Chips & Compute

HPE pairs hybrid supercomputing storage with liquid-cooling rack controller

HPE claims its hybrid E2000 storage delivers 116 GB/s reads per rack and up to 82% savings over all-flash, paired with a liquid-cooling rack controller shipping from March 2027.

HPE unveils hybrid supercomputing storage & rack controller
HPE unveils hybrid supercomputing storage & rack controllerfrabre / pixabay
By Rebecca Stone5 min read

Why it matters

  • HPE claims up to 116 GB/s read and 92 GB/s write throughput per rack for the hybrid E2000 configuration.
  • Customers could spend up to 82% less than all-flash, based on 15 PB of usable storage for a 1,024-GPU cluster meeting NVIDIA specs.
  • The hybrid setup offers up to 32% more usable capacity per rack than the current E2000.
  • The HPE Rack Management Controller ships March 2027; AMD and NVIDIA rack-scale support follows June 2027.
  • Hybrid E2000 configurations are planned for early 2027.

HPE has launched a hybrid configuration of its HPE Cray Supercomputing Storage Systems E2000 that it says delivers up to 116 GB per second of read throughput and 92 GB per second of write throughput per rack. The company announced the product alongside a new HPE Rack Management Controller for high-density liquid-cooled racks, targeting organisations that run dense high-performance computing and AI infrastructure.

The hybrid storage setup combines flash drives with 32-terabyte SAS-4 hard disk drives and new SAS-4 drive enclosures. The enclosures support 74 or 100 drives. HPE claims the configuration provides up to 41% more data throughput per storage rack than the nearest hybrid supercomputing storage competitor.

The launch matters because storage economics have become a constraint on AI and scientific computing buildouts. Prices for larger solid-state drives have risen as demand for data-intensive workloads has grown, HPE said, and that shift has pushed research institutions and national laboratories to look for alternatives that balance performance against cost.

How much can customers save?

HPE puts a concrete figure on the savings. Customers could spend up to 82% less than on a comparable all-flash system, according to the company. That figure comes from an analysis of 15 petabytes of usable storage meeting NVIDIA performance specifications for a cluster of 1,024 GPUs.

The hybrid design also packs more capacity into the same footprint. HPE said the new configuration offers up to 32% more usable storage capacity per rack than the current shipping hybrid HPE Cray Supercomputing Storage Systems E2000. Users can therefore support larger datasets without adding rack space — a direct constraint in datacenters where floor space, power and cooling budgets are already stretched.

The company frames the tradeoff in straightforward terms: spend less on storage, and direct more budget to compute nodes. That reflects a broader pattern in high-performance computing, where operators must balance spending across processors, networking, cooling and storage as AI and scientific workloads grow in scale. A cluster of 1,024 GPUs can quickly become compute-rich and storage-starved if flash prices dominate the budget.

Why liquid cooling is driving a new controller

The second product in the announcement addresses a different pressure point. The HPE Rack Management Controller is designed to monitor and manage high-density liquid-cooled racks. It provides real-time monitoring, leak detection, automated control and rack-level visibility in liquid-cooled environments, HPE said.

Liquid cooling has become a growing focus in the market as power densities rise. Operators of AI clusters and supercomputers are under pressure to improve cooling efficiency while maintaining visibility into systems that are becoming more complex at the rack level. HPE said it engineered the controller using experience from several decades of work in liquid cooling.

Leak detection carries particular weight in liquid-cooled estates. A single rack failure in a dense AI cluster can idle thousands of GPU-hours, and rack-level telemetry gives operators a way to catch faults before they spread.

The controller will initially support HPE Cray Supercomputing GX5000 systems. Support for AMD and NVIDIA rack-scale systems is due later, which would extend HPE's management layer onto competitor silicon inside mixed-vendor datacenters.

How does it fit HPE's software stack?

The controller integrates with the HPE Supercomputing Management Software suite. That suite includes:

  • Foundation, for infrastructure management
  • Composable, for resource sharing
  • Observability, for operational monitoring

Through that integration, users receive rack-level telemetry and control functions that HPE said are intended to make large liquid-cooled environments easier to run and expand. The pitch is a single pane of glass across storage, cooling and management — parts of one system approach for customers building larger HPC and AI estates.

HPE framed the launch as part of a broader push to supply infrastructure for scientific, commercial and high-performance computing workloads. The announcement links storage, cooling and management tools as components of a single system rather than as separate procurement decisions.

What did HPE executives say?

Trish Damkroger, Senior Vice President and General Manager, HPC and AI Infrastructure Solutions, HPE, tied the launch to customer planning horizons. "When making long-term investments in environments where advanced workloads are rapidly evolving, our customers want scalable, dense, and flexible systems to help them address dynamic shifts in requirements," Damkroger said.

She positioned the storage product as the anchor of the portfolio. "Our comprehensive portfolio, which now includes a world-leading hybrid supercomputing storage option, delivers powerful capabilities to scale, while maximizing capacity and optimizing costs."

The two statements map the two halves of the announcement: density and flexibility on the infrastructure side, cost optimisation on the storage side.

When will the products ship?

The release schedule stretches into 2027:

  • Early 2027: new hybrid configurations of HPE Cray Supercomputing Storage Systems E2000
  • March 2027: HPE Rack Management Controller for HPE Cray Supercomputing GX systems, plus the Supercomputing Management Software suite
  • June 2027: AMD and NVIDIA rack-scale system support for the controller

The sequencing signals where HPE expects demand to land. Storage arrives first, matching the immediate pressure from rising SSD prices and growing datasets. Rack management follows, timed to deployments of larger liquid-cooled estates.

What is the competitive position?

The 41% throughput claim and the 82% cost claim place HPE directly against other hybrid supercomputing storage vendors, though the company did not name the competitor it benchmarked against. The 82% figure rests on a defined test case — 15 petabytes of usable storage meeting NVIDIA performance specifications for 1,024 GPUs — which gives buyers a reference point for their own comparisons, even if the configuration is narrower than a general-purpose guarantee.

The controller, meanwhile, competes less on benchmarks and more on integration. Its value depends on the depth of its ties to the HPE Supercomputing Management Software suite and, from June 2027, its reach into AMD and NVIDIA rack-scale systems. Operators weighing single-vendor management against best-of-breed components will watch that AMD and NVIDIA support closely.

For national laboratories and research institutions already running Cray systems, the announcement mostly extends an existing stack. For organisations building their first GPU clusters at the 1,000-plus scale, it packages storage, cooling and management into one procurement path with a dated roadmap — early 2027 for storage, June 2027 for full rack-scale support.

Source: itbrief.com.au

Share this article:

More from Rebecca Stone

Rebecca Stone

Show full bio

Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

728 articles

Related articles

  1. Supermicro Ships Petascale All-Flash Server With PCIe Gen 6
  2. Microsoft prices Surface Laptop Ultra from $2,600, debuts RTX Spark workstation
  3. 62% of Enterprises Can't Meet AI's Storage Demands, Seagate Finds
  4. Lenovo Bets on Liquid Cooling and Services to Solve AI's Power Problem
  5. HPE’s private AI economics: $45 per node-hour, not just tokens

« Previous articleNext article »