Dell EMC and Fujitsu are the first top-tier server makers to incorporate programmable processors from Intel into their systems, a move that officials with the chip maker say is an initial step that will lead to other vendors offering the acceleration technology in their offerings.
Intel is announcing April 11 that the two system OEMs will offer servers with the field-programmable gate arrays (FPGAs), chips that can be used in conjunction with Intel’s Xeon processors to improve the performance and power efficiency of the servers while running compute- and data-intensive workloads that are becoming more commonplace in the data center.
Enterprises are becoming inundated with data with the rise of such trends as the internet of things (IoT) and cloud computing, and are having to deal with modern applications like machine learning and artificial intelligence (AI), data analytics, video transcoding and cyber-security. FPGAs can bring levels of flexibility, latency, performance and efficiency to enable organizations to better and more quickly handle those workloads and take advantage of the data, according to Sabrina Gomez, director of platform solutions marketing at Intel.
“There is so much data,” Gomez told eWEEK. “There is a need for faster analysis, and the demand for faster decision-making is growing.”
FPGAs, which are chips that can be programmed and reprogrammed through software to adapt to whatever workload enterprises are running at the time, are one of a growing number of accelerators used along with CPUs in systems to increase the performance of applications running on systems while keeping a lid on power consumption. Nvidia more than a decade ago began offering GPUs as accelerators, a move that caught on quickly in the high-performance computing (HPC) space and in supercomputers and has since cascaded down into mainstream enterprise data centers. Both Nvidia and Advanced Micro Devices offer such general-purpose GPUs as offload engines, enabling the CPU to push some tasks to the GPUs while working away on others.
Intel’s attempt to build its own GPU accelerators faltered, but the work led to the chip maker’s Xeon Phi efforts to develop x86-based many-core coprocessors that worked in similar fashion as the GPUs. Intel eventually made Xeon Phi “Knights Landing” chips that could be used as primary processors, though the company last year pulled back on the development of some of the coprocessors. In 2015, Intel paid $16.7 billion for Altera, one of two major FPGA makers. Xilinx is the other. Intel already offers some GPUs and ASIC chips, and buying Altera gave Intel another acceleration technology to offer enterprises.
Intel in 2017 expanded its FPGA portfolio, including rolling out the Programmable Acceleration Card (PAC) that included the vendor’s Arria 10GX FPGA and could be plugged into a server. The company also launched the Acceleration Stack for Intel Xeon Scalable processor with FPGA, a collection of firmware, software, tools, APIs and an FPGA interface manager to enable applications and accelerator function developers to more easily code for Intel FPGA platforms. Organizations have been able to use the programmable FPGA card with their systems, but Dell EMC and Fujitsu will now put the FPGA technology in some of their servers.
Dell EMC is offering the acceleration technology in its 2U (3.5-inch), dual-socket PowerEdge R740 and R740XD and 1U (1.75-inch), dual-socket R640 systems. The first two servers will hold up to four FPGAs, with the R640 supporting one, according to a Dell EMC spokesperson. Those systems are available now.
Fujitsu currently is offering FPGAs in servers as “early access for priority customers” now, with the accelerators coming in the Primergy RX2540 M4 in the near future.
Gomez said Intel is talking with other OEMs as well as system integrators and value-added resellers (VARs) and expects announcements similar to those from Dell EMC and Fujitsu down the road.
Dell EMC and Fujitsu putting the FPGAs into their systems is an important move, according to Patrick Moorhead, principal analyst with Moor Insights and Strategy.
“I see this as the first indication that FPGAs could become more mainstream in the data center,” Moorhead told eWEEK. “The industry is looking for different ways to accelerate workloads as Moore’s Law has slowed down a bit.”
Intel’s Gomez noted a range of use cases for FPGA acceleration, including data analytics, AI, video transcoding, genomics, cyber-security and financial services. In tests run by the vendor on relational databases using the Swarm64 technology, FPGAs helped increase real-time analytics by 20 times, traditional data warehousing by more than two times and storage compression threefold.
FPGAs give Intel another technology to offer as enterprise data centers become increasingly heterogeneous environments, where organizations want to have the right tools for a particular job. For acceleration, they now have more choices that include FPGAs, GPUs, ASICs and x86 coprocessors.
“GPUs and FPGAs are both programmable, and I believe it all comes down to the specific accelerated workload at a specific point in time,” Moorhead said. “One big benefit that FPGAs have is that they can be hardware programmable. One day they can be doing deep packet network inspection and the next hour they can be doing machine learning for inferencing. Not every data center needs morphing like that, but some will. I also like that FPGAs only consume power based on the gates used.”
The FPGAs also show a high level of flexibility within Intel to embrace heterogeneous computing, the analyst said.
“Less than three years ago, it can be argued that Intel really only cared about its x86 franchise,” he said. “Since then, Intel bought Altera, Nervana, Movidius and MobilEye and hired AMD’s chief of graphics, Raja Koduri. I’d say they have embraced accelerators as a core function for the future. Now the company has to deliver.”
Intel in 2016 bought startup Nervana to expand its capabilities in machine learning and AI and Movidius for computer vision technology used in drones and other systems. Last year, the company spent $15.3 billion for Mobileye, a company that developed computer vision technology for autonomous vehicles.