#ai-workloads

#ai-workloads

In terms of capacity, it will be the city's largest hyperscale data center, with a single major cloud player as its tenant. The project involves 78MW of new capacity, according to Reuters. That may sound modest, but in a European context, the figure is striking. The new capacity represents approximately 7 percent of the total 1,162MW of new live data center capacity added in continental Europe this year.

Datacenters are blocking other energy users from connecting to the grid by over-reserving capacity in case they need it for future growth, according to a new report. Advisory body Uptime Institute says that power use in modern datacenters tends to remain largely constant, with significant changes in demand typically coming from business expansion. To allow for this, operators often reserve significantly more power than they require, which prevents other users from being allocated that capacity by the grid, even though it is not actually being used.

Projections released by the IT analyst house suggest the electricity demands of datacentres will grow by 16% this year and are on course to double by 2030. At the same time, Gartner analysts estimate that the amount of electricity consumed by the global datacentre market will hit 448 terrawatt hours (TWh) in 2025, rising to 980 TWh by 2030, with much of this energy consumed by power-hungry AI workloads hosted in these datacentres.

Despite ongoing speculation around an investment bubble that may be set to burst, artificial intelligence (AI) technology is here to stay. And while an over-inflated market may exist at the level of the suppliers, AI is well-developed and has a firm foothold among organisations of all sizes. But AI workloads place specific demands on IT infrastructure and on storage in particular. Data volumes can start big and then balloon, in particular during training phases as data is vectorised and checkpoints are created.

The company's most recent products of note are the Gemini-I and Gemini-II associative processing units (APUs). These APUs use Compute-In-Memory (CIM) architectures, which can process tasks inside the memory itself without the need to transfer the data to an external processor. Shares in GSI Technology surged 155% yesterday after the company announced the publication of a paper by researchers at Cornell University.

With HyperScale Edge, Commvault is specifically targeting remote locations and smaller businesses where space and IT resources are limited. Think of retail stores, branches, and distribution centers. These environments often have different needs than large data centers. The solution works with validated hardware from Dell, HPE, and Lenovo. Like the other HyperScale products, it is delivered as a software image that is easy to install. This makes implementation faster and more predictable.

The more things change, the more they stay the same, as the French say. That's certainly the case in enterprise storage. Here we review the storage supplier profiles published this year on ComputerWeekly.com, and find all the key players building on key themes of the past decade. These include: flash storage (often QLC for increased density), hybrid cloud operations, storage and backup for containerised apps, as-a-service models of purchasing, and storage for AI workloads.

"One is just how fast the servers can give you back an answer so that you can make a decision. That's really important in a credit card authorization loop..."

Google plans to pause non-essential AI workloads to protect power grids, maintaining operations during peak energy demand by shifting tasks to available power sources.

IBM has introduced Power11 servers designed for AI workloads, offering bandwidth up to 1,200 GB/s, significantly outperforming x86 systems with a 99.9999% uptime promise.