Bitten: the computing power behind UCloud.
Bitten is a national AI and high-performance computing facility accessible through UCloud.
Bitten at a glance
Bitten combines large-scale computing capacity with flexible resource options for workloads ranging from transcription and smaller AI models to distributed training and advanced simulations.
- 20 CPU nodes, each equipped with two AMD EPYC 9635 processors at 2.25 GHz and 768 GB DDR5-6000 memory.
- 16 GPU nodes, each equipped with two AMD EPYC 9655 processors at 2.6 GHz, 2,304 GB DDR5-6400 memory and eight NVIDIA Blackwell B200 GPUs.
- 128 NVIDIA Blackwell B200 GPUs in total, each providing 192 GB of high-speed GPU memory for AI and other computationally demanding workloads.
- Flexible GPU sizes using NVIDIA MIG, allowing users to request between one and four smaller, isolated GPU instances instead of reserving an entire GPU.
- Up to seven MIG instances per B200 GPU, corresponding to a theoretical maximum of 896 smaller GPU instances when all GPUs are fully partitioned. Actual availability depends on the current system configuration and demand.
- 400 Gbit/s InfiniBand connectivity, enabling fast communication between nodes for distributed AI training, large-scale simulations and other multi-node workloads.
Kubernetes cluster is hosted and maintained at SDU. This provides the default storage and compute resources on the UCloud platform.
