Careers at OverGPU

Careers at OverGPU center on one job: keeping dedicated GPU servers running well for the people who rent them. We are not listing open roles, and we do not take applications through the website at the moment. The profiles on this page are the people we hire when we grow.

Careers at OverGPU: the profiles we hire

Rather than post roles we are not ready to fill, we describe the profiles we hire as the team grows: data-center and hardware engineers, GPU and HPC support engineers, Linux and network engineers, and customer-facing solutions engineers. Some people sit between two of them. We would rather work with people who already know why they want to work on GPU infrastructure.

What the work involves

We rent whole GPUs, from a single RTX 5090 to 8-GPU HGX B300 servers, in professional data centers with a 99.9% availability SLA and 24/7 support. Keeping those promises is the work.

In the data center, that means installing and testing servers, diagnosing GPU, NVLink, memory and network faults, and replacing failed parts as a priority. In support, it means helping customers untangle driver and CUDA versions, ROCm setups on AMD hardware, containers, and multi-GPU jobs that hang or run slowly; our guide to multi-GPU training with NCCL gives a sense of the problems. In systems and networking, it means maintaining OS images with the right drivers, CUDA and ROCm versions, keeping hosts healthy, and running IPv4 and IPv6 networking, DDoS mitigation and private networks between servers. With customers, it means sizing a workload honestly, even when the right answer is a smaller server, and, for rack-scale reservations, writing the confirmation they pay against.

To see what you would be supporting, read about our GPU server infrastructure and how we work.

What we value

Ownership. When a customer's job is down, the person who picks up the issue follows it to the end, including the follow-up nobody asked for and the fix that stops it from happening again.

Written communication. Our commitments are in writing: configurations, prices, delivery times and the SLA. We expect the same clarity in a support reply, an incident note or a handover. If you can explain a fault in five clear sentences, you will fit in.

Care for customers' workloads. A rented server is someone's training run, product launch or research deadline. We treat their time accordingly, we do not touch their data unless they ask us to operate or repair something or the law requires it, and we tell them the truth about trade-offs.

Applications

We do not take applications through the website at the moment, including open applications, so there is no way to apply from this page.