1. FPGA based accelerators have
- Low power acceleration
- High reconfigurability
- Low latency
- All of the above
Answer :- All of the above
2. Energy consumption for a 32-bit floating multiplication is
- 3.1 pJ
- 3.1 nJ
- 3.7 pJ
- 3.3 PJ
Answer :- 3.7 pJ
3. Which of the following is not an FPGA accelerator?
- Eyeriss
- Cerebras WSE
- EIE
- MAPLE
Answer :- Cerebras WSE
4. How many Streaming Processors (SPs) are present in Tesla V100 GPU’ SM (Streaming Multiprocessor)?
- 8
- 32
- 64
- 128
Answer :- 64
5. SPMD usually refers to
- Message Passing Interface
- Message Passing Protocol
- Message Passing Programming
- None
Answer :- Message Passing Programming
6. To access the services of the operating system, the interface is provided by the _____
- Library
- Assembly instructions
- API
- System calls
Answer :- System calls
7. Which one of the following is not a characteristic feature of Process VM?
- Performs binary translation
- Virtualizes the ABI
- Runs in non-privileged mode
- Runs in privileged mode
Answer :- Runs in privileged mode
8. AWS EC2 is
- PaaS
- IaaS
- SaaS
- Storage as a service
Answer :- IaaS
9. Cloud services have a _____________ relationship with their customers.
- Many-to-one
- One-to-many
- Many-to-many
- One-to-one
Answer :- One-to-many
10. One example of PaaS is:
- Google Docs
- Google Compute Engine
- Gmail
- Google App Engine
Answer :- Google App Engine
11. Generic ML workflow follows:
1. Test model
2. Deploy model
3. Source and Prepare Data
4. Design Model
5. Training and validation
2->3->4->1->5
3->4->5->1->2
1->2->3->4->5
5->4->3->2->1
Answer :- 3->4->5->1->2
12. What does command $ docker ps do?
- Display all container details
- Display all running container details only
- Display all images
- None of the above
Answer :- Display all running container details only
13. $docker run -it –gpus all -p ….
in the above command -p represents
Port
Path
Process
None of the above
Answer :- Port
14. The most prominent scheduler used in High Performance Computing and AI is
PBS
Kubernetes
SLURM
Docker
Answer :- SLURM
15. Which command is used to check the running batch and interactive jobs by current user?
- sinfo
- squeue -u$USER
- sbatch
- scancel
Answer :- squeue -u$USER
16. Kubernetes is an orchestration tool because
- It supports scheduling to improve cluster utilization
- Supports features like load balancing and auto scaling
- Supports On-Prem cluster only
- Both A&B
Answer :- Both A&B
17. ………. can be considered as the primary data store of Kubernetes.
pod
etcd
node
all of the above
Answer :- etcd
18. The different services within Kubernetes are _______.
- Cluster IP service
- Node Port service
- Load Balancer service
- All of the above
Answer :- All of the above
19. Command to create .yaml file in Kubernetes
- $kubectl create –filename {filename}.yaml
- $kubectl create –{filename}.yaml
- $kubectl create -filena
- none of the above
Answer :- none of the above
20. DeepOps is a …………………..
- tool to build a GPU cluster
- tool to build a CPU cluster
- both
- none of the above
Answer :- tool to build a GPU cluster
21. MLOps is a combination of …………
a. Machine Learning
b. DevOps
c. Data Engineering
- a, b, c
- b, c
- a, c
- none of the above
Answer :- a, b, c
22. How Kubernetes handle a sudden breakdown of node?
- using node port service
- using scheduler
- it fails to handle
- using load balancing technique
Answer :- using load balancing technique
23. Which statement best describes RDMA?
- Random Access across the cluster
- Capability of HPC network transport layer to directly access (read/write) to remote system memory
- Technology to accelerate TCP/IP stack by offloading it into the network adapter
- None of the above
Answer :- Capability of HPC network transport layer to directly access (read/write) to remote system memory
24. In a GPU accelerated cluster, what are the most important network characteristics affecting overall system performance?
- A balanced GPU: Network-bandwidth that can ensure data is being fed to GPUs at an optimum rate to keep all processing units busy
- RDMA access directly from the GPU high-bandwidth memory for efficient data transfer
- Use multi-rail connection between GPU accelerated servers with multiple GPUs on-board
- All of the above
Answer :- All of the above
25. Which is NOT the correct mechanism to increase network bandwidth on the physical layer?
- Increase the clock frequency
- Increase the transmission amplitude
- Increase the number of physical lanes/channels
- Use advance modulation scheme like PAM4
Answer :- Increase the transmission amplitude
26. Which technology improves the performance of MPI operations?
- SHIELD
- SHARP
- Multi-Host Technology
- Socket-Direct Technology
Answer :- SHARP
27. Total number of nodes in a n-dimensional hypercube network is
- N
- 2n-1
- N-1
- 2n
Answer :- 2n
28. Which among the following statement(s) is/are true about InfiniBand?
a. InfiniBand is the only architecture that can be used on the PCB
b. It has unicast and multicast support
c. High priority transactions between devices can be processed ahead of the lower priority items
- a, b
- b, c
- a, b, c
- a, c
Answer :- a, b, c
29. What are the three levels of Dragonfly network topology?
- Client, Server, Switch
- PC, Repeater, Transceiver
- Router, Group, System
- Workstation, Hub, Server
Answer :- Router, Group, System
30. Which among the following statement(s) is/are True?
- Zero-copy networking does not utilize computer processor.
- Data transfer between the systems without RDMA support is faster than both the systems that enables RDMA.
- IWARP is not compatible with RDMA.
- Both A and B
Answer :- Zero-copy networking does not utilize computer processor.