Optimize distributed AI training and inference workloads across Cornelis Omni-Path fabrics and GPU clusters.
About the role
As an AI Performance Engineer, you will measure and optimize training and multi-node inference across networking silicon and systems, including adapters, switches, and the supporting software stack. You will partner with architecture, firmware, software, and customers to evaluate distributed workloads on CPU and GPU clusters.
What you'll do
- Measure and optimize end-to-end performance for distributed AI training and multi-node inference across CPU and GPU clusters and fabrics including Omni-Path, Ethernet, and InfiniBand.
- Benchmark, characterize, and tune workloads and models such as Llama and Qwen using frameworks including Megatron-LM, TorchTitan, Composer, vLLM, and llama.cpp.
- Design and optimize distributed serving topologies, covering tensor and pipeline parallelism, MoE expert placement, continuous and adaptive batching, and KV-cache sharding and offload.
- Validate RDMA, GPUDirect RDMA, congestion control, and collective and point-to-point communication tradeoffs.
- Instrument and visualize performance using Nsight Systems, VTune, perf, application timers, Prometheus, and Grafana.
- Compare Cornelis Omni-Path to competing interconnects and produce technical documentation, briefs, READMEs, blogs, and peer-reviewed publications.
What you'll need
- BS in CS, EE, CE, Math, or related field and 2-3+ years of experience running AI/ML workloads at cluster scale.
- Hands-on experience with distributed training beyond single-GPU (DP, TP, PP, ZeRO, FSDP, sharded optimizers) and distributed inference architectures.
- Practical experience across AI frameworks and communications libraries, including PyTorch, DeepSpeed, Megatron-LM, RCCL, NCCL, vLLM, and llama.cpp.
- Experience with compilers (GCC, LLVM, Intel, OneAPI), MPI stacks, Python, and shell scripting.
- Familiarity with network architectures (Omni-Path/OPA, InfiniBand, Ethernet, RDMA, RoCE) and Linux system performance tuning, including NIC offloads, CQ moderation, pacing, and ECN.
Nice to have
- Cluster scheduler experience with SLURM or PBS.
- Hands-on profiling and tracing of GPU and communication paths using Nsight Systems, Nsight Compute, ROCm tools (rocprof, roctracer), VTune, or perf.
- Background in HPC performance engineering or storage technologies such as BeeGFS, Lustre, or NVMe-oF.
Skills: Omni-Path, RDMA, InfiniBand, NCCL, distributed training
More semiconductor jobs: San Jose jobs
This role has been open 12 days — well below the 38-day median for Infrastructure/Platform Software roles.
Infrastructure/Platform Software · Networking Software
|
Open roles in category
60
|
Median days open
38 d
|
Category median salary (all employers)
$264k
|
See the full market breakdown ▾Category comparison, skills in demand, and who else is hiring
| Metric | Cornelis Networks | All employers we track in this specialty (60 roles · 12 employers) |
|---|---|---|
| Open roles in this specialty | 1 | 60 |
| Open roles in the wider Software, Firmware & Systems family | 9 | 6531 · 156 employers |
| Median days open | 12 d | 38 d (−26 d vs this employer) |
| Median salary (USD postings) | — | $264k |
Skills observed across this category: Omni-Path, RDMA, InfiniBand, NCCL, distributed training
Who's hiring in this category
- NVIDIA · 40 open roles · median 38 d
- Arm Holdings · 3 open roles · median 191 d
- Eridu AI · 3 open roles · median 157 d
- AMD · 2 open roles · median 39 d
- Broadcom · 2 open roles · median 36 d
- Keysight Technologies · 2 open roles · median 7 d
How we counted: 60 open Infrastructure/Platform Software (Networking Software) roles from 12 employers tracked in the SemiconductorJobs index, counted 8 Nov 2026. Specialty figures count only roles carrying this exact specialty label, so an employer's related work in neighbouring specialties is not included there — it is counted in the wider Software, Firmware & Systems family row. Figures refresh nightly.