HPE is expanding its Cray Supercomputing Storage Systems E2000 with hybrid configurations that combine flash drives and 32TB SAS-4 hard drives. The company says new enclosures holding 74 or 100 drives can deliver up to 32% more usable capacity per rack than its currently shipping E2000 hybrid systems, without increasing the floor space required.
HPE rates a rack of the updated system for up to 116GB/s of read throughput and 92GB/s of write throughput. The company plans to begin shipping the configurations in early 2027. It has not disclosed pricing.
More capacity from a hybrid design
E2000 is HPE’s Lustre-based parallel file system for supercomputing environments. Its previous hybrid configurations supported hard drives with capacities of up to 24TB. The new 32TB SAS-4 drives and higher-density enclosures increase the amount of storage that can fit in a rack, while retaining flash for workloads that need faster access.
HPE vice president of product management for HPC and AI Infrastructure Solutions Marten Terpstra said the throughput figures are based on a configuration with roughly 10% of capacity on flash and 90% on hard drives. HPE adjusts that balance for individual customer workloads, he said.
The company attributes the product update in part to rising costs for high-capacity SSDs, which Terpstra said have increased over the past 12 to 18 months. HPE is targeting research institutions and national laboratories that must divide fixed procurement budgets between compute and storage. A higher-capacity hybrid rack could help those organizations add storage without displacing as many compute nodes, though the actual trade-off depends on each deployment.
HPE also makes two performance and cost comparisons. It claims the new E2000 configuration offers up to 41% more throughput, measured in gigabytes per second per storage rack, than the next hybrid supercomputing storage competitor, which it identifies as DDN. Separately, HPE says its system can have up to 82% lower acquisition cost than the next all-flash supercomputing storage competitor for the same capacity. That comparison is based on the street price of 15PB of usable capacity meeting performance specifications for an NVIDIA 1,024-GPU cluster.
Those figures are HPE’s claims; the company did not publish the competing configurations used to make either comparison. They should therefore be read as vendor-provided benchmarks rather than independently verified results.
Tiering and workload placement
How effectively a hybrid system serves workloads depends in part on how data moves between flash and hard-drive tiers. HPE says E2000 already includes tiering and data movers that monitor which data is actively being used.
Terpstra also described potential closer coordination with job schedulers, which could stage data onto flash before a compute job begins. He presented that as a possible direction for the system, not as a committed feature with a delivery date. For operators, such coordination could help reduce delays when jobs need data held on the slower-capacity tier, but HPE did not provide a specific implementation or performance guarantee.
Controller addresses liquid-cooled racks
Alongside the storage announcement, HPE introduced the Rack Management Controller, a rack-level system for monitoring liquid cooling, detecting leaks and automating responses. Its first supported platform will be the Cray Supercomputing GX5000. HPE says the controller will be offered in variants matched to different rack-scale systems.
The controller is designed to provide real-time monitoring and control, including the ability to respond to a leak by shutting down an individual blade or an entire rack and closing valves. Terpstra said the design builds on a controller HPE has used in Cray EX systems for six to seven years. He cited rack power levels of 300kW to 500kW as an example of the dense environments the controller is intended to manage.
HPE plans to integrate the controller with its Supercomputing Management Software suite. The suite includes Foundation for infrastructure management, Composable for resource sharing and Observability for operational monitoring. HPE says rack telemetry and control will be available through the management environment, which also supports open APIs for customers using their own operational tools.
The controller for Cray Supercomputing GX systems and its software integration are scheduled for March 2027. HPE expects support for AMD rack-scale systems and one additional platform in June 2027, but has not identified the specific AMD systems covered. The company has aligned with AMD and Broadcom on the Helios rack-scale architecture, though it did not say whether that architecture is the AMD platform planned for controller support. Terpstra also said HPE expects the controller could extend to the higher end of the ProLiant DL portfolio for customers buying servers in rack-scale deployments.
