CUDA availability check in PyTorch - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Check CUDA availability output

What is the output of this PyTorch code snippet that checks if CUDA is available?

PyTorch

import torch
print(torch.cuda.is_available())

A True if a CUDA-enabled GPU is available, otherwise False

B The number of CUDA devices available as an integer

C Raises a RuntimeError if CUDA is not installed

D Always prints False regardless of hardware

💡 Hint

This function returns a boolean indicating hardware availability.

torch.cuda.is_available() returns True if PyTorch detects a CUDA-capable GPU and the CUDA drivers are properly installed; otherwise, it returns False.

🧠 Conceptual

intermediate

Understanding CUDA device count

Which PyTorch function correctly returns the number of CUDA devices available on the system?

A torch.cuda.current_device()

B torch.cuda.is_available()

C torch.cuda.get_device_name()

D torch.cuda.device_count()

💡 Hint

This function returns an integer count of GPUs.

torch.cuda.device_count() returns the number of CUDA-capable devices detected by PyTorch.

🔧 Debug

advanced

Diagnose CUDA availability error

What error will this code raise if CUDA is not available on the system?

PyTorch

import torch
print(torch.cuda.current_device())

A IndexError: list index out of range

B RuntimeError: CUDA error: invalid device ordinal

C AssertionError: CUDA not available

D No error, prints 0

💡 Hint

current_device() requires at least one CUDA device.

If no CUDA device is available, torch.cuda.current_device() raises a RuntimeError indicating an invalid device ordinal.

❓ Model Choice

advanced

Choosing device for model training

Given this code snippet, which device will the model be moved to if CUDA is available?

PyTorch

import torch
model = torch.nn.Linear(10, 2)
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model.to(device)
print(next(model.parameters()).device)

A cuda:0

B cpu

C cuda:1

D Raises an error

💡 Hint

The device string 'cuda' defaults to the first GPU.

If CUDA is available, 'cuda' refers to the first GPU device (cuda:0), so the model parameters move there.

❓ Metrics

expert

Interpreting CUDA memory usage

What does the following PyTorch code output represent?

PyTorch

import torch
print(torch.cuda.memory_allocated())

A The peak memory usage during the program run

B The total GPU memory available on the system in bytes

C The number of bytes currently allocated by tensors on the default CUDA device

D The number of CUDA devices available

💡 Hint

This function reports memory used by tensors, not total or peak memory.

torch.cuda.memory_allocated() returns the current number of bytes allocated by tensors on the default CUDA device.