Solutions Architect with GPU infrastructure
astir it solutionsElizabeth (NJ)
Solutions Architect with GPU infrastructure
Posted 5 days ago
astir it solutionsElizabeth (NJ)
SENIORITY
Manager
About the role
Role: Lead Solution Architect Location: Remote
Contract Position. Key Performance Indicators· Technical win rate and contribution to closed contract value on strategic US GPUaaS pursuits· Size and complexity of deals shaped (committed capacity, multi-year value)· Accuracy of sizing versus delivered performance and cost, and quality of delivery outcomes· Team adoption of reference architectures, sizing tools and benchmark standards· Partner standing with NVIDIA and AMD, thought leadership output, and growth of architects mentored
Required Experience:
- 15+ years in enterprise infrastructure, cloud or solution architecture, including customer-facing pre-sales or consulting roles.· GPU infrastructure (mandatory): 5+ years hands-on designing, deploying or operating GPU clusters.· GPU selling: 1+ years selling GPU capacity, GPU cloud or AI infrastructure, with technical wins on multi-year GPU commitments.· AI data center: Understanding of AI data center nuances (high-density power, liquid cooling, power availability and lead times, facility readiness) and how they constrain GPU design, cost and timelines.· Scaling AI workloads: Experience scaling production AI workloads from single node to multi-node and multi-cluster, including autoscaling, multi-tenant GPU sharing and capacity planning.· LLM inference: 2+ years deploying and operating LLM inference in production (such as vLLM, SGLang, TensorRT-LLM or NIM).· Hands-on coding: Strong, current hands-on coding in Python, plus Bash and Infrastructure as Code (Terraform, Ansible, Helm). Personally builds POCs, benchmark harnesses, sizing tools and inference deployments, not only designs them.· Hands-on technical skills: Personally deploys and troubleshoots GPU environments: Linux administration, GPU drivers and CUDA/ROCm, containers, Kubernetes (GPU Operator, Helm), standing up and tuning serving engines (vLLM, SGLang), running benchmarks (GenAI-Perf, NCCL tests) and monitoring (DCGM, Prometheus, Grafana).· Technical Expertise· GPU infrastructure· Accelerators and their sizing characteristics (HBM, memory bandwidth, scale-up domain, power): NVIDIA H100/H200, B200/B300, GB200/GB300 NVL72; AMD Instinct MI300X/MI325X/MI355X; roadmap awareness (Vera Rubin, MI400).· Networking and storage: NVLink/NVSwitch, InfiniBand and RoCEv2/Spectrum-X, SuperNICs/DPUs; parallel and all-flash file systems (WEKA, VAST, DDN) and GPUDirect Storage.· Orchestration and multi-tenancy: Kubernetes with NVIDIA GPU Operator and AMD ROCm equivalents, MIG, time-slicing, bare-metal provisioning on OpenStack, tenant isolation and quotas.· Data center and validation: 40 kW to 140 kW+ racks, direct-to-chip liquid cooling, burn-in and acceptance testing (DCGM, NCCL/RCCL), NVIDIA and AMD reference architectures.· Inference· Serving engines: vLLM, SGLang, TensorRT-LLM, Triton and NIM, AMD ROCm stack; KServe and Gateway API Inference Extension; disaggregated serving (NVIDIA Dynamo, llm-d).· Optimization: continuous batching, KV-cache management, prefix caching, speculative decoding and quantization (FP8, FP4, INT4).· Sizing and benchmarking: GPU count, type and parallelism (tensor, pipeline, expert) to model, context length, traffic and SLA; TTFT, inter-token latency, goodput and cost per million tokens.· Open-weight models (Llama, Qwen, DeepSeek, Mistral, gpt-oss), including MoE and long-context models, plus RAG, agentic and model-gateway patterns. Thank you, Bharath MAccount Manager
Before you apply
Applying takes about a minute. These four things decide how fast it moves after that.
Your profile is current
It's what we read first. Occupations, seniority and locations matter more than a long history.
Two examples you can talk through
Not a portfolio — just two pieces of work where you can explain the decisions and what you'd change.
A number in mind
What you're on now and what would make you move. We negotiate better when we know both.
Your notice period
Employers plan around it, and it's the question that stalls offers most often.
Once you apply, someone reads it and calls you before anything reaches the employer — usually within two working days.