VK_EXT_gpu_assisted_validation overhead and default activation state

Answer

GPU‑assisted validation should be turned on by default in debug builds only when the target device has enough dedicated VRAM to absorb the extra validation buffers without risking out‑of‑memory (OOM) failures, and when the measured CPU‑time savings outweigh the added GPU workload. On hardware that does not meet those conditions the extension must remain opt‑in.

Confirmed facts

• The VK_EXT_gpu_assisted_validation extension is explicitly opt‑in via VK_VALIDATION_FEATURE_ENABLE_GPU_ASSISTED_EXT during vkCreateInstance; it is not enabled by default in any Vulkan SDK release.
• When active, validation work is moved from the CPU to the GPU, reducing CPU overhead but increasing GPU memory consumption roughly proportional to the amount of validation data (command buffers, descriptors, pipeline state, etc.).
• On low‑end or integrated GPUs with limited VRAM (< 1 GB dedicated), the extra memory pressure can cause validation failures or frame‑rate drops.

Likely explanation / criteria for default enablement

To decide whether to flip the default to “on” for debug builds, evaluate the following:

1. Available dedicated VRAM – query VkMemoryHeap::flags for VK_MEMORY_HEAP_DEVICE_LOCAL_BIT and sum the size of all device‑local heaps. If the total exceeds a safe threshold (commonly ≈ 1 GB – 2 GB, depending on the application’s baseline usage), the extra validation buffers are unlikely to cause OOM.
2. Validation complexity – enable only the validation features you actually need (e.g., parameter validation, synchronization) using VK_VALIDATION_FEATURE_ENABLE_* flags. Fewer active checks mean lower GPU memory demand.
3. Measured CPU‑GPU trade‑off – run a short benchmark with and without VK_VALIDATION_FEATURE_ENABLE_GPU_ASSISTED_EXT. If the CPU time saved is significant (> 10 % of frame time) and GPU frame time does not increase beyond acceptable limits, the extension is a net win.
4. Target audience – if the debug build is intended for developers on workstation‑class GPUs, enabling by default is reasonable. For builds that may run on consumer laptops or mobile‑class GPUs, keep it opt‑in.

Steps to mitigate VRAM overhead when the extension is active

If you determine that GPU‑assisted validation is beneficial but VRAM is tight, apply these scoped actions:

• Limit validation features – enable only a subset, e.g.:

VkValidationFeatureEnableEXT enables[] = {
    VK_VALIDATION_FEATURE_ENABLE_GPU_ASSISTED_EXT,
    VK_VALIDATION_FEATURE_ENABLE_SYNCHRONIZATION_VALIDATION_EXT
};
VkValidationFeaturesEXT features = {
    .sType = VK_STRUCTURE_TYPE_VALIDATION_FEATURES_EXT,
    .enabledValidationFeatureCount = 2,
    .pEnabledValidationFeatures = enables
};
VkInstanceCreateInfo createInfo = { .sType = VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO, .pNext = &features };

• Reduce validation data size – turn off verbose logging or debug‑message callbacks that generate large amounts of validation payloads.
• Use memory budgeting – after instance creation, call vkGetPhysicalDeviceMemoryProperties2 with VK_KHR_get_memory_requirements2 to monitor device‑local heap usage and dynamically disable the extension if usage approaches a safe limit (e.g., 80 % of total device‑local memory).
• Profile and adjust – capture GPU memory usage with tools such as Radeon GPU Profiler, NVIDIA Nsight Graphics, or Intel GPU Viewer while the extension is active; adjust the enabled validation features until the memory footprint stays within budget.

Missing diagnostic detail

To give a definitive recommendation for a specific device, we need to know the total dedicated VRAM size reported by the physical device. If you can provide that value, we can confirm whether the default‑on threshold is satisfied.