✅ Custom Hardware for AI – Solved MCQs 1. What is the primary purpose of custom AI hardware ? A. Playing games B. General-purpose proce...
✅ Custom Hardware for AI – Solved MCQs
1. What is the primary purpose of custom AI hardware?
A. Playing games
B. General-purpose processing
C. Accelerating machine learning computations
D. Memory optimization
✅ Correct Answer: C. Accelerating machine learning computations
2. TPUs (Tensor Processing Units) are developed by:
A. Apple
B. Microsoft
C. Google
D. IBM
✅ Correct Answer: C. Google
3. What is the core advantage of using a GPU over a CPU for AI?
A. Faster clocks
B. Better for sequential tasks
C. Parallel processing capability
D. Lower cost
✅ Correct Answer: C. Parallel processing capability
4. Which of the following is not a hardware accelerator?
A. CPU
B. GPU
C. TPU
D. FPGA
✅ Correct Answer: A. CPU
5. Which of the following is an example of FPGA use in AI?
A. Data center cooling
B. Image filtering
C. Reconfigurable inference acceleration
D. Database storage
✅ Correct Answer: C. Reconfigurable inference acceleration
6. NPUs (Neural Processing Units) are optimized for:
A. Gaming
B. Database querying
C. Deep learning inference
D. Compiling code
✅ Correct Answer: C. Deep learning inference
7. A key benefit of using ASICs in AI is:
A. Reprogrammability
B. General use
C. High performance and low power consumption
D. Upgradability
✅ Correct Answer: C. High performance and low power consumption
8. What does "edge AI" typically require?
A. Cloud servers
B. Power-hungry processors
C. Lightweight inference on local devices
D. Human supervision
✅ Correct Answer: C. Lightweight inference on local devices
9. Which is a custom AI chip for Apple devices?
A. Bionic A-series
B. Snapdragon
C. Tensor SoC
D. Xeon
✅ Correct Answer: A. Bionic A-series
10. What is the primary advantage of using TPUs in neural networks?
A. Fast rendering
B. Low latency inference for deep learning
C. Operating system emulation
D. Browser speed
✅ Correct Answer: B. Low latency inference for deep learning
✅ Inference Optimization – Solved MCQs
11. In AI, “inference” refers to:
A. Training a model
B. Guessing parameters
C. Running a trained model on new data
D. Creating training labels
✅ Correct Answer: C. Running a trained model on new data
12. Quantization in inference optimization reduces:
A. Training time
B. Model accuracy
C. Precision to improve speed and memory
D. Model depth
✅ Correct Answer: C. Precision to improve speed and memory
13. What does pruning do in model optimization?
A. Adds more layers
B. Reduces model size by removing unimportant weights
C. Increases memory usage
D. Reduces bias
✅ Correct Answer: B. Reduces model size by removing unimportant weights
14. Which data type is commonly used in quantized inference?
A. float64
B. float32
C. int8
D. double
✅ Correct Answer: C. int8
15. Which of the following techniques is NOT used in inference optimization?
A. Quantization
B. Pruning
C. Dropout
D. Model distillation
✅ Correct Answer: C. Dropout
16. What is the purpose of TensorRT?
A. Running JavaScript
B. Optimizing TensorFlow training
C. High-performance inference on NVIDIA GPUs
D. Database acceleration
✅ Correct Answer: C. High-performance inference on NVIDIA GPUs
17. ONNX stands for:
A. Open Neural Network Exchange
B. Optimized Neural Network Extension
C. Operational Node eXecution
D. Offline Node Network
✅ Correct Answer: A. Open Neural Network Exchange
18. Model distillation is a technique to:
A. Add noise to models
B. Train a smaller student model from a larger teacher model
C. Increase training loss
D. Visualize datasets
✅ Correct Answer: B. Train a smaller student model from a larger teacher model
19. In deep learning deployment, the biggest bottleneck is often:
A. CPU heat
B. I/O throughput
C. Inference latency
D. Label generation
✅ Correct Answer: C. Inference latency
20. Which platform is not used for inference optimization?
A. TensorRT
B. OpenVINO
C. ONNX Runtime
D. MySQL
✅ Correct Answer: D. MySQL
✅ Mixed Questions – Custom Hardware + Optimization
21. Which hardware is best for training very large models?
A. CPU
B. FPGA
C. High-end GPU
D. NPU
✅ Correct Answer: C. High-end GPU
22. Which framework supports edge AI inference on microcontrollers?
A. TensorFlow
B. TensorFlow Lite
C. PyTorch
D. Scikit-learn
✅ Correct Answer: B. TensorFlow Lite
23. Why is model compression important in edge AI?
A. For faster internet
B. To match screen size
C. To reduce memory and power usage
D. To improve sound
✅ Correct Answer: C. To reduce memory and power usage
24. What does OpenVINO focus on?
A. Audio enhancement
