Qualcomm Datacenter Technologies May Have Improved Server TCO With Arm

Qualcomm Datacenter Technologies (QDT) announced real-world server workload benchmarks running on its Qualcomm Centriq 2400 system-on-chip (SoC) products. QDT also ran the same set of performance and power consumption benchmarks on competitive Intel Xeon Scalable systems. TIRIAS Research evaluated the test results and created a total cost of ownership (TCO) model for cloud service providers (CSPs), including the hosting community.

CSPs can use 1S Qualcomm Centriq 2400 per server for same HHVM perf per rack as Xeon Scalable 2S but using half power

Server Workload TCO Example: HHVM

One of the workloads QDT tested for cloud datacenter TCO is the popular HHVM web server framework, originally written by Facebook .

Our TCO analysis demonstrated that using only one Qualcomm Centriq 2452 SoC per server chassis, a 12kW rack full of 36 46-core SoCs should show slightly better performance than a rack full of Intel Xeon Silver 4110 dual-socket server chassis, at only 51% of the power consumption. That’s similar performance with about half the power consumption.

Using two Qualcomm Centriq 2452 SoCs per server chassis in a 12kW rack should yield a little over double the performance of the dual-socket Intel Xeon Silver 4110 servers at 88% of the power consumption. A key factor is that only 35 of the Intel Xeon Silver 4110 systems can fit within the 12kW rack power budget. In this scenario, Qualcomm Centriq 2400 offers double the performance with less power consumption.

The results of QDT’s Armv8 core and SoC design decisions are clear in our TCO analysis of several workloads running on Qualcomm Centriq 2452 server SoCs. If density is critical, a CSP can achieve twice the performance using dual Qualcomm Centriq 2400s per chassis. Otherwise, a CSP can provide the same performance in the same overall rack space using roughly half the power by using single Qualcomm Centriq 2400s per chassis.

Arm-based Server Ecosystem

Microsoft has already endorsed the Centriq 2452 for use in Azure’s OCP Project Olympus designs, in addition to other Arm-based server processors. Microsoft noted that the new crop of Armv8-based SoCs offers a compelling value proposition, more than just a price per performance play. Additionally, Hatch Entertainment has announced that it will use Qualcomm Centriq 2400 SoCs to power cloud gaming.

Qualcomm Centriq 2400's Advantage

I think the standout decisions in QDT’s SoC design were going 64-bit only (eliminating the extra logic required for legacy Arm instructions), eliminating scale-up in favor of scaling out single-socket server nodes and initial manufacturing in Samsung’s 10nm silicon process. For certain workloads, these decisions contributed to maximizing performance per core and the number of cores on a cost-effective SoC, plus minimizing per core and overall SoC power consumption.

For more detail than is appropriate in this post, continue reading here.

-- The author and members of the TIRIAS Research staff do not hold equity positions in any of the companies mentioned. TIRIAS Research tracks and consults for companies throughout the electronics ecosystem from semiconductors to systems and sensors to the cloud.