AMD · Phone screen → Onsite
In processAMD Experienced Full-Time Interviews by Role through Oct 2026: 16 Experiences (Part 1 :)
Note: This is the most complete collection of interview information I could find in this round of searching. If friends need more information in the future, I may start another round of research. Good luck with your interviews! — 11-02-2026 (jane)
In this series
- Part 1 (current): Senior Software System Designer; GPU Software Engineer; Senior Software Engineer - uProf; Compiler / Performance; Mid-level Software Engineer.
- Part 2: SDE2; Senior Software Engineer - AI; MTS Software Development Engineer - GPU; Role not reported — post-offer start delay; Software System Designer 2 — Test Automation; Verification — PCIe; Role not reported — C and memory panel.
- Part 3: GPU team — exact role not reported; MTS Verification; Member of Technical Staff - Performance.
Coverage: 6 of 16 AMD Experienced Full-Time accounts in 5 separate role sections.
Other AMD hiring levels
- Hiring Level Not Reported: 9 accounts.
- Intern / Co-op: 18 accounts in 2 parts.
- New Grad: 11 accounts in 2 parts.
Roles in this article
- Senior Software System Designer — 2 accounts.
- GPU Software Engineer — 1 account.
- Senior Software Engineer - uProf — 1 account.
- Compiler / Performance — 1 account.
- Mid-level Software Engineer — 1 account.
Senior Software System Designer
2 accounts in this role.
What came up
| Topic | Reports | Most recent report |
|---|---|---|
| Coding tasks | 2 of 2 | 2026-10-07 (report) |
| Projects or prior work | 2 of 2 | 2026-10-07 (report) |
| OS, memory, architecture or concurrency | 2 of 2 | 2026-10-07 (report) |
| C/C++ explicitly reported | 2 of 2 | 2026-10-07 (report) |
| GPU or ML questions | 1 of 2 | 2026-02-13 (report; offer month 2026-02) |
Questions we found
Algorithms / Data Structures
| Question | Frequency | All last seen dates |
|---|---|---|
| Lowest Common Ancestor of two binary-tree nodes | 1 | Reports: 2026-10-07 (report) |
| Sort an array of zeroes and ones | 1 | Reports: 2026-10-07 (report) |
C/C++ / OS Questions
| Question | Frequency | All last seen dates |
|---|---|---|
| Mutex versus semaphore | 1 | 2026-10-07 (report) |
| fork and child processes | 1 | 2026-10-07 (report) |
| User versus kernel mode | 1 | 2026-10-07 (report) |
| free/delete and malloc/new | 1 | 2026-10-07 (report) |
01: Bengaluru: LCA, a binary array, then systems work
Report date: October 7, 2026 · Senior Software System Designer · About 5 years of experience · Selected and joined
Round 1 — online coding and resume
The coding question was Lowest Common Ancestor of two nodes in a binary tree. The interviewer also went through the candidate's projects and resume, looking at the approach, code quality and complexity.
Round 2 — OS, concurrency and coding, in person, about 90 minutes
They covered mutexes versus semaphores, fork() and child processes, user versus kernel mode, process states, threads, synchronization and CPU caches. LRU cache also appeared as a discussion topic; this account does not say the candidate implemented it.
The coding task was to sort an array containing only zeroes and ones.
Round 3 — systems and current work, in person
The interviewer dug into the candidate's current project and responsibilities. The account names a producer-consumer problem, but does not say whether the expected deliverable was code or design. C/C++ questions included free versus delete, malloc versus new, and high-performance system programming.
Hiring manager and final managerial rounds — online
The hiring manager discussed current responsibilities, teamwork and collaboration. The final conversation covered daily work, releases, cross-team dependencies and joining availability, with no major coding question reported.
02: Hyderabad: three rounds across C++, DNNs and architecture
Offer: February 2026 · Report date: February 13, 2026 · Senior Software System Designer · 2.5 years of experience · Selected
This shorter account describes three rounds covering DSA in C++, current work, deep neural networks and architecture. It does not identify the coding problems or individual architecture questions.
GPU Software Engineer
1 account in this role.
What came up
This GPU software loop moved from inference-engine internals to agent design, profiling and a Top-K task.
