https://docs.hpc.gwdg.de/services/ftp/index.htmlThe Future Technology Platform (FTP) is a service that offers a test platform for researchers and developers to work with advanced and prototype architectures. Among other systems, FTP provides access to: ET-SoC-1 Platform Gaudi2 Graphcore Neuromorphic Computing NVIDIA Bluefield-2 DPU NVIDIA GH200 Grace Hopper If you have any question, feel free to contact us at support@gwdg.de. Or join our community chat on Matrix.Hugode-dehttps://docs.hpc.gwdg.de/services/ftp/esperanto/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://docs.hpc.gwdg.de/services/ftp/esperanto/index.htmlIntroduction The ET-SoC-1 (ET) is an experimental “manycore” chip originally developed by Esperanto Technologies for applications in HPC and AI. In 2025, Ainekko has acquired the IP and plans to open-source the platform. The design leverages over 1000 RISC-V “ET-Minion” processing cores on a single chip for massive parallelization of workloads. Each core contains a vector processing unit (VPU) as well as a tensor unit (TU) specifically optimized for machine learning operations. A network-on-chip (NoC) interconnects the “ET-Minion” cores and 32 GB of distributed, energy-efficient LPDDR4X RAM allow for high throughput in a power envelope of around 40 W per card. Each card is connected to the system via a PCIe 4.0 x8 interface. On the FTP, we currently host 4 compute nodes equipped with 8 ET-SoC-1 cards each, which are available for researchers and developers especially to evaluate AI deployment on the edge and energy-efficient small language models.https://docs.hpc.gwdg.de/services/ftp/gaudi2/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://docs.hpc.gwdg.de/services/ftp/gaudi2/index.htmlIntroduction Gaudi2 is Intel’s second-generation deep learning accelerator, developed by Habana Labs (now part of Intel). Unlike traditional GPUs, Gaudi2 has been designed from the ground up for large-scale AI training. Each device is powered by Habana Processing Units (HPUs), its purpose-built AI training cores. The memory-centric architecture and Ethernet-based scale-out enable efficient training of today’s large and complex models, while offering a favorable power-to-performance ratio. The platform provides 96 GB of on-chip high-bandwidth memory per device, together with 24×100 Gbps standard Ethernet interfaces. This combination eliminates the need for proprietary interconnects and allows flexible integration into existing cluster infrastructures. On the FTP, we currently host a single Gaudi2 node equipped with 8 HL-225 HPUs, available for researchers and developers to evaluate distributed AI training.https://docs.hpc.gwdg.de/services/ftp/graphcore/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://docs.hpc.gwdg.de/services/ftp/graphcore/index.htmlGraphcore Intelligence Processing Unit (IPU) is a highly parallel processor which is specifically designed to accelerate Machine Learning and Artificial Intelligence applications. IPU has a unique memory architecture which allows it to hold much more data within IPU than other processors. IPU-Machine is a compute platform consisting of 1U chassis that includes 4 IPUs and up to 260 GB of memory. IPU-Machines can also be used to make larger compute systems. Multiple IPUs can be used together on a single task where they communicate through IPU-Fabric as shown in the image below.https://docs.hpc.gwdg.de/services/ftp/neuromorphic-computing/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://docs.hpc.gwdg.de/services/ftp/neuromorphic-computing/index.htmlNeuromorphic Computing Tools and Libraries SpiNNaker Neuromorphic computing is an alternative way of computing, centered around the concept of the spiking neuron, inspired by the way biological neurons work. It can be used not only to perform simulations of nervous tissue, but also to solve constraint and graph optimization problems, run network simulations, process signals in real time, and perform various AI/ML tasks. Additionally, it is known to require lower energy consumption when compared to more traditional algorithms and computing architectures. For more information, please read the article in the January/February 2024 issue of GWDG News.https://docs.hpc.gwdg.de/services/ftp/bluefield/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://docs.hpc.gwdg.de/services/ftp/bluefield/index.htmlIntroduction The BlueField‑2 DPU is a purpose‑built processor that moves networking, storage and security functions off the host CPU, giving you more cycles for compute‑intensive workloads. How to Get Access To get access to a Bluefield node, please contact us at hpc-support@gwdg.de. For any other questions, suggestions or feedback, you can also get in touch with us via our community chat on Matrix. Resources Bluefield-2 DPU Datasheet Siddharth Simediya: Benchmarking Network Acceleration in Kubernetes Clusters with 2nd Gen NVIDIA BlueField DPUshttps://docs.hpc.gwdg.de/services/ftp/gh200/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://docs.hpc.gwdg.de/services/ftp/gh200/index.htmlIntroduction NVIDIA’s GH200 Grace Hopper superchip fuses the Grace CPU and Hopper GPU with NVIDIA NVLink‑C2C, creating a single, coherent CPU‑GPU memory space. This design theoretically enables workloads to run faster and with less programming overhead compared to traditional systems with separate, discrete CPUs and GPUs. Feature What it means for you Coherent memory model CPU and GPU can share data directly, no explicit copies or staging buffers 900 GB/s coherent interconnect Up to 7x the bandwidth of PCIe Gen5, delivering fast memory access for data‑intensive tasks HBM3 / HBM3e GPU memory High‑capacity, high‑bandwidth memory that accelerates large ML models and simulations GH200 is used, e.g., in Forschungszentrum Jülich’s JUPITER exascale class supercomputer and NVIDIA’s DGX GH200 system.