+{
+  if (m_levels == 1) [[likely]]
+    return ((static_cast<size_t>(m_width * m_height) * GetPixelSize(m_format)) * m_layers * m_samples);
+
+  const size_t ps = GetPixelSize(m_format) * m_layers * m_samples;
+  u32 width = m_width;
+  u32 height = m_height;
+  size_t ts = 0;
+  for (u32 i = 0; i < m_levels; i++)
+  {
+    width = (width > 1) ? (width / 2) : width;
+    height = (height > 1) ? (height / 2) : height;
+    ts += static_cast<size_t>(width * height) * ps;
+  }
+
+  return ts;
+}
+
 u32 GPUTexture::GetPixelSize(GPUTexture::Format format)
 {
   static constexpr std::array<u8, static_cast<size_t>(Format::MaxCount)> sizes = {{
diff --git a/src/util/gpu_texture.h b/src/util/gpu_texture.h
index 566dc55fd..9dec6e654 100644
--- a/src/util/gpu_texture.h
+++ b/src/util/gpu_texture.h
@@ -73,9 +73,18 @@ public:
   };
 
 public:
+  GPUTexture(const GPUTexture&) = delete;
   virtual ~GPUTexture();
 
   static const char* GetFormatName(Format format);
+  static u32 GetPixelSize(GPUTexture::Format format);
+  static bool IsDepthFormat(GPUTexture::Format format);
+  static bool IsDepthStencilFormat(GPUTexture::Format format);
+  static bool ValidateConfig(u32 width, u32 height, u32 layers, u32 levels, u32 samples, Type type, Format format);
+
+  static bool ConvertTextureDataToRGBA8(u32 width, u32 height, std::vector<u32>& texture_data, u32& texture_data_stride,
+                                        GPUTexture::Format format);
+  static void FlipTextureDataRGBA8(u32 width, u32 height, std::vector<u32>& texture_data, u32 texture_data_stride);
 
   ALWAYS_INLINE u32 GetWidth() const { return m_width; }
   ALWAYS_INLINE u32 GetHeight() const { return m_height; }
@@ -122,16 +131,9 @@ public:
     m_clear_value.depth = depth;
   }
 
-  static u32 GetPixelSize(GPUTexture::Format format);
-  static bool IsDepthFormat(GPUTexture::Format format);
-  static bool IsDepthStencilFormat(GPUTexture::Format format);
-  static bool ValidateConfig(u32 width, u32 height, u32 layers, u32 levels, u32 samples, Type type, Format format);
+  size_t GetVRAMUsage() const;
 
-  static bool ConvertTextureDataToRGBA8(u32 width, u32 height, std::vector<u32>& texture_data, u32& texture_data_stride,
-                                        GPUTexture::Format format);
-  static void FlipTextureDataRGBA8(u32 width, u32 height, std::vector<u32>& texture_data, u32 texture_data_stride);
-
-  virtual bool IsValid() const = 0;
+  GPUTexture& operator=(const GPUTexture&) = delete;
 
   virtual bool Update(u32 x, u32 y, u32 width, u32 height, const void* data, u32 pitch, u32 layer = 0,
                       u32 level = 0) = 0;
@@ -144,11 +146,8 @@ public:
   virtual void SetDebugName(const std::string_view& name) = 0;
 
 protected:
-  GPUTexture();
   GPUTexture(u16 width, u16 height, u8 layers, u8 levels, u8 samples, Type type, Format format);
 
-  void ClearBaseProperties();
-
   u16 m_width = 0;
   u16 m_height = 0;
   u8 m_layers = 0;
diff --git a/src/util/metal_device.h b/src/util/metal_device.h
index b121cbcd5..5b5ac32f3 100644
--- a/src/util/metal_device.h
+++ b/src/util/metal_device.h
@@ -107,9 +107,6 @@ public:
 
   bool Create(id<MTLDevice> device, u32 width, u32 height, u32 layers, u32 levels, u32 samples, Type type,
               Format format, const void* initial_data = nullptr, u32 initial_data_stride = 0);
-  void Destroy();
-
-  bool IsValid() const override;
 
   bool Update(u32 x, u32 y, u32 width, u32 height, const void* data, u32 pitch, u32 layer = 0, u32 level = 0) override;
   bool Map(void** map, u32* map_stride, u32 x, u32 y, u32 width, u32 height, u32 layer = 0, u32 level = 0) override;
diff --git a/src/util/metal_device.mm b/src/util/metal_device.mm
index b9a144d9f..7be3ec508 100644
--- a/src/util/metal_device.mm
+++ b/src/util/metal_device.mm
@@ -220,6 +220,7 @@ void MetalDevice::SetFeatures(FeatureMask disabled_features)
   m_features.partial_msaa_resolve = false;
   m_features.shader_cache = true;
   m_features.pipeline_cache = false;
+  m_features.prefer_unused_textures = true;
 }
 
 bool MetalDevice::LoadShaders()
@@ -871,13 +872,11 @@ MetalTexture::MetalTexture(id<MTLTexture> texture, u16 width, u16 height, u8 lay
 