Questions we found
Algorithms / Data Structures
| Question | Frequency | All last seen dates |
|---|---|---|
| Find the Top-K elements; heap answer followed by a request for partitioning | 1 report | Reports: 2026-09-17 (report); Jul–Sep 2026 (header interview period); Additional same-company practice history: 2026-09-18 |
LLM Inference Questions
| Question | Frequency | All last seen dates |
|---|---|---|
| Trace a prompt through vLLM, including PagedAttention, KV-cache block tables and continuous batching | 1 report | 2026-09-17 (report); Jul–Sep 2026 (header interview period) |
| Use a roofline model to discuss prefill, decoding and memory versus compute limits | 1 report | 2026-09-17 (report); Jul–Sep 2026 (header interview period) |
Agentic AI Questions
| Question | Frequency | All last seen dates |
|---|---|---|
| Explain skills, progressive disclosure and how skills differ from prompts | 1 report | 2026-09-17 (report); Jul–Sep 2026 (header interview period) |
| Explain the ReAct architecture | 1 report | 2026-09-17 (report); Jul–Sep 2026 (header interview period) |
| Compare a single agent with multiple agents | 1 report | 2026-09-17 (report); Jul–Sep 2026 (header interview period) |
Compiler / GPU Performance Questions
| Question | Frequency | All last seen dates |
|---|---|---|
| Explain register spilling and why it happens | 1 report | 2026-09-17 (report); Jul–Sep 2026 (header interview period) |
| Profile a bottleneck and explain the parallelism strategy chosen | 1 report | 2026-09-17 (report); Jul–Sep 2026 (header interview period) |
| Analyze noncontiguous access in a ROCm matrix-transpose kernel | 1 report | 2026-09-17 (report); Jul–Sep 2026 (header interview period) |
| Explain a cache-line conflict with a diagram | 1 report | 2026-09-17 (report); Jul–Sep 2026 (header interview period) |
Behavioral (BQ)
| Question | Frequency | All last seen dates |
|---|---|---|
| Introduce yourself | 1 report | 2026-09-17 (report); Jul–Sep 2026 (header interview period) |
| Discuss your motivation for the role | 1 report | 2026-09-17 (report); Jul–Sep 2026 (header interview period) |
| How do you learn new technologies? | 1 report | 2026-09-17 (report); Jul–Sep 2026 (header interview period) |
| Share your impressions of the interviews so far | 1 report | 2026-09-17 (report); Jul–Sep 2026 (header interview period) |
03: Toronto / Markham: inference systems, a transpose kernel and Top-K
Interview period: Jul–Sep 2026 (header) · Report: Sep 17, 2026 · GPU Software Engineer · Toronto / Markham, Canada · Offer
A recruiter contacted the candidate on LinkedIn. The recruiter screen took about 15 minutes. A hiring-manager screen booked for 45 minutes ran for more than an hour; the candidate was told the following week that they had passed. The onsite followed roughly two weeks later. Because the candidate was local and had offered to meet in person, that part of the process was held in person.
Recruiter and hiring-manager screens
The recruiter covered motivation and previous work, explained the team, and took a question about export-control licensing. The manager returned to the candidate's experience and asked about inference engines.
Onsite 1 — ML frameworks and inference
After a self-introduction and a detailed resume discussion, the interviewer asked what happens from the moment a prompt enters vLLM. The conversation covered PagedAttention, the KV-cache block table and continuous batching. Roofline reasoning led to memory-bound versus compute-bound work and the different constraints of prefill and decoding.
Onsite 2 — agents and compilers
After another self-introduction, the candidate drew the inference-engine diagram again. Questions covered skills, progressive disclosure, skills versus prompts, ReAct, and the trade-offs between one agent and several agents. The discussion then moved to register spills and why they occur.
Onsite 3 — profiling, GPU behavior and coding
The interviewer asked about profiling tools, finding a bottleneck, parallelism strategies and why those strategies helped. They supplied a ROCm transpose kernel for analysis; the candidate identified that one side of the read/write pattern would be noncontiguous.
The named coding task was Top-K elements. The candidate used a heap. The interviewer wanted an approach based on quickselect or quicksort-style partitioning. A cache-line-conflict question ended with a whiteboard explanation.
Onsite 4 — hiring-manager follow-up
The manager asked about the interview, learning new technology, the team's work, kernel agents, compiler agents and newer inference directions. This slot was booked for 15 minutes and ran close to an hour. The first three onsite slots had been booked for 45 minutes and had also run toward an hour.
The offer arrived about three weeks after the onsite. During that wait, the candidate followed up with the recruiter about once a week.
Senior Software Engineer - uProf
1 account in this role.
What came up
Reported topics: coding tasks; projects or prior work; OS, memory, architecture or concurrency.
