Multicore and many-core processors have appeared in widespread use in both deskt

Question

Multicore and many-core processors have appeared in widespread use in both desktop computers and HPC systems. Answer the following questions regarding advanced processors, memory devices, and system interconnects: a. What are the differences between multicore CPUs and GPUs in terms of architecture and usage? b. Explain why parallel programming cannot match the progress of processor technology. c. Suggest ideas and defend your argument with some plausible solutions to this mismatch problem between core scaling and effective programming and use of multicores. d. Explain why flash memory SSD can deliver better speedups in some HPC or HTC applications. e. Justify the prediction that InfiniBand and Ethernet will continue to dominate the HPC market.

Explanation / Answer

a)

GPUs have far more processor cores than CPUs, but because each GPU core runs significantly slower than a CPU core and do not have the features needed for modern operating systems, they are not appropriate for performing most of the processing in everyday computing. They are most suited to compute-intensive operations such as video processing and physics simulations.In other words, CPUs and GPUs have significantly different architectures that make them better suited to different tasks. A GPU can handle large amounts of data in many streams, performing relatively simple operations on them, but is ill-suited to heavy or complex processing on a single or few streams of data. A CPU is much faster on a per-core basis (in terms of instructions per second) and can perform complex operations on a single or few streams of data more easily, but cannot efficiently handle many streams simultaneously.

GPUs lack:

You need these to be able to implement anything like a modern operating system.

1. CPU is the brain of the computer the GPU is only meant to complement it.
2. GPUs are specialized and cannot replace the function of a CPU.
3. CPUs can perform the functions of a GPU but at a much slower speed.
4. GPUs can rival CPUs in transistor count.
5. GPUs can also work in tandem just like the CPU’s multi core capability

d) Solid-state drives actually aren't hard drives in the
traditional sense of the term, as there are no moving parts involved.
A traditional hard disk drive (HDD) consists of a spinning disk with a
read/write head on a mechanical arm. An SSD, on the other hand,
has an array of semiconductor memory organized as a disk drive,
using integrated circuits (ICs) rather than magnetic or optical storage media.

using solid-state drive storage(SSD) as an alternative to DRAM and hard drives, which could help speed up internal data transfers because SSDs are lightning fast. this is close, or as fast as the RAM, we can take the RAM out and replace it with an SSD. Because it is non-volatile and has great storage, it brings better performance and saving more power (since sleeping will be considered as shutting down because there is no RAM to power up) and speed up booting times.

e)

Ethernet cannot kill InfiniBand. For the foreseeable future, the very high-end of the server, storage, and database cluster spaces will need a network interconnect that can deliver the same or better bandwidth at lower latency than can Ethernet gear. That latter bit is the important part, and it is what is driving InfiniBand forward to 100 Gb/sec speeds now and 200 Gb/sec speeds a few years hence.

The biggest advantage for Ethernet in the cluster environment is the seamless integration with the wider network, freeing storage from geographical constraints. In supercomputing, storage is used to hold "data sets," which are moved from storage to the cluster and back again after processing.

Infiniband pros:

• Trade association is defining the architecture

• Multi vendor support (but single chip manufacturer)

• Native storage connections

• Low latency and high bandwidth simultaneously

• QoS • Dual-port adapters provide 2x 10 Gigabit bandwidth

Infiniband cons: • Cumbersome and clumsy requiring significant reinvestment in software development • Scaling is challenging in both logical and physical design • Need a gateway to talk to WAN or Campus • Expensive, cumbersome cabling • Limited distance – 17 meters for copper cable, no fiber currently available • Single vendor for Infiniband chips (Mellanox)

Ethernet pros: • Management and debugging tools are well established • No need to hire specialists • One network can provide interconnect, storage, I/O, and management • No need for specialized gateways • It interconnects with anything • Low cost, readily available cable • Higher-level protocols provide a rich set of application support • Equipment often has high-availability features • iSCSI gaining momentum for storage over Ethernet • Multiple clusters can be combined across LAN or WAN • LAN on Motherboard (LOM) reduces cost at node • You already know how it works – training is minimal Ethernet cons: • Traditionally the cons for Ethernet were two: 10 Gigabit Ethernet was expensive and the latency of 10 Gigabit Ethernet switches was high. Both of these have been recently removed. • There remains some confusion about the improvements in 10 Gigabit NICs, with RDMA, TOE, etc., and their prices. The price and latency of 10 Gigabit Ethernet NICs has fallen dramatically and their latency performance has increased substantially. These continued improvements are removing the price and latency advantage of proprietary technologies.

Related Questions

Navigate

• Browse (All)
• Browse M
• Subjects

Integrity-first tutoring: explanations and feedback only — we do not complete graded work. Learn more.