diff --git a/Source/Core/VideoBackends/D3D/TextureCache.cpp b/Source/Core/VideoBackends/D3D/TextureCache.cpp
index d6369f290f..de22a2c04f 100644
--- a/Source/Core/VideoBackends/D3D/TextureCache.cpp
+++ b/Source/Core/VideoBackends/D3D/TextureCache.cpp
@@ -129,10 +129,10 @@ void TextureCache::TCacheEntry::CopyRectangleFromTexture(const TCacheEntryBase*
   g_renderer->RestoreAPIState();
 }
 
-void TextureCache::TCacheEntry::Load(const u8* buffer, u32 width, u32 height, u32 expanded_width,
-                                     u32 level)
+void TextureCache::TCacheEntry::Load(u32 level, u32 width, u32 height, u32 row_length,
+                                     const u8* buffer, size_t buffer_size)
 {
-  unsigned int src_pitch = 4 * expanded_width;
+  u32 src_pitch = row_length * 4;
   D3D::context->UpdateSubresource(texture->GetTex(), level, nullptr, buffer, src_pitch, 0);
 }
 
diff --git a/Source/Core/VideoBackends/D3D/TextureCache.h b/Source/Core/VideoBackends/D3D/TextureCache.h
index 01b229cf72..78450eefdd 100644
--- a/Source/Core/VideoBackends/D3D/TextureCache.h
+++ b/Source/Core/VideoBackends/D3D/TextureCache.h
@@ -30,7 +30,8 @@ private:
                                   const MathUtil::Rectangle<int>& srcrect,
                                   const MathUtil::Rectangle<int>& dstrect) override;
 
-    void Load(const u8* buffer, u32 width, u32 height, u32 expanded_width, u32 levels) override;
+    void Load(u32 level, u32 width, u32 height, u32 row_length, const u8* buffer,
+              size_t buffer_size) override;
 
     void FromRenderTarget(bool is_depth_copy, const EFBRectangle& srcRect, bool scaleByHalf,
                           unsigned int cbufid, const float* colmat) override;
diff --git a/Source/Core/VideoBackends/D3D12/TextureCache.cpp b/Source/Core/VideoBackends/D3D12/TextureCache.cpp
index d626cb7732..77009112cd 100644
--- a/Source/Core/VideoBackends/D3D12/TextureCache.cpp
+++ b/Source/Core/VideoBackends/D3D12/TextureCache.cpp
@@ -175,10 +175,10 @@ void TextureCache::TCacheEntry::CopyRectangleFromTexture(const TCacheEntryBase*
   g_renderer->RestoreAPIState();
 }
 
-void TextureCache::TCacheEntry::Load(const u8* buffer, u32 width, u32 height, u32 expanded_width,
-                                     u32 level)
+void TextureCache::TCacheEntry::Load(u32 level, u32 width, u32 height, u32 row_length,
+                                     const u8* buffer, size_t buffer_size)
 {
-  unsigned int src_pitch = 4 * expanded_width;
+  u32 src_pitch = 4 * row_length;
   D3D::ReplaceRGBATexture2D(m_texture->GetTex12(), buffer, width, height, src_pitch, level,
                             m_texture->GetResourceUsageState());
 }
diff --git a/Source/Core/VideoBackends/D3D12/TextureCache.h b/Source/Core/VideoBackends/D3D12/TextureCache.h
index f6921afe5b..fb8cdf3385 100644
--- a/Source/Core/VideoBackends/D3D12/TextureCache.h
+++ b/Source/Core/VideoBackends/D3D12/TextureCache.h
@@ -39,7 +39,8 @@ private:
                                   const MathUtil::Rectangle<int>& src_rect,
                                   const MathUtil::Rectangle<int>& dst_rect) override;
 
-    void Load(const u8* buffer, u32 width, u32 height, u32 expanded_width, u32 levels) override;
+    void Load(u32 level, u32 width, u32 height, u32 row_length, const u8* buffer,
+              size_t buffer_size) override;
 
     void FromRenderTarget(bool is_depth_copy, const EFBRectangle& src_rect, bool scale_by_half,
                           unsigned int cbuf_id, const float* colmat) override;
diff --git a/Source/Core/VideoBackends/Null/NullBackend.cpp b/Source/Core/VideoBackends/Null/NullBackend.cpp
index 7892b86927..bcbbd1aaab 100644
--- a/Source/Core/VideoBackends/Null/NullBackend.cpp
+++ b/Source/Core/VideoBackends/Null/NullBackend.cpp
@@ -45,6 +45,7 @@ void VideoBackend::InitBackendInfo()
   g_Config.backend_info.bSupportsMultithreading = false;
   g_Config.backend_info.bSupportsInternalResolutionFrameDumps = false;
   g_Config.backend_info.bSupportsGPUTextureDecoding = false;
+  g_Config.backend_info.bSupportsST3CTextures = false;
 