 MetalTexture::~MetalTexture()
 {
-  MetalDevice::GetInstance().UnbindTexture(this);
-  Destroy();
-}
-
-bool MetalTexture::IsValid() const
-{
-  return (m_texture != nil);
+  if (m_texture != nil)
+  {
+    MetalDevice::GetInstance().UnbindTexture(this);
+    MetalDevice::DeferRelease(m_texture);
+  }
 }
 
 bool MetalTexture::Update(u32 x, u32 y, u32 width, u32 height, const void* data, u32 pitch, u32 layer /*= 0*/,
@@ -1029,16 +1028,6 @@ void MetalTexture::SetDebugName(const std::string_view& name)
   }
 }
 
-void MetalTexture::Destroy()
-{
-  if (m_texture != nil)
-  {
-    MetalDevice::DeferRelease(m_texture);
-    m_texture = nil;
-  }
-  ClearBaseProperties();
-}
-
 std::unique_ptr<GPUTexture> MetalDevice::CreateTexture(u32 width, u32 height, u32 layers, u32 levels, u32 samples,
                                                        GPUTexture::Type type, GPUTexture::Format format,
                                                        const void* data, u32 data_stride)
diff --git a/src/util/opengl_device.cpp b/src/util/opengl_device.cpp
index da8f52cc1..0c550947f 100644
--- a/src/util/opengl_device.cpp
+++ b/src/util/opengl_device.cpp
@@ -52,11 +52,7 @@ std::unique_ptr<GPUTexture> OpenGLDevice::CreateTexture(u32 width, u32 height, u
                                                         GPUTexture::Type type, GPUTexture::Format format,
                                                         const void* data, u32 data_stride)
 {
-  std::unique_ptr<OpenGLTexture> tex(std::make_unique<OpenGLTexture>());
-  if (!tex->Create(width, height, layers, levels, samples, type, format, data, data_stride))
-    tex.reset();
-
-  return tex;
+  return OpenGLTexture::Create(width, height, layers, levels, samples, type, format, data, data_stride);
 }
 
 bool OpenGLDevice::DownloadTexture(GPUTexture* texture, u32 x, u32 y, u32 width, u32 height, void* out_data,
@@ -533,6 +529,9 @@ bool OpenGLDevice::CheckFeatures(bool* buggy_pbo, FeatureMask disabled_features)
                       "startup will be slow due to compiling shaders.");
   }
 
+  // Mobile drivers prefer textures to not be updated mid-frame.
+  m_features.prefer_unused_textures = is_gles || vendor_id_arm || vendor_id_powervr || vendor_id_qualcomm;
+
   return true;
 }
 
diff --git a/src/util/opengl_texture.cpp b/src/util/opengl_texture.cpp
index d487442fb..dafafcd2e 100644
--- a/src/util/opengl_texture.cpp
+++ b/src/util/opengl_texture.cpp
@@ -82,11 +82,22 @@ const std::tuple<GLenum, GLenum, GLenum>& OpenGLTexture::GetPixelFormatMapping(G
   return gles ? mapping_gles[static_cast<u32>(format)] : mapping[static_cast<u32>(format)];
 }
 
-OpenGLTexture::OpenGLTexture() = default;
+OpenGLTexture::OpenGLTexture(u32 width, u32 height, u32 layers, u32 levels, u32 samples, Type type, Format format,
+                             GLuint id)
+  : GPUTexture(static_cast<u16>(width), static_cast<u16>(height), static_cast<u8>(layers), static_cast<u8>(levels),
+               static_cast<u8>(samples), type, format),
+    m_id(id)
+{
+}
 
 OpenGLTexture::~OpenGLTexture()
 {
-  Destroy();
+  if (m_id != 0)
+  {
+    OpenGLDevice::GetInstance().UnbindTexture(this);
+    glDeleteTextures(1, &m_id);
+    m_id = 0;
+  }
 }
 
 bool OpenGLTexture::UseTextureStorage(bool multisampled)
@@ -99,16 +110,16 @@ bool OpenGLTexture::UseTextureStorage() const
   return UseTextureStorage(IsMultisampled());
 }
 
