HPE is launching new high-performance computing (HPC) and AI solutions aimed at systems with supercomputer allure. The focus is on liquid-cooled systems and servers optimized for AI training and tuning.
HPE is revamping its HPC portfolio with new Cray Supercomputing EX systems. These are designed for complex science and technology workloads. The new portfolio includes:
- EX4252 Gen 2 Compute Blade: which provides up to 98,304 cores in one cabinet, powered by AMD EPYC processors. These servers will be available starting in spring 2025.
- EX154n Accelerator Blade: equipped Nvidia B200 GPUs and Nvidia Grace CPUs. With these blades, up to 224 GPUs can fit in a single cabinet. This solution will also be available in late 2025.
- Slingshot Interconnect 400: offers 400 Gbps network capacity and more efficient congestion management, available from fall 2025.
- Cray Supercomputing Storage Systems E2000: increases I/O performance for large-scale workloads and will be available in early 2025.
The portfolio offers air cooling options and HPE’s proprietary fanless liquid cooling technology. The latter reduces energy consumption and optimizes performance in data centers. Given the immense density of high-performance systems, liquid cooling is the only viable solution in many cases.
read also
Hot water, cool servers: Nvidia’s Blackwell chips as pool heaters
ProLiant
In addition to HPC systems, HPE is launching two new ProLiant Compute XD servers. These are designed specifically for AI training and tuning, with a focus on flexibility and energy efficiency:
- XD680 server: focused on AI training with Intel Gaudi 3 accelerators. This server is ak available from December 2024.
- XD685 server: This one will be available in two flavors: with Nvidia GPUs or AMD Instinct accelerators. The systems will be available in early 2025.
Both servers integrate HPE-optimized cooling and use HPE’s iLO management technology, which facilitates remote management and security.
Big boys
With this portfolio, HPE is targeting organizations looking to accelerate large-scale AI initiatives, such as government agencies and research organizations. HPE emphasizes that the new systems address the growing need for AI-supported supercomputing. That liquid cooling plays an important role in this should come as no surprise. HPC and AI servers today have an unprecedented density, with a lot of computing power but also a lot of heat concentrated in a single cabinet. Removing that heat efficiently is really only possible with liquid cooling.