   // aamodes: We only support 1 sample, so no MSAA
   g_Config.backend_info.Adapters.clear();
diff --git a/Source/Core/VideoBackends/Null/TextureCache.h b/Source/Core/VideoBackends/Null/TextureCache.h
index 48d111dce6..399af04764 100644
--- a/Source/Core/VideoBackends/Null/TextureCache.h
+++ b/Source/Core/VideoBackends/Null/TextureCache.h
@@ -31,7 +31,10 @@ private:
   {
     TCacheEntry(const TCacheEntryConfig& _config) : TCacheEntryBase(_config) {}
     ~TCacheEntry() {}
-    void Load(const u8* buffer, u32 width, u32 height, u32 expanded_width, u32 level) override {}
+    void Load(u32 level, u32 width, u32 height, u32 row_length, const u8* buffer,
+              size_t buffer_size) override
+    {
+    }
     void FromRenderTarget(bool is_depth_copy, const EFBRectangle& src_rect, bool scale_by_half,
                           unsigned int cbufid, const float* colmat) override
     {
diff --git a/Source/Core/VideoBackends/OGL/TextureCache.cpp b/Source/Core/VideoBackends/OGL/TextureCache.cpp
index 3bf1342809..e49483a593 100644
--- a/Source/Core/VideoBackends/OGL/TextureCache.cpp
+++ b/Source/Core/VideoBackends/OGL/TextureCache.cpp
@@ -202,8 +202,8 @@ void TextureCache::TCacheEntry::CopyRectangleFromTexture(const TCacheEntryBase*
   g_renderer->RestoreAPIState();
 }
 
-void TextureCache::TCacheEntry::Load(const u8* buffer, u32 width, u32 height, u32 expanded_width,
-                                     u32 level)
+void TextureCache::TCacheEntry::Load(u32 level, u32 width, u32 height, u32 row_length,
+                                     const u8* buffer, size_t buffer_size)
 {
   if (level >= config.levels)
     PanicAlert("Texture only has %d levels, can't update level %d", config.levels, level);
@@ -216,8 +216,8 @@ void TextureCache::TCacheEntry::Load(const u8* buffer, u32 width, u32 height, u3
   glActiveTexture(GL_TEXTURE9);
   glBindTexture(GL_TEXTURE_2D_ARRAY, texture);
 
-  if (expanded_width != width)
-    glPixelStorei(GL_UNPACK_ROW_LENGTH, expanded_width);
+  if (row_length != width)
+    glPixelStorei(GL_UNPACK_ROW_LENGTH, row_length);
 
   if (g_ogl_config.bSupportsTextureStorage)
   {
@@ -230,7 +230,7 @@ void TextureCache::TCacheEntry::Load(const u8* buffer, u32 width, u32 height, u3
                  GL_UNSIGNED_BYTE, buffer);
   }
 
-  if (expanded_width != width)
+  if (row_length != width)
     glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
 
   TextureCache::SetStage();
diff --git a/Source/Core/VideoBackends/OGL/TextureCache.h b/Source/Core/VideoBackends/OGL/TextureCache.h
index aece35cd6c..de0de485f5 100644
--- a/Source/Core/VideoBackends/OGL/TextureCache.h
+++ b/Source/Core/VideoBackends/OGL/TextureCache.h
@@ -45,7 +45,8 @@ private:
                                   const MathUtil::Rectangle<int>& srcrect,
                                   const MathUtil::Rectangle<int>& dstrect) override;
 
-    void Load(const u8* buffer, u32 width, u32 height, u32 expanded_width, u32 level) override;
+    void Load(u32 level, u32 width, u32 height, u32 row_length, const u8* buffer,
+              size_t buffer_size) override;
 
     void FromRenderTarget(bool is_depth_copy, const EFBRectangle& srcRect, bool scaleByHalf,
                           unsigned int cbufid, const float* colmat) override;
diff --git a/Source/Core/VideoBackends/Software/SWmain.cpp b/Source/Core/VideoBackends/Software/SWmain.cpp
index 1b26f1d40a..e004a0cac3 100644
--- a/Source/Core/VideoBackends/Software/SWmain.cpp
+++ b/Source/Core/VideoBackends/Software/SWmain.cpp
@@ -65,7 +65,10 @@ private:
   {
     TCacheEntry(const TCacheEntryConfig& _config) : TCacheEntryBase(_config) {}
     ~TCacheEntry() {}
-    void Load(const u8* buffer, u32 width, u32 height, u32 expanded_width, u32 level) override {}
+    void Load(u32 level, u32 width, u32 height, u32 row_length, const u8* buffer,
+              size_t buffer_size) override
+    {
+    }
     void FromRenderTarget(bool is_depth_copy, const EFBRectangle& srcRect, bool scaleByHalf,
                           unsigned int cbufid, const float* colmat) override
     {
diff --git a/Source/Core/VideoBackends/Vulkan/TextureCache.cpp b/Source/Core/VideoBackends/Vulkan/TextureCache.cpp
index 08c370b48f..62c0228010 100644
--- a/Source/Core/VideoBackends/Vulkan/TextureCache.cpp
+++ b/Source/Core/VideoBackends/Vulkan/TextureCache.cpp
@@ -341,8 +341,8 @@ TextureCache::TCacheEntry::~TCacheEntry()
     g_command_buffer_mgr->DeferFramebufferDestruction(m_framebuffer);
 }
 
