diff --git a/src/xenia/gpu/vulkan/buffer_cache.cc b/src/xenia/gpu/vulkan/buffer_cache.cc
index 1def6d26f..359f90819 100644
--- a/src/xenia/gpu/vulkan/buffer_cache.cc
+++ b/src/xenia/gpu/vulkan/buffer_cache.cc
@@ -344,11 +344,12 @@ VkDeviceSize BufferCache::TryAllocateTransientData(VkDeviceSize alignment,
   if (transient_tail_offset_ >= transient_head_offset_) {
     // Tail follows head, so things are easy:
     // |    H----T   |
-    if (transient_tail_offset_ + length <= transient_capacity_) {
+    if (xe::round_up(transient_tail_offset_, alignment) + length <=
+        transient_capacity_) {
       // Allocation fits from tail to end of buffer, so grow.
       // |    H----**T |
-      VkDeviceSize offset = transient_tail_offset_;
-      transient_tail_offset_ += length;
+      VkDeviceSize offset = xe::round_up(transient_tail_offset_, alignment);
+      transient_tail_offset_ = offset + length;
       return offset;
     } else if (length + kDeadZone <= transient_head_offset_) {
       // Can't fit at the end, but can fit if we wrap around.
@@ -360,11 +361,12 @@ VkDeviceSize BufferCache::TryAllocateTransientData(VkDeviceSize alignment,
   } else {
     // Head follows tail, so we're reversed:
     // |----T    H---|
-    if (transient_tail_offset_ + length + kDeadZone <= transient_head_offset_) {
+    if (xe::round_up(transient_tail_offset_, alignment) + length + kDeadZone <=
+        transient_head_offset_) {
       // Fits from tail to head.
       // |----***T H---|
-      VkDeviceSize offset = transient_tail_offset_;
-      transient_tail_offset_ += length;
+      VkDeviceSize offset = xe::round_up(transient_tail_offset_, alignment);
+      transient_tail_offset_ = offset + length;
       return offset;
     }
   }
diff --git a/src/xenia/gpu/vulkan/render_cache.cc b/src/xenia/gpu/vulkan/render_cache.cc
index a9595741f..5637d44eb 100644
--- a/src/xenia/gpu/vulkan/render_cache.cc
+++ b/src/xenia/gpu/vulkan/render_cache.cc
@@ -293,6 +293,10 @@ bool CachedFramebuffer::IsCompatible(
     const RenderConfiguration& desired_config) const {
   // We already know all render pass things line up, so let's verify dimensions,
   // edram offsets, etc. We need an exact match.
+  if (desired_config.surface_pitch_px != width ||
+      desired_config.surface_height_px != height) {
+    return false;
+  }
   // TODO(benvanik): separate image views from images in tiles and store in fb?
   for (int i = 0; i < 4; ++i) {
     // Ensure the the attachment points to the same tile.
diff --git a/src/xenia/gpu/vulkan/vulkan_command_processor.cc b/src/xenia/gpu/vulkan/vulkan_command_processor.cc
index 33483791a..e568df482 100644
--- a/src/xenia/gpu/vulkan/vulkan_command_processor.cc
+++ b/src/xenia/gpu/vulkan/vulkan_command_processor.cc
@@ -163,11 +163,6 @@ bool VulkanCommandProcessor::IssueDraw(PrimitiveType primitive_type,
                                        IndexBufferInfo* index_buffer_info) {
   auto& regs = *register_file_;
 
-  // TODO(benvanik): move to CP or to host (trace dump, etc).
-  if (FLAGS_vulkan_renderdoc_capture_all && device_->is_renderdoc_attached()) {
-    device_->BeginRenderDocFrameCapture();
-  }
-
 #if FINE_GRAINED_DRAW_SCOPES
   SCOPE_profile_cpu_f("gpu");
 #endif  // FINE_GRAINED_DRAW_SCOPES
@@ -182,6 +177,11 @@ bool VulkanCommandProcessor::IssueDraw(PrimitiveType primitive_type,
     return IssueCopy();
   }
 
+  // TODO(benvanik): move to CP or to host (trace dump, etc).
+  if (FLAGS_vulkan_renderdoc_capture_all && device_->is_renderdoc_attached()) {
+    device_->BeginRenderDocFrameCapture();
+  }
+
   // Shaders will have already been defined by previous loads.
   // We need the to do just about anything so validate here.
   auto vertex_shader = static_cast<VulkanShader*>(active_vertex_shader());