Cybersecurity laboratory - target computer

1) Product properties regarding the 28 Cybersecurity test target computer units: Requirement that - the system be ultra-small form factor, applicable in desktop environments. Maximum 155x155x55mm – USFF. - the system sustainably provide continuous high AI computing performance, with a capacity of at least 980 TOPS in inference and fine-tuning profiles, and a minimum of ~0.9 petaFLOP FP4 computing performance. - the system have a high-performance CPU built on the Arm v9.2-A architecture, with at least 18 cores and a clock speed of 3.0 GHz. - the system have a dedicated GPU for AI tasks, containing at least 5th generation (or equivalent) Tensor Core units, and supporting the FP4 computing format. - the CPU and GPU operate in close integration, with a unified memory architecture. NVIDIA GB10 Grace–Blackwell Superchip or an equivalent solution can be considered equivalent. - the system have at least 120 GB of high-bandwidth system memory, uniformly accessible to the CPU and GPU. - the system have internal storage capability in the form of at least 960 GB NVMe SSD, with an M.2 interface and a minimum PCIe Gen4x4 standard. - the system ensure high-speed CPU–GPU data transfer using a coherent unified memory architecture, with a data transfer bandwidth of at least 650 GB/s. NVIDIA NVLink C2C or equivalent technology can be considered equivalent. - the system be scalable for running large language models, and have an interconnect solution providing at least 160 Gb/s network bandwidth, enabling the connection of multiple processing units. The solution must support the execution of language models with at least 160 billion parameters, depending on workload and quantization settings. NVIDIA ConnectX-7 or equivalent technology can be considered equivalent. - the system have wired network connectivity, providing at least a 10 Gigabit Ethernet LAN port. - the system also have wireless connectivity, with at least Intel Wi-Fi 7 (Gig+) 2x2 + Bluetooth 5.4 support. - the system have energy-efficient and reliable cooling, ensuring sustainable stable high performance even in a compact design. The power consumption of individual units of the proposed system must not exceed 250 W. - the system run on a Linux-based operating system. Ubuntu Linux or an equivalent distribution can be considered equivalent. - the system support an AI development environment that includes leading AI frameworks (e.g., CUDA or equivalent computing platform, PyTorch, TensorFlow, Jupyter), and provides containerization and cache optimization capabilities. NVIDIA AI Software Stack (NIM™, Blueprints, TensorRT, Ollama, full NVIDIA AI ecosystem support) or an equivalent solution can be considered equivalent. - The device must have the CE certification required for placing on the market in the European Union, as well as at least one internationally recognized conformity certificate related to intended use (e.g., BSMI, CB, FCC, UL, CCC, C-Tick, Wi-Fi, RF, VCCI). Other certificates equivalent to the listed ones are also acceptable if they attest to the same safety and quality requirements. 2) Provision of a description in English or Hungarian; 3) Commitment to a minimum of 36 months warranty in accordance with Government Decree No. 235/2024. (VIII. 8.). Extended warranty is an evaluation criterion. 4) For the purpose of comparing offers, the contracting authority has prepared a specification table in Word format. The professional offer must be submitted by filling out this specification table. The table must record the name, manufacturer, country of origin of the offered device, and the specific type of product must be indicated. The fact of custom-made production must also be stated in the specification...