YourAlly

NVIDIA Staff coding interview questions

These are the 3 coding questions YourAlly has tagged for NVIDIA at Staff. On the YourAlly ladder this is Staff (E6), the same level as Google L6, Meta E6 and Amazon L7.

A coding round at a large tech company usually runs 45 to 60 minutes with one or two problems. You write code that runs, test it on your own examples and state its time and space cost. At this level interviewers also expect clean code quickly and edge cases you raise before they do.

Practice a mock NVIDIA Staff coding loopPractice NVIDIA coding questions

Dynamic Batcher Completion TimesFree to read

Simulate an inference server's dynamic batcher. Requests arrive in time order and queue FIFO. A batch is sent to the earliest free model instance once the queue holds max_batch requests or the oldest request has waited max_delay. A batch of k requests takes base_cost + per_item_cost * k. Return each request's completion time.

Input: arrivals = [0,1,2,3], max_batch = 2, max_delay = 10, base_cost = 5, per_item_cost = 1, instances = 1 Output: [8,8,15,15]

What a strong answer covers

  • Trigger as min of the delay deadline and the fill time
  • Dispatch at the later of trigger and earliest free instance
  • Batch limited to arrived requests and max_batch
  • Min-heap of instance free times
  • Argument that FIFO order and monotone dispatch times hold

2 more coding questions

  1. Buddy Allocator for Device Memory Locked
  2. Pipeline Stages Under a Memory Cap Locked

Sign up free to practice all 3

Related pages

More NVIDIA Staff rounds

NVIDIA coding at other levels

Other companies at this level