The latest addition to Amazon Web Services Inc.’s ever-growing selection of cloud instances is a pair of beefy bare-metal servers aimed at enterprises running SAP SE applications.

The machines, unveiled today, are joining the EC2 High Memory instance family AWS launched last year. They both pack eight Intel Corp. central processing units with a 2.7-gigahertz base clock rate and a combined 448 processing cores.

The main difference between the two instances is the memory allocation. AWS said that the first, dubbed 18tb1.metal, packs 18 tebibytes of memory while the bigger 18tb1.metal is configured with a hefty 24 tebibytes.

That translates to roughly 20 terabytes and 26 terabytes, respectively. This generous RAM pool makes the instances highly suitable for running SAP’s widely-deployed business management applications, which use a memory-hungry database called HANA under the hood to process data.

Another advantage the instances have over the older members of the EC2 High Memory family is that they provide four times as much network bandwidth. Each server can transfer data at rates of up to 100 gigabits per second. This means that applications can not only ingest more information from external systems, but also do so faster.

AWS is initially making the instances available from its North Virginia data center with more locations set to be added later. The new hardware options effectively one-up rival Google LLC, which last month introduced two HANA-optimized instances with 5.8 and 11.7 terabytes of memory. AWS now matches Microsoft Corp., its other major competitor in the cloud market, whose own HANA-optimized virtual machines max out at 26 terabytes of RAM.

The instances continue the trend of cloud providers developing highly niche hardware offerings to compete for high-value enterprise workloads. Another notable example is Google’s TPU Pods. Introduced in May, TPU Pods are essentially supercomputers for rent that allow organizations to provision up to 107.5 petaflops of processing power for machine learning projects. 

Photo: Robert Hof/SiliconANGLE