Check your understanding of this module before moving on.
?
0/3
Checkpoint quiz 3 questions
A quick check on this module's key ideas before you move on
Question 01
Why does training deep networks typically rely on GPUs (or TPUs) rather than CPUs alone?
A is correct. Training a neural network is dominated by large matrix multiplications, which decompose into millions of independent multiply-add operations. GPUs are designed with massive numbers of simple cores that excel at exactly this kind of parallel, repetitive arithmetic.
Question 02
What is the purpose of a data loader/pipeline that prefetches and batches data during training?
A is correct. If data loading and preprocessing happen serially before each training step, the GPU sits idle waiting. A well-designed pipeline prepares the next batch in the background while the current one is being processed, keeping the GPU busy.
Question 03
Why might a training run use mixed-precision (e.g. float16/bfloat16 combined with float32) rather than full float32 precision throughout?
A is correct. Reduced-precision formats use less memory and compute faster on hardware built to accelerate them. Mixed-precision training keeps certain sensitive computations (like gradient accumulation) in higher precision to avoid numerical issues while reaping most of the speed benefit elsewhere.