Intel Corp. today introduced the PAC D5005, a new accelerator card that enables data center servers to run workloads such as analytics and artificial intelligence software faster.

An accelerator card is a device that can be plugged into a server to supplement its built-in processor. Habana Labs Ltd.’s recently unveiled Gaudi card, for instance, provides extra computing power for machine learning models.

The D5005 is different in that it’s not built to speed up a specific workload type such as AI. Instead, it’s programmable, meaning companies can use it for a range of applications.

Intel has a few specific use cases in mind. Patrick Dorsey, the head of the company’s Programmable Solutions Group, told SDxCentral that the D5005 lends itself particularly well to streaming analytics, media transcoding and AI workloads. He listed network security and financial software as two other workload categories the product can help speed up as well. 

The D5005 is a major improvement over Intel’s previous-generation programmable accelerator. It has four times as much memory, for a total of 32 gigabytes, and twice the number of programmable circuits that companies can customize for their workloads. There are also two 100-gigabit Ethernet ports where there was previously only one 40-gigabit port, which allows the card to ingest data at a significantly faster rate.

Under the hood, the D5005 is powered by a Xeon server CPU and a Stratix 10 SX field-programmable gate array. It’s the latter chip that provides the configurable circuits at the heart of the accelerator, while the Xeon processor handles more general-purpose tasks.

The D5005 will initially become available for Hewlett Packard Enterprise Co.’s ProLiant DL380 Gen10 server. Multiple other, unnamed server makers are currently certifying systems to work with the product as well. 

Intel is investing significant resources into its push to capture a bigger part of the programmable silicon market. Earlier this year, the company launched another accelerator card that is built to boost the performance of carrier networks. More recently, in April, Intel acquired a U.K.-based firm called Omnitech that developed field-programmable gate array technology for video and AI applications. 

Photo: Intel