Warning: "continue" targeting switch is equivalent to "break". Did you mean to use "continue 2"? in /homepages/9/d659054407/htdocs/app660029368/wp-content/plugins/magneto-sidebar-generator/html.php on line 83

Warning: session_start(): Cannot start session when headers already sent in /homepages/9/d659054407/htdocs/app660029368/wp-content/plugins/wpmu-dev-facebook/lib/external/facebook.php on line 60

Warning: session_start(): Cannot start session when headers already sent in /homepages/9/d659054407/htdocs/app660029368/wp-content/plugins/wpmu-dev-facebook/lib/external/facebook.php on line 60

Warning: session_start(): Cannot start session when headers already sent in /homepages/9/d659054407/htdocs/app660029368/wp-content/plugins/wpmu-dev-facebook/lib/external/facebook.php on line 60

Warning: session_start(): Cannot start session when headers already sent in /homepages/9/d659054407/htdocs/app660029368/wp-content/plugins/wpmu-dev-facebook/lib/external/facebook.php on line 60
ARM servers becoming more of a reality with X-Gene 3 - Tech Altar

There have been several different companies who have been interested in taking ARM architectures and applying them to the server, or micro-server, environment. The natural efficiency inherent in current designs could make them a great candidate for certain types of workloads. MACOM has more details concerning their upcoming X-Gene 3, 32-core Server-on-a-Chip, looks to compete with low-power Intel E5 Xeons in addition to the lower power Atom-based Xeons.

X-Gene 3

X-Gene 3 Server-on-a-Chip boasts big improvements for all workloads

One of the large improvements that should benefit a number of different server workloads is the massive amounts of memory bandwidth available through 8 channels of DDR4 running nominally at 2667MHz. Compared to even Intel’s Skylake-EP, it can achieve higher theoretical throughput, capable of having 170GB/s of total memory bandwidth with which to use for various kinds of workloads. Big data appreciates high memory bandwidth, as do databases and scientific computing. But there’s more significance to the X-Gene 3 and it’s lower power design coming to servers.

X-Gene 3

Though the TDP is a rather high 125W (which refers to the heat output, not power consumption) the ARMv8-based SoC is being touted as being faster than than current ARM designs, and not merely due to the higher clock speeds they’ve achieved. The X-Gene architecture, is said to be on par with the E5-2680v4 in terms of integer performance, though it may lag behind slightly with floating-point performance. The only benchmark they’ve quoted is SPEC_int2006 which achieved a peak score of 24. For comparison, an AMD Opteron 5180, based on Bulldozer, achieves a peak INT score of 23.8.

Aside from pure CPU performance, the 42-PCIe lanes could enable co-processor arrangements that are suitable for low-power specific workloads, such as even connecting a bevy of NVIDIA P100’s or AMD’s Instinct Deep Learning Accelerators.

The plan, however, is to compete in a ~$1700 monetary bracket, though they note that the final price will likely be 1/3 the price of the 14-core Broadwell-E 2680 v4, use less power at a similar TDP and potentially be a great competitor.

ARM has made some headway in the server world with partnerships from HP and their Moonshot device. Facebook, Microsoft and others have expressed an interest in the micro-server deployment model, saving power where the actual workload may be slightly smaller. Here they’ll be able to compete in a similar performance threshold for slightly less and with just as much potential for a wide-range of activities.

This is good news and provides more motivation for Intel to improve their own hyper-efficient server CPUs. Competition can only be a good thing in this sector, especially with the impending return of AMD to the server-space.