NVMe Caching Architecture for Object Storage Reads
GPU training wastes half its cycles waiting for data from cheap object storage.
Theo Vásquez-Kline
Contributing Writer
Theo Vásquez-Kline
Theo Vásquez-Kline is a researcher and writer who spent four years studying autonomous agent pipelines at a university HPC lab before pivoting to independent technical coverage. He focuses on the emerging and often underspecified problem of how AI agents read, write, and reason about persistent storage.
4 stories
Data Loading Bottlenecks in PyTorch and TensorFlow Training Pipelines
GPU utilization dashboards miss data loading bottlenecks that waste most training time.
Theo Vásquez-Kline
GCS vs S3 vs Azure Blob API Differences That Break Portability
Migration costs compound across authentication, tiering, and event models that don't translate.
Theo Vásquez-Kline
How Read-Ahead and Prefetching Work in Storage Systems
Storage systems guess your next read by prefetching data into memory before you ask for it.
Theo Vásquez-Kline