Questions we found
Algorithms / Data Structures
| Question | Frequency | All last seen dates |
|---|---|---|
| Implement a growing producer-consumer circular queue | 1 | Reports: 2026-02-10 (report); Additional same-company practice history: 2026-02-12 |
Behavioral (BQ)
| Question | Frequency | All last seen dates |
|---|---|---|
| Why leave your current company, and what do you know about this team? | 1 | 2026-02-10 (report) |
C/C++ / OS Questions
| Question | Frequency | All last seen dates |
|---|---|---|
| Mutexes and semaphores | 1 | 2026-02-10 (report) |
| Child processes and fork | 1 | 2026-02-10 (report) |
| Kernel/user modes and process states | 1 | 2026-02-10 (report) |
04: uProf: a queue that had to survive a burst of 500 items
Report date: February 10, 2026 · Senior SWE, uProf · Bengaluru · More than 3.5 years of experience · Rejected
Round 1 — screen
The candidate described easy/medium coding questions, CV discussion and current work, but did not name the problems.
Round 2 — OS and concurrency, about 90 minutes
Topics included mutexes and semaphores, child processes and fork, kernel/user modes, process states, caches and LRU awareness. There was also a modified question from a LeetCode concurrency list; its identity was not given.
Round 3 — in-person coding, about 90 minutes
After a detailed discussion of current work, the candidate had to write a runnable producer-consumer circular queue. It started with capacity 10. The source describes normal occupancy of 7–8 items in 90% of cases, but bursts up to 500 in the remaining 10%, without losing data.
The Growing Circular Queue practice exercise preserves that growth requirement. Its ordered operation interface and doubling policy are defined for practice; the report does not prescribe those rules.
Hiring manager
They discussed why the candidate wanted to leave and what they understood about the team's work. The author reported positive feedback before receiving a rejection.
Compiler / Performance
1 account in this role.
What came up
Matrix tiling appeared in two of the candidate’s five interviews. Resume work was examined throughout.
Questions we found
Algorithms / Performance Coding
| Question | Frequency | All last seen dates |
|---|---|---|
| Tile a matrix multiplication | 1 report | Reports: 2025-10-05 (report); Oct–Dec 2025 (header interview period); Additional same-company practice history: 2024-03-31, 2021-04-07, 2021-04-06 |
05: Compiler / performance: matrix tiling appeared in two rounds
Interview period: Oct–Dec 2025 (header) · Report: Oct 5, 2025 · Compiler / performance · Result not specified
The candidate described five interviews of about 45 minutes each, with detailed resume questions throughout. They said the coding was connected to NPUs and did not follow the usual LeetCode format.
Tiling a matrix multiplication came up twice in that one loop. The report does not preserve the two versions, tile sizes, cache parameters or the candidate's solution. The two rounds count as one candidate sighting of the question.
Mid-level Software Engineer
1 account in this role.
What came up
Reported topics: coding tasks; projects or prior work; OS, memory, architecture or concurrency; named design round.
Questions we found
Algorithms / Data Structures
| Question | Frequency | All last seen dates |
|---|---|---|
| Implement an allocator in the round described as LLD | 1 | Reports: 2026-09-06 (same account’s report publication); 2025-01 (interview); Additional same-company practice history: 2026-02-12 |
| Climbing Stairs | 1 | Reports: 2026-09-06 (same account’s report publication); 2025-01 (interview); Additional same-company practice history: 2024-03-31 |
| Coin Change in the U.S. mid-level loop (variant not specified) | 1 | Reports: 2026-09-06 (same account’s report publication); 2025-01 (interview); Additional same-company practice history: 2025-06-12 |
System Design
| Question | Frequency | All last seen dates |
|---|---|---|
| Design WhatsApp and justify the stack | 1 | Reports: 2026-09-06 (same account’s report publication); 2025-01 (interview) |
Behavioral (BQ)
| Question | Frequency | All last seen dates |
|---|---|---|
| Discuss strengths, weaknesses and possible teams | 1 | 2025-01 (interview) |
06: A U.S. mid-level loop with an allocator and WhatsApp design
Interview: January 2025 · Mid-level Software Engineer · United States · Offer declined
The recruiter discussed earlier technical projects and motivation.
The first technical round was described as LLD and asked for a memory allocator implementation. The interviewer offered hints; the API and constraints were not shared. The linked first-fit arena’s API and allocation rules are authored for practice.
The next round asked for a high-level design of WhatsApp and reasons for the chosen technology stack. The linked messaging exercise supplies its own workload assumptions. A further technical/behavioral round included Climbing Stairs and Coin Change, then strengths, weaknesses and possible teams. The Coin Change variant was not specified; the linked exercise explicitly uses minimum coins.
The candidate declined the offer over compensation. The account was posted in September 2026, but describes a January 2025 interview.
Comments
0No comments yet — be the first to share what you know.