FPGA vs GPU: who's the future?

Compile & have spent | have fish

Product & have spent | & have spent Minersight (ID: minersight)

Eg. All the while, The main application field is electronic engineering. But when Intel gets it right & NBSP; Altera (Altera  Is one of the largest field programmable gate array manufacturers) was acquired when things changed slightly.

Intel pair & NBSP; FPGA  The potential is strong sense of smell and insight, and immediately after the acquisition, it will help Microsoft build data centers and use cloud services to help amazon.

But many miners may still have questions about what an FPGA is. What are the advantages of FPGA? What are the disadvantages of FPGA? Will FPGA stand out in the GPU mining camp and become the leading role? This article will give an overview of FPGA on the above questions of miners.

1. What is FPGA?

FPGA is an integrated circuit that is easier to manufacture than ASIC. Unlike ASIC, FPGA can be redesigned and configured. For FPGA miners, this allows them to mine multiple currencies supported by multiple algorithms.

2. Strengths and weaknesses

01. Delay

One of the biggest advantages of fpgas is their minimal latency. The reason field programmable gate arrays have such minimal latency is that they do not need to rely on a common OS or communicate via a common bus, such as USB or pci-express.

This is difficult to achieve on other hardware devices. Even high-quality cpus have a latency of 50 microseconds, but on fpgas, this time is reduced to 1 microsecond.

Field programmable gate arrays with extremely low latency can shine in situations where minimal latency is required. Fighter jets, for example, can be seen in autonomous operations and high-frequency algorithmic trading.

02. Efficiency

Dedicated integrated circuits (asics) are faster than fpgas in a single algorithm, but their initial investment in larger machine clusters is extremely high. Fpgas are better than gpus at mining digital currency because they run well and are highly efficient.

In addition to the benefits of efficient computing, fpgas have the advantage of being efficient without the need to run a host machine.

03. Engineering fees

Although FPGA has great advantages in the above aspects, the cost of configuration and programming is much higher than that of GPU based on instruction architecture. Hardware circuits are usually written using hardware description languages (HDL) such as VHDL and Verilog.

When it comes to programming, it's done in one of several programming languages (such as Java, C, and Python).

04. High level synthesis (HLS)

High-level synthesis (HLS) is an expanding trend, manifested in the ability to program fpgas using regular programming languages (such as Open CL or C ++) and high-level abstractions.

But despite all the technical advances and breakthroughs in programming, FPGA programming is still a tough job.

In addition, compilation time is a major disadvantage of field programmable gate arrays. For example, if you use the Intel Open CL compiler, it takes between 4 and 12 hours to compile a typical FPGA program. To achieve performance optimization, programmers often need to compile overnight.

05. Connectivity

Looking back at the GPU, it relies more on OS and standardized buses (such as USB and pci-express) for data transfer. Fpgas do not rely on these parameters, and their advantage is that they allow users to get higher bandwidth with less latency.

In fact, fpgas can directly connect to data sources, such as network interfaces and sensors.

3. FPGA and GPU floating-point number debate

In order to use high-performance computing functions such as deep learning, users have to rely heavily on gpus with excellent floating-point performance, while fpgas are less efficient at floating-point algorithms.

In general floating-point operations, floating-point units require a lot of resources and must be assembled from logical blocks. However, the script changes when it comes to FPGA. Since the FPGA structure already contains floating point units, the energy efficiency can be significantly improved.

Although, as things stand, advanced gpus are more energy efficient than fpgas, in the near future, field programmable gate arrays will certainly surpass gpus in floating-point computing and win the energy efficiency battle.

4, endnotes

Obviously, fpgas have a future. Gpus are likely to be replaced as fpgas become easier to program, faster to compile, and more efficient for floating point computing.

It's a big deal. It's a big deal. It's a big deal

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia