About the role
Positions: 1 | Experience: 5+ years
About the role You will design high-throughput GPU networking and distributed KV-cache systems for inference.
Key responsibilities
- Design GPU interconnect and high-speed network topologies.
- Configure and tune RDMA, InfiniBand, RoCE v2 and Ethernet communications.
- Evaluate LMCache, Mooncake, NIXL and remote KV-cache architectures.
- Optimise transfers among GPU HBM, host DRAM and remote memory nodes.
- Diagnose bandwidth, congestion, packet loss and latency issues.
- Tune NCCL/RCCL communication and evaluate 100G/200G/400G or faster links.
- Test network failures, recovery and distributed cache behaviour.
Required technical skills Linux networking, RDMA verbs, InfiniBand, RoCE v2, ConnectX NICs, NCCL/RCCL, C/C++, Python, GPU memory architecture.
- Expected deliverables Network architecture, KV-cache POCs, capacity recommendations and validated transfer performance.
Skills
Interested in this role?
You'll apply directly on Spydra's careers site.