In today’s world where big data analytics have become an indispensable part of the routine in numerous industries, the hunger for faster computing speed and storage can be hard to satisfy. The advent of PCIe 4.0, a new generation Peripheral Component Interconnect Express standard, is welcomed by many as it can improve the performance of complex applications such as machine learning, big data analytics, and high-resolution visualization with faster speed. The PCIe standard was first proposed in 2003 to interconnect CPU, storage, networking, IO and GPUs. It has evolved 4 times, with data transfer rates and bandwidth being doubled each time (see Figure 1 for details). Currently, PCIe 3.0 remains the mainstream standard on the market, but the long lapse between its release in 2010 and the announcement of PCIe 4.0 in 2017 has rendered users impatient and curious about the performance of PCIe 4.0.
Figure 1. Evolution of PCIe standard
https://pcisig.com/sites/default/files/files/PCI-SIG%20DevCon%202018_Briefing%20Presentation.pdf
QCT pioneered validating the superior performance of PCIe 4.0
Although the PCIe 4.0 standard was released in 2017, compatible hardware components were not available until 2019. To evaluate the actual performance of the latest technology, Quanta Cloud Technology (QCT) took the lead to conduct tests on its PCIe 4.0-supporting servers and data farms used for cloud storage, services, and software in concert with our ecosystem partners. Theoretically, servers and data farms using PCIe 4.0 network adapters should run twice as fast as those using PCIe 3.0. To validate this assumption, we built a test environment with two QUANTAGRID D43K-1U servers powered by AMD EPYC processors and installed them with the latest Microsoft Windows Server 2019 OS. Two tests were subsequently conducted.
The first test focused on data transfer. Two PCIe 3.0 network adaptors spent 93 and 88 seconds to transfer 1TB of data; however, their PCIe 4.0 counterparts only took 47 seconds to get the job done, saving 48% of the time (see Figure 2). The bandwidth also increased from 11.5 and 12.2 GB/s for PCIe 3.0 adaptors to 22.5 GB/s for PCIe 4.0 adaptors (see Figure 3) which is 1.9 times faster.
Figure 2. PCIe 3.0 and 4.0 data transfer time
Figure 3. Bandwidth of PCIe 3.0 and 4.0 during 1 TB data transfer
The second test integrated the same AMD EPYC servers into a 2-node Azure Stack HCI cluster to compare PCIe 3.0 and 4.0 adaptors for storage using VM Fleet. The result was 2x faster storage, with PCIe 3.0 achieving a data throughput of around 3 GB/s and PCIe 4.0 reaching over 6 GB/s, marking a significant benchmark breakthrough (see Figure 4).
Figure 4. Bandwidth of PCIe 3.0 and 4.0 on 2-node clusters
Prepare for a PCIe 4.0-ready future
QCT’s test results validated that the PCIe 4.0-based architecture can deliver the claimed performance gains. In these early days of adoption, PCIe 4.0 will mostly be applied to mid-range to high-performance computing for the first few years after its debut. And of all the performance gains, PCIe 4.0 achieves the most remarkable benefit in storage, which makes it highly suitable for I/O bound applications. With reference to the timeline of PCIe 3.0, the adoption and popularization of PCIe 4.0 will likely be a long march, but wherever you are located in the evolution roadmap, QCT is ready to tailor the most appropriate solutions for you.