Senior Solutions Architect, HPC and AI
RemoteBerlin, State of Berlin, Germany or France
Job Summary
Collaborate with NVIDIA's training framework developers to adopt new features and assist partners in scaling AI workloads on large-scale GPU clusters. Deploy, debug, and optimize training and inference workloads while benchmarking framework performance and sharing actionable insights with customers and internal teams. Directly engage external clients to resolve cluster stability issues, identify bottlenecks, and implement effective solutions for high-performance computing challenges. Contribute to Europe's Sovereign AI initiative by guiding customers in implementing advanced resiliency features within AI training pipelines.
Required Qualifications
- BS, MS, PhD or equivalent experience in Computer Science, Electrical/Computer Engineering, Physics, Mathematics, or a related engineering field—or equivalent practical experience
- 8+ years of experience in accelerated computing technologies at cluster scale, ideally including work with NVIDIA platforms
- Strong programming skills in at least one of the following languages: C, C++, or Python
- Practical experience identifying and resolving bottlenecks in large-scale training workloads or parallel applications
- Hands-on experienced in profiling and debugging large parallel applications
- Solid understanding of CPU and GPU architectures, CUDA, parallel filesystems, and high-speed interconnects
- Experienced in working with large compute clusters with an understanding of their internal scheduling and resource management mechanisms (e.g. SLURM or Cloud based clusters)
- Proficient knowledge of training pipelines and frameworks, encompassing their internal operations and performance attributes
Desired Qualifications
- Experience in debugging training pipelines running on thousands of GPUs in production environment
- Hands-on experience with performance profiling and optimizations using tools like Nsight Systems, Nsight Compute and good understanding of NCCL, MPI and low-level communication libraries
- Ability to debug stability issues across the entire stack: parallel application, training frameworks, runtime libraries, schedulers, and hardware
- Solid understanding of the internal workings of LLM frameworks such as PyTorch, Megatron-LM, or NeMo, and how they affect compute layers like CPUs, GPUs, network and storage or understanding of inference tools such as vLLM, Dynamo, TensorRT-LLM, RedHat Inference Server or SGLang
Hiring someone like this?
Get your role in front of qualified candidates on Sorce.