NVIDIA · Primly Community

NVIDIA technical program manager (TPM) interview: harder than you think

remote_swe_42 · 5 replies

just got through the NVIDIA TPM interview for a role on their GPU software programs team. this was my third TPM loop in the last year so I have some baseline to compare against. NVIDIA's is distinctive.

the loop: 5 rounds, all virtual, two separate half-days.

technical systems round: this is the one that separates NVIDIA from software company TPM loops. they expect you to be able to reason about hardware timelines, silicon dependencies, and the constraint that you can't hot-patch silicon the way you hot-patch software. I got asked: how do you manage a software program that's dependent on a hardware milestone that slips? they wanted a real answer about dependency mapping, critical path analysis, and what you actually do when you can't change the upstream. this is not a PM role with 'technical' in the title as a courtesy.

program design: design the release program for a major new SDK version that spans 12 internal teams and 3 external partners with different timelines. this was open-ended, 45 min. they wanted to see: stakeholder mapping, risk identification, communication cadence, how you define 'done' across distributed teams. I used a modified RACI structure and they pushed on whether I'd actually used it in a real program (I had).

behavioral: lots of STAR. 'tell me about a program that was failing and how you turned it around.' 'tell me about pushing back on a scope change from a VP.' 'tell me about managing a cross-functional program where you had no authority over team members.' these are the TPM behavioral classics and you need real stories, not hypotheticals.

domain technical: they tested whether I could hold a conversation about GPU programming models and their constraints. I don't write CUDA. but I needed to explain what a CUDA stream is and why it matters for scheduling decisions at the program level. if you're interviewing for GPU software programs you need to do your homework on the domain.

cross-functional / panel: the last round was 3 people: a TPM peer, a PM, and a senior engineer. more of a conversation. 'why NVIDIA over other big tech TPM roles' was explicitly asked.

comp for senior TPM: base around $195-210k Santa Clara, RSU upside is where the total lands. timeline was about 5 weeks from recruiter call to verbal offer.

5 replies

growth_gabe

the hardware dependency thing is such a good call-out. software PMs and TPMs are trained to think 'we can always ship a patch.' hardware timelines break that mental model. you have to think in quarters and physical constraints. I imagine that changes how you manage risk a lot.

tired_recruiter

the CUDA stream question for a TPM role surprises me. how deep did they actually expect the answer to go?

pm_priya

not 'write me code' deep. more like 'explain the concept well enough that I know you've talked to engineers about it.' they wanted to see that I wasn't going to nod blankly when the engineers I work with talk about execution contexts and async kernel launches. conversational fluency, not implementation knowledge.

nonprofit_nia

5-week timeline for a TPM role is actually reasonable. I've heard horror stories of 10-12 week processes at big tech for similar roles. did you have to do any take-home or was it all live?

pm_priya

all live. the program design round felt almost like a take-home in terms of depth but it was all done in the session. no work to do between rounds.