-void TextureCache::TCacheEntry::Load(const u8* buffer, unsigned int width, unsigned int height,
-                                     unsigned int expanded_width, unsigned int level)
+void TextureCache::TCacheEntry::Load(u32 level, u32 width, u32 height, u32 row_length,
+                                     const u8* buffer, size_t buffer_size)
 {
   // Can't copy data larger than the texture extents.
   width = std::max(1u, std::min(width, m_texture->GetWidth() >> level));
@@ -371,7 +371,7 @@ void TextureCache::TCacheEntry::Load(const u8* buffer, unsigned int width, unsig
   u32 upload_pitch = upload_width * sizeof(u32);
   u32 upload_size = upload_pitch * height;
   u32 upload_alignment = static_cast<u32>(g_vulkan_context->GetBufferImageGranularity());
-  u32 source_pitch = expanded_width * 4;
+  u32 source_pitch = row_length * 4;
   if ((upload_size + upload_alignment) <= STAGING_TEXTURE_UPLOAD_THRESHOLD &&
       (upload_size + upload_alignment) <= MAXIMUM_TEXTURE_UPLOAD_BUFFER_SIZE)
   {
diff --git a/Source/Core/VideoBackends/Vulkan/TextureCache.h b/Source/Core/VideoBackends/Vulkan/TextureCache.h
index 5fd25c23d8..884d82ddf4 100644
--- a/Source/Core/VideoBackends/Vulkan/TextureCache.h
+++ b/Source/Core/VideoBackends/Vulkan/TextureCache.h
@@ -27,8 +27,8 @@ public:
 
     Texture2D* GetTexture() const { return m_texture.get(); }
     VkFramebuffer GetFramebuffer() const { return m_framebuffer; }
-    void Load(const u8* buffer, unsigned int width, unsigned int height,
-              unsigned int expanded_width, unsigned int level) override;
+    void Load(u32 level, u32 width, u32 height, u32 row_length, const u8* buffer,
+              size_t buffer_size) override;
     void FromRenderTarget(bool is_depth_copy, const EFBRectangle& src_rect, bool scale_by_half,
                           unsigned int cbufid, const float* colmat) override;
     void CopyRectangleFromTexture(const TCacheEntryBase* source,
diff --git a/Source/Core/VideoCommon/TextureCacheBase.cpp b/Source/Core/VideoCommon/TextureCacheBase.cpp
index 89d473cfec..e3a9daa69d 100644
--- a/Source/Core/VideoCommon/TextureCacheBase.cpp
+++ b/Source/Core/VideoCommon/TextureCacheBase.cpp
@@ -750,8 +750,6 @@ TextureCacheBase::TCacheEntryBase* TextureCacheBase::Load(const u32 stage)
       }
       expandedWidth = level.width;
       expandedHeight = level.height;
-      CheckTempSize(level.data_size);
-      memcpy(temp, level.data.get(), level.data_size);
     }
   }
 
@@ -774,6 +772,7 @@ TextureCacheBase::TCacheEntryBase* TextureCacheBase::Load(const u32 stage)
   config.width = width;
   config.height = height;
   config.levels = texLevels;
+  config.format = hires_tex ? hires_tex->GetFormat() : HostTextureFormat::RGBA8;
 
