I applied through college or university. The process took 2 weeks. I interviewed at NVIDIA (Boston, MA) in May 2017
Interview
2 phone screens, followed by 5 onsite interviews. Overall it was a pleasant experience. All the interviewers were polite. Questions were mostly testing the architecture , SystemVerilog knowledge. Less of problem solving type questions
I applied online. The process took 2 weeks. I interviewed at NVIDIA (Shanghai, Shanghai) in Sep 2018
Interview
There are 2 round technique phone interviews, lasting for 1 hour for each. The first is conducted by the team member, the second is conducted by the team manager
Interview questions [1]
Question 1
phone interviews are all about then experience in resume
I applied through other source. The process took 2 weeks. I interviewed at NVIDIA in Apr 2018
Interview
The phone screen was average but had too many puzzles.
The 5 f2f interviews were on projects, SV and UVM but the interviewers always take you away from your area of expertise.
That's okay to some extent but annoying if concrete questions are asked for which one needs to consider many factors, apart from what the interviewer had in mind, to select a correct implementation. The interviewers want a clever answer not the correct one. Also some of their questions are vague and they will expect you to find all the unasked,implicit sub questions and solutions. Then will interrupt your explanation to go off on another tangential detail before you finish.
In programming and methodology interview where they wanted to know details found in nooks and crannies of reference manuals. dismissive of implementations other than their own even when pros and cons are outlined. On top of that sophomoric puzzles /cs algorithm questions are tossed out in midst of serious discussions.
Seems they want people who spend their time looking up puzzles and appearing smart.To deal with the interview one would need everything from asic design(rtl and gate level) to verification to data structures/algorithms. Mostly asic interviews are enjoyable irrespective of rigor or result but this one was too tiresome. Looks like this style of interviewing has become a fad.
Interview questions [1]
Question 1
Test bench architecture, systemverilog minutae, polymorphism in uvm, fifo design(including the generation of control), details on clock domain crossing, clock gating, setup and hold times, flops and resets, serial protocol and clock recovery,cache coherency, power saving methods, assertions, soc verification techniques, data structures,sort and search algorithms, quantum mechanics(no.. just kidding about the last one)
cover each "gotcha" point on all of the topics(all of which I answered correctly), but few with real depth or understanding. I got the feeling that they knew stuff as solutions to puzzles . When asked about their "solutions" they gave weak plausibility examples.
Also, hashing algorithms and applications, as usual an eternal fad.