Optane was targeting the latency gap, while HBF targets bandwidth. Given how LLMs work, HBF is perfect for offloading.