GPU Optimization | Dataford Interview Questions

Tell me about a time you optimized GPU performance for an application, and what changes you made.

For this coding exercise, implement a function that converts profiling observations into prioritized optimization recommendations. Each observation contains memory_bound, coalesced, occupancy, launch_overhead, and branch_divergence fields. Return applicable recommendations ordered by descending evidence, using the tie order defined below.

Function signature: recommend_gpu_optimizations(profile) Input is a list of dictionaries, and output is a list of recommendation strings.

Constraints

0 <= len(profile) <= 10^4
Each observation contains all five required fields
memory_bound, coalesced, launch_overhead, and branch_divergence are booleans
occupancy is a number in the range [0, 1]
Return recommendations in descending evidence order, using the specified priority for ties

Function Signature

def recommend_gpu_optimizations(profile):

Your question is GPU Optimization. Start with the requirements in the Question tab.

Run and submit as often as you like. When you're ready, talk me through your approach or go straight to the code.

Problem

Tell me about a time you optimized GPU performance for an application, and what changes you made.

Function signature: recommend_gpu_optimizations(profile) Input is a list of dictionaries, and output is a list of recommendation strings.

Constraints

0 <= len(profile) <= 10^4
Each observation contains all five required fields
memory_bound, coalesced, launch_overhead, and branch_divergence are booleans
occupancy is a number in the range [0, 1]
Return recommendations in descending evidence order, using the specified priority for ties

Function Signature

def recommend_gpu_optimizations(profile):

Your question is GPU Optimization. Start with the requirements in the Question tab.

Run and submit as often as you like. When you're ready, talk me through your approach or go straight to the code.