B. Vision inference optimization on Intel hardware
C. Web development
D. Chatbot training
✅ Correct Answer: B. Vision inference optimization on Intel hardware
25. Which AI deployment tool is optimized for Apple devices?
A. PyTorch
B. Core ML
C. TensorFlow Hub
D. JAX
✅ Correct Answer: B. Core ML
26. Which optimization reduces both compute and storage requirements?
A. Hyperparameter tuning
B. Dropout
C. Quantization
D. Oversampling
✅ Correct Answer: C. Quantization
27. A common problem with aggressive quantization is:
A. Higher power usage
B. Increased training time
C. Accuracy degradation
D. More layers
✅ Correct Answer: C. Accuracy degradation
28. Which of the following is a hardware-aware optimization technique?
A. Pruning
B. Quantization
C. Neural Architecture Search (NAS)
D. Normalization
✅ Correct Answer: C. Neural Architecture Search (NAS)
29. Jetson Nano is a product of:
A. AMD
B. Intel
C. NVIDIA
D. Microsoft
✅ Correct Answer: C. NVIDIA
30. What does latency refer to in inference?
A. Model depth
B. Memory allocation
C. Delay between input and output
D. Temperature increase
✅ Correct Answer: C. Delay between input and output
✅ Advanced MCQs (31–50)
31. Bit-width reduction in quantization is done to:
A. Add more neurons
B. Decrease computation cost
C. Increase precision
D. Improve training loss
✅ Correct Answer: B. Decrease computation cost
32. NVIDIA’s DLSS uses AI for:
A. Speech recognition
B. Game performance boosting via deep learning
C. Hardware repair
D. Virus scanning
✅ Correct Answer: B. Game performance boosting via deep learning
33. ASICs are best suited for:
A. General computation
B. Flexible model development
C. Specific and repetitive AI workloads
D. Debugging
✅ Correct Answer: C. Specific and repetitive AI workloads
34. A major challenge in edge inference is:
A. Too much electricity
B. Cooling fans
C. Limited compute and memory
D. Lack of internet
✅ Correct Answer: C. Limited compute and memory
35. Neural Engine in iPhones is a type of:
A. NPU
B. CPU
C. TPU
D. ASIC
✅ Correct Answer: A. NPU
36. TFLite is mainly used for:
A. Real-time OS control
B. Mobile and edge inference
C. Cloud training
D. Robotic movement
✅ Correct Answer: B. Mobile and edge inference
37. Which one offers reconfigurability at runtime?
A. ASIC
B. FPGA
C. TPU
D. GPU
✅ Correct Answer: B. FPGA
38. Quantization-aware training helps by:
A. Avoiding overfitting
B. Reducing training time
C. Improving accuracy post-quantization
D. Slowing down inference
✅ Correct Answer: C. Improving accuracy post-quantization
39. Which is an example of layer fusion?
A. Merging Conv + BatchNorm
B. Repeating max pool
C. Adding dropout
D. Increasing depth
✅ Correct Answer: A. Merging Conv + BatchNorm
40. A good reason to use ONNX is:
A. Reduces image noise
B. Helps convert models between frameworks
C. Offers GPU drivers
D. Provides voice assistant
✅ Correct Answer: B. Helps convert models between frameworks
41. Which hardware type has fixed logic for AI models?
A. FPGA
B. CPU
C. ASIC
D. NPU
✅ Correct Answer: C. ASIC
42. Edge TPUs are designed for:
A. Cloud training
B. Mobile games
C. Inference at the edge with low power
D. Database lookup
✅ Correct Answer: C. Inference at the edge with low power
43. LLMs require which optimization for deployment?
A. Upsampling
B. Weight sharing
C. Inference acceleration
D. Token removal
✅ Correct Answer: C. Inference acceleration
44. Which is a risk of over-optimizing inference?
A. Bigger model
B. Better performance
C. Accuracy drop
D. GPU freezing
✅ Correct Answer: C. Accuracy drop
45. Which toolkit is used with NVIDIA for deployment?
A. DeepStream
B. Xcode
C. PyCaret
D. Hugging Face
✅ Correct Answer: A. DeepStream
46. Power-efficient AI hardware is crucial for:
A. Laptops only
B. Edge and embedded devices
C. Cloud clusters
D. Model tuning
✅ Correct Answer: B. Edge and embedded devices
47. Real-time AI applications need:
A. High throughput
B. High latency
C. Post-training only
D. Lower resolution
✅ Correct Answer: A. High throughput
48. ARM-based AI chips are used for:
A. Supercomputers only
B. Mobile devices and edge
C. Satellites
D. Legacy systems
✅ Correct Answer: B. Mobile devices and edge
49. One downside of using full-precision models in deployment is:
A. High accuracy
B. Slow inference and high resource consumption
C. Compatibility
D. Faster quantization
✅ Correct Answer: B. Slow inference and high resource consumption
50. Efficient AI inference allows:
A. More frequent retraining
B. Larger datasets
C. Real-time decision-making in low-resource environments
D. Storing user passwords
✅ Correct Answer: C. Real-time decision-making in low-resource environments
COMMENTS