-bool OpenGLTexture::Create(u32 width, u32 height, u32 layers, u32 levels, u32 samples, Type type, Format format,
-                           const void* data, u32 data_pitch)
+std::unique_ptr<OpenGLTexture> OpenGLTexture::Create(u32 width, u32 height, u32 layers, u32 levels, u32 samples,
+                                                     Type type, Format format, const void* data, u32 data_pitch)
 {
   if (!ValidateConfig(width, height, layers, levels, samples, type, format))
-    return false;
+    return nullptr;
 
   if (layers > 1 && data)
   {
     Log_ErrorPrintf("Loading texture array data not currently supported");
-    return false;
+    return nullptr;
   }
 
   const GLenum target =
@@ -211,34 +222,10 @@ bool OpenGLTexture::Create(u32 width, u32 height, u32 layers, u32 levels, u32 sa
   {
     Log_ErrorPrintf("Failed to create texture: 0x%X", error);
     glDeleteTextures(1, &id);
-    return false;
+    return nullptr;
   }
 
-  if (IsValid())
-    Destroy();
-
-  m_id = id;
-  m_width = static_cast<u16>(width);
-  m_height = static_cast<u16>(height);
-  m_layers = static_cast<u8>(layers);
-  m_levels = static_cast<u8>(levels);
-  m_samples = static_cast<u8>(samples);
-  m_type = type;
-  m_format = format;
-  m_state = GPUTexture::State::Dirty;
-  return true;
-}
-
-void OpenGLTexture::Destroy()
-{
-  if (m_id != 0)
-  {
-    OpenGLDevice::GetInstance().UnbindTexture(this);
-    glDeleteTextures(1, &m_id);
-    m_id = 0;
-  }
-
-  ClearBaseProperties();
+  return std::unique_ptr<OpenGLTexture>(new OpenGLTexture(width, height, layers, levels, samples, type, format, id));
 }
 
 void OpenGLTexture::CommitClear()
diff --git a/src/util/opengl_texture.h b/src/util/opengl_texture.h
index 336c49487..b8fc3e382 100644
--- a/src/util/opengl_texture.h
+++ b/src/util/opengl_texture.h
@@ -15,7 +15,6 @@ class OpenGLTexture final : public GPUTexture
   friend OpenGLDevice;
 
 public:
-  OpenGLTexture();
   OpenGLTexture(const OpenGLTexture&) = delete;
   ~OpenGLTexture();
 
@@ -23,16 +22,15 @@ public:
   static const std::tuple<GLenum, GLenum, GLenum>& GetPixelFormatMapping(Format format, bool gles);
 
   ALWAYS_INLINE GLuint GetGLId() const { return m_id; }
-  bool IsValid() const override { return m_id != 0; }
+
   bool Update(u32 x, u32 y, u32 width, u32 height, const void* data, u32 pitch, u32 layer = 0, u32 level = 0) override;
   bool Map(void** map, u32* map_stride, u32 x, u32 y, u32 width, u32 height, u32 layer = 0, u32 level = 0) override;
   void Unmap() override;
 
   void SetDebugName(const std::string_view& name) override;
 
-  bool Create(u32 width, u32 height, u32 layers, u32 levels, u32 samples, Type type, Format format,
-              const void* data = nullptr, u32 data_pitch = 0);
-  void Destroy();
+  static std::unique_ptr<OpenGLTexture> Create(u32 width, u32 height, u32 layers, u32 levels, u32 samples, Type type,
+                                               Format format, const void* data = nullptr, u32 data_pitch = 0);
 
   bool UseTextureStorage() const;
 
@@ -46,6 +44,8 @@ public:
   OpenGLTexture& operator=(const OpenGLTexture&) = delete;
 
 private:
+  OpenGLTexture(u32 width, u32 height, u32 layers, u32 levels, u32 samples, Type type, Format format, GLuint id);
+
   GLuint m_id = 0;
 
   u32 m_map_offset = 0;
diff --git a/src/util/vulkan_device.cpp b/src/util/vulkan_device.cpp
index 49d4cf233..bf91d0e4b 100644
--- a/src/util/vulkan_device.cpp
+++ b/src/util/vulkan_device.cpp
@@ -2414,6 +2414,7 @@ bool VulkanDevice::CheckFeatures(FeatureMask disabled_features)
   m_features.partial_msaa_resolve = true;
   m_features.shader_cache = true;
   m_features.pipeline_cache = true;
+  m_features.prefer_unused_textures = true;
 
   return true;
 }
diff --git a/src/util/vulkan_texture.h b/src/util/vulkan_texture.h
index 0dc84670e..f33b8c472 100644
--- a/src/util/vulkan_texture.h
+++ b/src/util/vulkan_texture.h
@@ -48,7 +48,6 @@ public:
 
   VkImageLayout GetVkLayout() const;
 
-  bool IsValid() const override { return (m_image != VK_NULL_HANDLE); }
   bool Update(u32 x, u32 y, u32 width, u32 height, const void* data, u32 pitch, u32 layer = 0, u32 level = 0) override;
   bool Map(void** map, u32* map_stride, u32 x, u32 y, u32 width, u32 height, u32 layer = 0, u32 level = 0) override;
   void Unmap() override;