diff --git a/src/shader_recompiler/info.h b/src/shader_recompiler/info.h
index 12e48c8e4..ba28d7e43 100644
--- a/src/shader_recompiler/info.h
+++ b/src/shader_recompiler/info.h
@@ -62,7 +62,14 @@ struct BufferResource {
     }
 
     bool IsStorage(const AmdGpu::Buffer& buffer, const Profile& profile) const noexcept {
-        return buffer.GetSize() > profile.max_ubo_size || is_written;
+        // When using uniform buffers, a size is required at compilation time, so we need to
+        // either compile a lot of shader specializations to handle each size or just force it to
+        // the maximum possible size always. However, for some vendors the shader-supplied size is
+        // used for bounds checking uniform buffer accesses, so the latter would effectively turn
+        // off buffer robustness behavior. Instead, force storage buffers which are bounds checked
+        // using the actual buffer size. We are assuming the performance hit from this is
+        // acceptable.
+        return true; // buffer.GetSize() > profile.max_ubo_size || is_written;
     }
 
     [[nodiscard]] constexpr AmdGpu::Buffer GetSharp(const Info& info) const noexcept;