Performance Engineering Intern

Astera Labs · San Jose, CA · Internship · On-site

About Astera Labs

Astera Labs builds connectivity solutions that help power modern AI infrastructure. Our products connect processors, accelerators, memory, and networks in systems where throughput, latency, bandwidth, utilization, and power efficiency directly affect customer outcomes.

The opportunity

Join the Performance Engineering team and help measure, understand, and improve system performance across AI and data-center workloads. You’ll work with real platforms and representative workloads to build reproducible benchmarks, analyze results, investigate bottlenecks, and communicate what the data means.

This internship is designed for someone who enjoys combining systems knowledge, experimentation, and data analysis: define a question, design a fair test, run it repeatedly, inspect the traces and metrics, form a hypothesis, and recommend the next engineering action. You’ll work with experienced engineers on a meaningful project involving CPUs, GPUs, memory, CXL, PCIe, Ethernet, NVMe, or related connectivity systems.

What you’ll do

Learn the platform architecture, workload, benchmark methodology, tools, and success criteria for your assigned project.
Design and execute reproducible performance experiments across hardware configurations, software stacks, workloads, and operating conditions.
Measure and analyze metrics such as throughput, latency, tail latency, IOPS, memory bandwidth, GPU utilization, power, CPU utilization, and system-level performance.
Run or support benchmark suites and workloads such as MLPerf, CXL and memory benchmarks, FIO, MLC, SPEC, database or vector-database workloads, AI inference, and custom application tests.
Explore the performance impact of variables such as GPU count, batch size, memory placement, NUMA configuration, read/write mix, interleaving mode, link topology, and workload size.
Capture and analyze system traces, PCIe or GPU data flows, memory behavior, logs, counters, and other evidence to identify bottlenecks and performance regressions.
Develop Python or shell scripts to automate test execution, collect results, extract metrics from logs, convert data to analysis-ready formats, and generate plots or reports.
Compare results across configurations, reference systems, software versions, or competitor platforms while documenting assumptions and ensuring valid comparisons.
Investigate unexpected results by reproducing the behavior, narrowing the variables, and working with system, hardware, software, validation, and product teams to identify root causes.
Help build repeatable performance test plans and methodologies that can be applied to new GPUs, systems, workloads, or product generations.
Create clear technical documentation that explains the setup, workload, metrics, results, limitations, and recommendations.
Present your project, results, and key learnings at the end of the internship.

What you bring

Coursework, projects, research, or hands-on experience in computer science, computer engineering, electrical engineering, data science, or a related field.
Foundational understanding of computer architecture, operating systems, computer networks, memory systems, GPUs, storage, or distributed systems.
Programming or scripting experience in Python, C/C++, Bash, or another language used for automation and data analysis.
Comfort working in Linux and using command-line tools, configuration files, logs, and source control.
Curiosity about why a system performs the way it does and a methodical approach to designing experiments and interpreting evidence.
Ability to work with data, identify trends and anomalies, question assumptions, and communicate technical conclusions clearly.
Ownership, persistence, and a collaborative mindset in an environment where performance questions often require iteration across multiple layers of the stack.

Helpful experience

Benchmarking, profiling, performance modeling, workload characterization, or system-level debugging.
MLPerf, FIO, Intel Memory Latency Checker, Linux perf, Nsight, NCCL, SPEC, or related performance tools.
CPU/GPU architecture, PCIe, CXL, Ethernet, RDMA, NUMA, memory bandwidth, storage, or interconnects.
Throughput, latency, IOPS, bandwidth, QPS, p99 latency, utilization, power, or performance-per-watt analysis.
Python data analysis and visualization with tools such as pandas, NumPy, Matplotlib, or comparable libraries.
Slurm or other distributed-compute environments, automated test infrastructure, or CI workflows.
Experience presenting technical results through concise reports, dashboards, plots, or demonstrations.

Internship experience and compensation

This is a paid, hourly U.S. internship. The applicable hourly rate is determined as part of the offer process and specified in the offer letter, depending on education level ranges from $35/hour to $55/hour. U.S. interns currently receive a $500-per-week housing/relocation stipend for the duration of the internship. Interns participate in Early Career programming, including technical and professional-development sessions, community events, and an end-of-program project presentation.

Equal opportunity

Astera Labs is committed to providing equal employment opportunities to all applicants and employees. We value diverse perspectives and consider qualified candidates without regard to legally protected characteristics.

The opportunity

What you’ll do

Learn the platform architecture, workload, benchmark methodology, tools, and success criteria for your assigned project.

Design and execute reproducible performance experiments across hardware configurations, software stacks, workloads, and operating conditions.

Measure and analyze metrics such as throughput, latency, tail latency, IOPS, memory bandwidth, GPU utilization, power, CPU utilization, and system-level performance.

Run or support benchmark suites and workloads such as MLPerf, CXL and memory benchmarks, FIO, MLC, SPEC, database or vector-database workloads, AI inference, and custom application tests.

Explore the performance impact of variables such as GPU count, batch size, memory placement, NUMA configuration, read/write mix, interleaving mode, link topology, and workload size.

Capture and analyze system traces, PCIe or GPU data flows, memory behavior, logs, counters, and other evidence to identify bottlenecks and performance regressions.

Develop Python or shell scripts to automate test execution, collect results, extract metrics from logs, convert data to analysis-ready formats, and generate plots or reports.

Compare results across configurations, reference systems, software versions, or competitor platforms while documenting assumptions and ensuring valid comparisons.

Investigate unexpected results by reproducing the behavior, narrowing the variables, and working with system, hardware, software, validation, and product teams to identify root causes.

Help build repeatable performance test plans and methodologies that can be applied to new GPUs, systems, workloads, or product generations.

Create clear technical documentation that explains the setup, workload, metrics, results, limitations, and recommendations.

Present your project, results, and key learnings at the end of the internship.

What you bring

Coursework, projects, research, or hands-on experience in computer science, computer engineering, electrical engineering, data science, or a related field.

Foundational understanding of computer architecture, operating systems, computer networks, memory systems, GPUs, storage, or distributed systems.

Programming or scripting experience in Python, C/C++, Bash, or another language used for automation and data analysis.

Comfort working in Linux and using command-line tools, configuration files, logs, and source control.

Curiosity about why a system performs the way it does and a methodical approach to designing experiments and interpreting evidence.

Ability to work with data, identify trends and anomalies, question assumptions, and communicate technical conclusions clearly.

Ownership, persistence, and a collaborative mindset in an environment where performance questions often require iteration across multiple layers of the stack.

Helpful experience

Benchmarking, profiling, performance modeling, workload characterization, or system-level debugging.

MLPerf, FIO, Intel Memory Latency Checker, Linux perf, Nsight, NCCL, SPEC, or related performance tools.

CPU/GPU architecture, PCIe, CXL, Ethernet, RDMA, NUMA, memory bandwidth, storage, or interconnects.

Throughput, latency, IOPS, bandwidth, QPS, p99 latency, utilization, power, or performance-per-watt analysis.

Python data analysis and visualization with tools such as pandas, NumPy, Matplotlib, or comparable libraries.

Slurm or other distributed-compute environments, automated test infrastructure, or CI workflows.

Experience presenting technical results through concise reports, dashboards, plots, or demonstrations.

Internship experience and compensation