Lead Performance and Optimization Engineer

AMD · Bangalore

Collected role description

Role overview

Lead Performance and Optimization Engineer at AMD

Performance Engineering & Optimization

THE ROLE:

The role includes handson work in performance debugging, OS/BIOS tuning, thread/core affinity, multiinstance execution, and Python/scriptingbased automation. .

The ideal candidate should be passionate about software engineering and possess leadership skills to drive sophisticated issues to resolution.

Responsibilities

Run and optimize ML inference workloads on CPUs using EPYC optimized libraries ( ZenDNN ), improving throughput/latency across single instance and multi instance scenarios.
Configure and tune NUMA, HugePages , SMT, power/performance modes, CPU isolation, scheduler settings, scaling governors, and other OS/ BIOS parameters.
Design and validate thread/core affinity strategies for single instance , multi instance , multi socket , and framework level multi instance execution models.
Optimize workload behaviour through NUMA aware locality, thread scheduling/pinning, batch size tuning, operator level parallelism, and other CPU focused techniques.
Contribute to multi instance execution framework development, including policies for instance partitioning, core allocation, memory distribution, and orchestration of parallel runs on large EPYC systems.
Benchmarking & Analysis
Develop and run structured benchmarks across EPYC SKUs, core counts, caching/topology variations, sockets, and diverse batch sizes.
Analyze scaling for single instance vs. multi instance execution, instance placement strategies, and workload isolation.
Use perf, VTune , ftrace / trace cmd , PMU counters, flame graphs to identify bottlenecks in compute, memory, thread scheduling, or instance level competition.
Perform root cause analysis for regressions in latency, throughput, multi instance efficiency, memory bandwidth, and pipeline behaviour .
ML Inference Domain Knowledge
Understand how ML frameworks execute models on CPU, including tensor shapes/layouts, operator behavior , threading models, kernel dispatch, scheduling strategies, and multi instance runtime interactions.
Interpret how model architecture and operator composition influence performance across single and multiple concurrent inference instances.
Collaborate with ZenDNN and kernel/ops teams to relay findings and help guide kernel/ operator level improvements.
Automation & Tooling
Build automation pipelines for single instance and multi instance benchmarking, profiling, orchestration, scaling studies, and regression detection.
Develop Python/Bash tooling that manages instance spawning, CPU core partitioning, memory pinning, performance data capture, reporting, and visual dashboards.
Maintain reproducible experiment workflows for both single instance and multi instance configurations.

Required qualifications

Strong understanding of CPU architecture: pipelines, caches, TLB, NUMA, SMT/HT, vector units (AVX2/AVX 512/VNNI/BF16/INT8), and memory hierarchy.
8 to 12 years in performance engineering, systems optimization, or low level execution on Linux.
Hands on experience with Linux tuning and server class OS/ BIOS configuration.
Proficiency with perf, VTune , PMU counters, ftrace / trace cmd , flame graphs, and multi instance profiling.
Strong knowledge of ML inference execution (tensors, operators, threading models) on CPU backends.
Strong Python and Bash for automation and performance tooling.
Experience in multi core scaling, thread affinity, scheduler behavior , concurrency techniques, and multi instance execution strategies.
Familiarity with PyTorch , TensorFlow, ONNX Runtime for running inference workloads.
#LI-SG2
Benefits offered are described: AMD benefits at a glance .
AMD does not accept unsolicited resumes from headhunters, recruitment agencies, or fee-based recruitment services.
AMD and its subsidiaries are equal opportunity, inclusive employers and will consider all applicants without regard to age, ancestry, color, marital status, medical condition, mental or physical disability, national origin, race, religion, political and/or third-party affiliation, sex, pregnancy, sexual orientation, gender identity, military or veteran status, or any other characteristic protected by law.
We encourage applications from all qualified candidates and will accommodate applicants’ needs under the respective laws throughout all stages of the recruitment and selection process.
AMD may use Artificial Intelligence to help screen, assess or select applicants for this position.
AMD’s “Responsible AI Policy” is available here.
This posting is for an existing vacancy.
Minimum qualification: Benefits offered are described: AMD benefits at a glance . AMD does not accept unsolicited resumes from headhunters, recruitment agencies, or fee-based recruitment services. AMD and its subsidiaries are equal opportunity, inclusive employers and will consider all applicants without regard to age, ancestry, color, marital status, medical condition, mental or physical disability, national origin, race, religion, political and/or third-party affiliation, sex, pregnancy, sexual orientation, gender identity, military or veteran status, or any other characteristic protected by law. We encourage applications from all qualified candidates and will accommodate applicants’ needs under the respective laws throughout all stages of the recruitment and selection process. AMD may use Artificial Intelligence to help screen, assess or select applicants for this position. AMD’s “Responsible AI Policy” is available here. This posting is for an existing vacancy.

