Advanced Memory Architecture

One of the main challenges limiting the attainable performance for real-world applications on HPC systems is the von-Neumann bottleneck: Data located on storage (and then memory) needs to reach the computational units as quickly as possible, but bandwidth and latency are never good enough on powerful CPUs and GPUs. For applications dealing with very large amounts of data (e.g., AI training) this bottleneck is even more painful. This panel will bring together researchers and vendors of memory devices working on various innovative approaches to tackle the memory wall such as memristors, in-memory computing, or memory disaggregation.