   TCacheEntryBase* entry = AllocateTexture(config);
   GFX_DEBUGGER_PAUSE_AT(NEXT_NEW_TEXTURE, true);
@@ -784,17 +783,20 @@ TextureCacheBase::TCacheEntryBase* TextureCacheBase::Load(const u32 stage)
   const u8* tlut = &texMem[tlutaddr];
   if (hires_tex)
   {
-    entry->Load(temp, width, height, expandedWidth, 0);
+    const auto& level = hires_tex->m_levels[0];
+    entry->Load(0, level.width, level.height, level.row_length, level.data.get(), level.data_size);
   }
-  else if (decode_on_gpu)
+  if (!hires_tex && decode_on_gpu)
   {
     u32 row_stride = bytes_per_block * (expandedWidth / bsw);
     g_texture_cache->DecodeTextureOnGPU(
         entry, 0, src_data, texture_size, static_cast<TextureFormat>(texformat), width, height,
         expandedWidth, expandedHeight, row_stride, tlut, static_cast<TlutFormat>(tlutfmt));
   }
-  else
+  else if (!hires_tex)
   {
+    size_t decoded_texture_size = expandedWidth * sizeof(u32) * expandedHeight;
+    CheckTempSize(decoded_texture_size);
     if (!(texformat == GX_TF_RGBA8 && from_tmem))
     {
       TexDecoder_Decode(temp, src_data, expandedWidth, expandedHeight, texformat, tlut,
@@ -807,7 +809,7 @@ TextureCacheBase::TCacheEntryBase* TextureCacheBase::Load(const u32 stage)
       TexDecoder_DecodeRGBA8FromTmem(temp, src_data, src_data_gb, expandedWidth, expandedHeight);
     }
 
-    entry->Load(temp, width, height, expandedWidth, 0);
+    entry->Load(0, width, height, expandedWidth, temp, decoded_texture_size);
   }
 
   iter = textures_by_address.emplace(address, entry);
@@ -837,9 +839,8 @@ TextureCacheBase::TCacheEntryBase* TextureCacheBase::Load(const u32 stage)
     for (u32 level_index = 1; level_index != texLevels; ++level_index)
     {
       const auto& level = hires_tex->m_levels[level_index];
-      CheckTempSize(level.data_size);
-      memcpy(temp, level.data.get(), level.data_size);
-      entry->Load(temp, level.width, level.height, level.width, level_index);
+      entry->Load(level_index, level.width, level.height, level.row_length, level.data.get(),
+                  level.data_size);
     }
   }
   else
@@ -877,9 +878,11 @@ TextureCacheBase::TCacheEntryBase* TextureCacheBase::Load(const u32 stage)
       }
       else
       {
+        // No need to call CheckTempSize here, as mips will always be smaller than the base level.
+        size_t decoded_mip_size = expanded_mip_width * sizeof(u32) * expanded_mip_height;
         TexDecoder_Decode(temp, mip_src_data, expanded_mip_width, expanded_mip_height, texformat,
                           tlut, (TlutFormat)tlutfmt);
-        entry->Load(temp, mip_width, mip_height, expanded_mip_width, level);
+        entry->Load(level, mip_width, mip_height, expanded_mip_width, temp, decoded_mip_size);
       }
 
       mip_src_data += mip_size;
diff --git a/Source/Core/VideoCommon/TextureCacheBase.h b/Source/Core/VideoCommon/TextureCacheBase.h
index 4c9582079b..f621f7d92a 100644
--- a/Source/Core/VideoCommon/TextureCacheBase.h
+++ b/Source/Core/VideoCommon/TextureCacheBase.h
@@ -27,16 +27,16 @@ public:
 
     bool operator==(const TCacheEntryConfig& o) const
     {
-      return std::tie(width, height, levels, layers, rendertarget) ==
-             std::tie(o.width, o.height, o.levels, o.layers, o.rendertarget);
+      return std::tie(width, height, levels, layers, format, rendertarget) ==
+             std::tie(o.width, o.height, o.levels, o.layers, o.format, o.rendertarget);
     }
 
     struct Hasher : std::hash<u64>
     {
       size_t operator()(const TCacheEntryConfig& c) const
       {
-        u64 id = (u64)c.rendertarget << 63 | (u64)c.layers << 48 | (u64)c.levels << 32 |
-                 (u64)c.height << 16 | (u64)c.width;
+        u64 id = (u64)c.rendertarget << 63 | (u64)c.format << 50 | (u64)c.layers << 48 |
+                 (u64)c.levels << 32 | (u64)c.height << 16 | (u64)c.width;
         return std::hash<u64>::operator()(id);
       }
     };
@@ -45,6 +45,7 @@ public:
     u32 height = 0;
     u32 levels = 1;
     u32 layers = 1;
+    HostTextureFormat format = HostTextureFormat::RGBA8;
     bool rendertarget = false;
   };
 
@@ -129,7 +130,8 @@ public:
                                           const MathUtil::Rectangle<int>& srcrect,
                                           const MathUtil::Rectangle<int>& dstrect) = 0;
 
-    virtual void Load(const u8* buffer, u32 width, u32 height, u32 expanded_width, u32 level) = 0;
+    virtual void Load(u32 level, u32 width, u32 height, u32 row_length, const u8* buffer,
+                      size_t buffer_size) = 0;
     virtual void FromRenderTarget(bool is_depth_copy, const EFBRectangle& srcRect, bool scaleByHalf,
                                   unsigned int cbufid, const float* colmat) = 0;