Employment details

Location: Bangalore [India]

About the employer

WHAT YOU DO AT AMD CHANGES EVERYTHING
When you join AMD, you’ll discover the real differentiator is our culture.
Able to communicate effectively and work optimally with different teams across AMD.

Additional role details

At AMD, our mission is to build great products that accelerate next-generation computing experiences—from AI and data centers, to PCs, gaming and embedded systems.
Grounded in a culture of innovation and collaboration, we believe real progress comes from bold ideas, human ingenuity and a shared passion to create something extraordinary.
We push the limits of innovation to solve the world’s most important challenges—striving for execution excellence, while being direct, humble, collaborative, and inclusive of diverse perspectives.
Join us as we shape the future of AI and beyond. Together, we advance your career.
Lead Performance and Optimization Engineer
We are seeking a Performance Engineer with strong expertise in serverclass CPUs, CPU microarchitecture, and ML inference, responsible for benchmarking, analysing, and optimizing CPU inference performance using EPYCoptimized ML libraries (e.g., ZenDNN) with common frameworks (PyTorch, TensorFlow, ONNX Runtime).
THE PERSON:

Collected minimum qualification

Benefits offered are described:  <a href="https://www.amd.com/en/corporate/careers/benefits.html" target="_blank" rel="noopener">AMD benefits at a glance</a>. AMD does not accept unsolicited resumes from headhunters, recruitment agencies, or fee-based recruitment services. AMD and its subsidiaries are equal opportunity, inclusive employers and will consider all applicants without regard to age, ancestry, color, marital status, medical condition, mental or physical disability, national origin, race, religion, political and/or third-party affiliation, sex, pregnancy, sexual orientation, gender identity, military or veteran status, or any other characteristic protected by law.   We encourage applications from all qualified candidates and will accommodate applicants’ needs under the respective laws throughout all stages of the recruitment and selection process. AMD may use Artificial Intelligence to help screen, assess or select applicants for this position.  AMD’s “Responsible AI Policy” is available <a href="https://www.amd.com/en/legal/responsible-ai-use-policy.html" target="_blank" rel="noopener">here.</a> This posting is for an existing vacancy.

