There is a test lurking here. A few times throughout the kernel's history have the post-resume GTT speed been an order of magnitude slower, hence why this exists in the first place. That test needs to be automated. -Chris