HPC Engineer

Nex Gen Cloud is a fast-growing GPU cloud and AI infrastructure business. We’re the company behind Hyperstack — a self-serve, on-demand cloud platform built for high-performance AI, HPC, and compute-heavy workloads. Powered by 100% renewable energy and NVIDIA architecture, Hyperstack gives teams from AI startups to enterprise the compute they need, without the cost or complexity of legacy cloud.

We move fast, solve hard problems, and build things that matter. If you want ownership, real impact, and a front-row seat to one of the most consequential infrastructure shifts of the decade — this is the place.

We’re deploying more and more large-scale NVIDIA GPU clusters connected by Infini Band, and the gap between a cluster that’s built and one that delivers its expected training performance is where customer experience is won or lost. We’re hiring an HPC Engineer to close that gap: validating and tuning clusters before handover, then owning the HPC layer in production.

This is a role for someone who follows a problem across GPUs, fabric, storage and software until it’s fixed rather than handing it off, who won’t sign off a cluster that isn’t ready, and who’s comfortable operating without heavy oversight in a fast-moving environment. It’s a permanent, remote role based in the UK, with an on-call rota and occasional travel to data centre sites.

Rather than a checklist, here’s what success in this role actually looks like:

• Own cluster validation before handover: running burn-in, health checks and NCCL, HPL and storage benchmarks against agreed acceptance criteria, and resolving underperformance before it reaches the customer
• Collaborate closely with Deployment Solution Architects, Network Engineers, Validation and Support teams, and with NVIDIA and OEMs on hardware faults and RMAs
• Run production HPC clusters day to day, including Slurm, GPU drivers and the CUDA stack, Infini Band fabrics (subnet management, monitoring, firmware) and the parallel storage clusters depend on
• Tune clusters for AI training and HPC workloads, from NCCL settings and GPUDirect RDMA to topology-aware scheduling and storage throughput
• Build GPU, fabric and job-level monitoring that catches faults before customers notice, and lead incident response from Xid errors and fabric degradation through to root cause
• Automate provisioning, configuration and validation with infrastructure-as-code, creating a standard, repeatable validation suite for each GPU platform and fabric type
• Turn incident learning into automated checks, runbooks and documentation others can follow, and feed what you learn in production back into how clusters are designed and deployed

We care more about how you think and work than a perfect CV. You’ll likely bring a combination of the following:

Essential:

• Hands-on experience operating NVIDIA GPU infrastructure at scale, including drivers, CUDA, DCGM and diagnosing GPU hardware faults
• Proven experience running production HPC clusters with a workload manager such as Slurm
• Strong practical Infini Band experience, including subnet management, RDMA, fabric monitoring and fault-finding
• Experience benchmarking and tuning performance for distributed workloads using NCCL tests, HPL or similar
• Strong Linux systems administration skills, including OS-level networking and storage, plus infrastructure automation and Git-based workflows

Nice to Have:

• Experience with large NVIDIA platforms such as HGX H100/H200/B200 or GB200 rack-scale systems, NVIDIA UFM and GPUDirect RDMA
• Parallel file systems such as WEKA, VAST, DDN/Lustre or GPFS
• Running GPU workloads on Kubernetes, or integrating HPC clusters with Kubernetes or Open Stack
• A background supporting AI/ML training,…