Collected preferred qualification

Lead Performance and Optimization Engineer  THE ROLE: We are seeking a Performance Engineer with strong expertise in serverclass CPUs, CPU microarchitecture, and ML inference, responsible for benchmarking, analysing, and optimizing CPU inference performance using EPYCoptimized ML libraries (e.g., ZenDNN) with common frameworks (PyTorch, TensorFlow, ONNX Runtime). The role includes handson work in performance debugging, OS/BIOS tuning, thread/core affinity, multiinstance execution, and Python/scriptingbased automation.  .<span style="margin: 0px; padding: 0px;" data-ccp-charstyle="eop" data-ccp-charstyle-defn="{"ObjectId":"065fb625-426b-4ae2-8ee9-a7c4873db88c|130","ClassId":1073872969,"Properties":[469775450,"eop",201340122,"1",134233614,"true",469778129,"eop",335572020,"1",469778324,"Default Paragraph Font"]}">   THE PERSON: The ideal candidate should be passionate about software engineering and possess leadership skills to drive sophisticated issues to resolution. Able to communicate effectively and work optimally with different teams across AMD.   KEY RESPONSIBILITIES: <ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 24px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" data-leveltext="" data-font="Symbol" data-listid="5" data-list-defn-props="{"335552541":1,"335559684":-2,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"hybridMultilevel"}" aria-setsize="-1" data-aria-posinset="1" data-aria-level="1">Performance Engineering & Optimization <ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 48px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="7" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="1" data-aria-level="1">Run and optimize ML inference workloads on CPUs using EPYCoptimized libraries (ZenDNN), improving throughput/latency across singleinstance and multiinstance scenarios. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 48px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="7" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="2" data-aria-level="1">Configure and tune NUMA, HugePages, SMT, power/performance modes, CPU isolation, scheduler settings, scaling governors, and other OS/BIOS parameters. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 48px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="7" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="3" data-aria-level="1">Design and validate thread/core affinity strategies for singleinstance, multiinstance, multisocket, and frameworklevel multiinstance execution models. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 48px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="7" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="4" data-aria-level="1">Optimize workload behaviour through NUMAaware locality, thread scheduling/pinning, batch size tuning, operatorlevel parallelism, and other CPUfocused techniques. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 48px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="7" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="5" data-aria-level="1">Contribute to multiinstance execution framework development, including policies for instance partitioning, core allocation, memory distribution, and orchestration of parallel runs on large EPYC systems. </li></ul>Benchmarking & Analysis <ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 48px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="6" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="1" data-aria-level="1">Develop and run structured benchmarks across EPYC SKUs, core counts, caching/topology variations, sockets, and diverse batch sizes. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 48px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="6" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="2" data-aria-level="1">Analyze scaling for singleinstance vs. multiinstance execution, instance placement strategies, and workload isolation. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 48px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="6" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="3" data-aria-level="1">Use perf, VTune, ftrace/tracecmd, PMU counters, flame graphs to identify bottlenecks in compute, memory, thread scheduling, or instancelevel competition. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 48px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="6" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="4" data-aria-level="1">Perform rootcause analysis for regressions in latency, throughput, multiinstance efficiency, memory bandwidth, and pipeline behaviour. </li></ul>ML Inference Domain Knowledge <ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 48px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="10" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="1" data-aria-level="1">Understand how ML frameworks execute models on CPU, including tensor shapes/layouts, operator behavior, threading models, kernel dispatch, scheduling strategies, and multiinstance runtime interactions. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 48px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="10" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="2" data-aria-level="1">Interpret how model architecture and operator composition influence performance across single and multiple concurrent inference instances. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 48px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="10" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="3" data-aria-level="1">Collaborate with ZenDNN and kernel/ops teams to relay findings and help guide kernel/operatorlevel improvements. </li></ul>Automation & Tooling <ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 48px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="2" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="1" data-aria-level="1">Build automation pipelines for singleinstance and multiinstance benchmarking, profiling, orchestration, scaling studies, and regression detection. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 48px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="2" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="2" data-aria-level="1">Develop Python/Bash tooling that manages instance spawning, CPU core partitioning, memory pinning, performance data capture, reporting, and visual dashboards. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 48px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="2" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="3" data-aria-level="1">Maintain reproducible experiment workflows for both singleinstance and multiinstance configurations. </li></ul> Required Skills & Qualifications <ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 24px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="11" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="1" data-aria-level="1">Strong understanding of CPU architecture: pipelines, caches, TLB, NUMA, SMT/HT, vector units (AVX2/AVX512/VNNI/BF16/INT8), and memory hierarchy. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 24px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="11" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="2" data-aria-level="1">8 to 12 years in performance engineering, systems optimization, or lowlevel execution on Linux. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 24px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="11" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="3" data-aria-level="1">Handson experience with Linux tuning and serverclass OS/BIOS configuration. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 24px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="11" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="4" data-aria-level="1">Proficiency with perf, VTune, PMU counters, ftrace/tracecmd, flame graphs, and multiinstance profiling. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 24px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="11" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="5" data-aria-level="1">Strong knowledge of ML inference execution (tensors, operators, threading models) on CPU backends. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 24px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="11" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="6" data-aria-level="1">Strong Python and Bash for automation and performance tooling. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 24px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="11" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="7" data-aria-level="1">Experience in multicore scaling, thread affinity, scheduler behavior, concurrency techniques, and multiinstance execution strategies. </li></ul><ul style="margin: 0px; padding: 0px; font-family: verdana;"><li style="margin: 0px 0px 0px 24px; padding: 0px; font-size: 12pt; font-family: arial, helvetica, sans-serif;" aria-setsize="-1" data-leveltext="" data-font="Symbol" data-listid="11" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"multilevel"}" data-aria-posinset="8" data-aria-level="1">Familiarity with PyTorch, TensorFlow, ONNX Runtime for running inference workloads. </li></ul> </li></ul>#LI-SG2

Collected experience requirement

12 years