Dedicated nodes
Reserve whole B300 nodes on bare metal, by the month or longer, with full root access and nothing shared.
EnquireGPU cloud
Dedicated Blackwell Ultra capacity for training, fine-tuning and inference. Reserve a single node or a full cluster — run it yourself on bare metal, or let us manage it end to end.
01The platform
Specifications per NVIDIA HGX B300 node.
02The network
Our B300 clusters are networked to NVIDIA’s reference architecture on NVIDIA Spectrum-X Ethernet, designed and run by our operations partner Amaze.
All eight B300 GPUs in a node share one fifth-generation NVLink domain, so they work together as a single large accelerator.
Every GPU has its own ConnectX-8 SuperNIC, linked over NVIDIA Spectrum-X Ethernet with RDMA (RoCE) for direct GPU-to-GPU transfers across nodes.
Each cluster gets dedicated fabric bandwidth, so throughput stays predictable from one job to the next.
Spectrum-X adaptive routing and congestion control keep all-reduce and other collective traffic flowing evenly at multi-node scale.
03Ways to use it
Reserve whole B300 nodes on bare metal, by the month or longer, with full root access and nothing shared.
EnquireMulti-node clusters linked by high-bandwidth GPU networking, sized for large training and fine-tuning runs.
EnquireShort-term access to prove out a model on B300 before you commit to a longer reservation.
EnquireWe handle provisioning, orchestration, monitoring and lifecycle operations, so your team can focus on the models.
Enquire04Why SYNKO
Workloads run on Australian soil, under Australian jurisdiction.
Our B300 fleet belongs to SYNKO — no resold capacity and no hidden middle layer.
Every customer is screened, and we operate in line with NVIDIA’s requirements and US export controls.
Our sites are planned to step up to NVIDIA’s next platform, Vera Rubin, as it becomes available.
Contact
Reach out about AI computing, IDC hosting, cloud services, or a bespoke deployment across Asia-Pacific.
24-26 Bowden St, Alexandria NSW 2015, Australia