How To Get Over The Memory Wall

SPONSORED: MemCon 2024 is billed as a one stop shop for emerging technologies in the memory and storage domain, and a hub for efficient data movement and management.

Because it’s all too easy to lose track of compute utilization and ownership costs when you’re working with data-intensive workloads, so you might want to get more familiar with emerging memory technologies like Compute Express Link (CXL) to see what they can do to help.

This year’s MemCon 2024 event will be held at the Computer History Museum in Silicon Valley, California on 26th and 27th March. It will feature speakers from some of the worlds’ biggest IT companies, including Microsoft, Meta, Oracle and Huawei, but the real draw is the customer stories.

You’ll hear Netflix reveal how it handled memory allocation and deallocation when training Machine Learning (ML) models for example, and how using techniques like data quantization, model pruning and efficient mini-batch selection helped the streaming giant conserve memory resources without significantly impacting model performance.

This is a theme which is also touched on by Berkeley Research Lab. The research institute will explain how it analyzed thirteen diverse workloads – including AI training, data analysis, genomics, protein, fusion, atomic nuclei, and traditional HPC bookends – to better understand the potential and pitfalls of a disaggregated memory system.

Shell has been processing large subsurface datasets for decades, but the increasing size and complexity of the algorithms it used posed significant memory and data management challenges. Hear tips on how the company addressed them to accelerate AI-based workflows using technology improvements over the years.

Elsewhere healthcare provider Roche will share an overview of the memory-bound obstacles it faced in scaling up its genomics computations and how it ported key kernels to GPUs and wrote pipelines which could be ported to either clusters or clouds to help take the strain. The full agenda is available here.

So why not join a thriving network of systems and AI/ML experts to help you overcome your data management roadblocks and get a better grasp of ownership costs? Head here to register for MemCon 2024 and you won’t be disappointed (you can save 15 percent on the cost of your ticket when you quote NEXTPLATFORM15 on checkout).

Sponsored by Kisaco.