IBM has just announced a new, high-end, Power10-based server, the Power E1080. The company’s press release positions the new server as a “frictionless,” scalable, hybrid cloud consolidation server – with improvements in security, artificial intelligence, scalability, performance and metering. I, however, see this new server somewhat differently…
Power10 is the hugest differentiator
The big news to me regarding the Power E1080 is the use of IBM’s Power10 processor. As I stated in last year’s Power review (found here), “Power Systems offer a microprocessor [now the Power10] that is distinctly different from z & x86 processors. Power processors offer extremely fast I/O; they can be closely integrated with graphical processing units (GPUs) to achieve extremely high performance levels; they offer very tight security and have access to massive amounts of main memory (see here). These differences mean that there are workloads that Power Systems can process most assuredly better than x86-based servers.” To me, the emphasis of this announcement should have been not only on the “hybridcloudedness” of the new servers but also on the types of workloads these new, high-end servers can now handle.
What types of workloads will Power10-based servers excel at processing? The most obvious are data-intensive workloads that benefit from large memory configurations. For instance, Power systems excel at processing SAP HANA in-memory database applications (as described here), which is why IBM is reporting that its 8-socket Power E1080 servers now provide “world-record performance” measured by the SAP SD standard application benchmark running SAP ERP 6.0 EHP5 on a Power10 3.55-4.0 GHz processor, with 4,096 GB memory, 8p/120c/960t, with 174,000 SD benchmark users (955,050 SAPS), using AIX 7.2, with DB2 11.5. (Certification # 2021059). All results can be found at sap.com/benchmark.
Further, note the processing power and memory speed advantages that the new system offers. With more processing power per core, the Power E1080 provides 4X the socket performance of a 16-socket Intel2 and 2.7X the socket performance of an 8-socket Intel3. To further extend its advantage over Intel-based servers, IBM has also introduced a new memory architecture (with a new Open Memory Interface [OMI]) that delivers twice the memory RAS [reliability, availability, serviceability] as compared to standard DIMMS). Compute-intensive applications that can exploit large memory belong on Power systems.
Other compute-intensive applications that run ideally on a Power E1080 include business analytics applications that use data sampling (a statistical analysis technique) to select, massage and analyze data point subsets to identify patterns in larger data supersets. The Power10 microprocessor includes four matrix math accelerators per core that enable five times faster inference performance than the previous generation Power9-based Power system (the Power E980). Inference performance is measured in data samples processed per second (higher is better). A 5:1 performance improvement for business analytics applications over previous generation Power systems running pattern analysis, Power10-based servers offer an especially noteworthy improvement in analytics performance.
Not only do Power10-based servers have significant performance advantages, but they also burn less energy – providing the ability to add 50% more capacity while using the same amount of power as the previous generation.
The hybrid cloud dream is a reality – Now simplifying pay-for-use pricing
Way back in 2009, IDG News Service Green Computing quoted me as saying: “If you kept installing all of these different hypervisor products on your x86 and then a different management scheme on your Power systems and a different one on your mainframe and a different one on your Itanium systems … you’re going to end up with a whole bunch of different stacks.” What a difference a decade makes…
Fast forward to today. Now, in 2021, IBM’s dream of a transparently blended “frictionless” cloud environment (hybrid public/private clouds) has become a reality. As I explained last year in this Clabby Analytics Report, containerization is key to the hybrid cloud. “A container is a software environment that contains a complete deployment unit that allows an application to be automated, tracked and rapidly deployed. It differs from a virtual machine by allowing multiple workloads to run on an operating system, rather than running multiple OS instances on underlying virtual machines.) In short, containers are more efficient (less resource intensive), more flexible (from a development/deployment perspective), and more secure than the now “traditional” virtualized resource approach to computing.”
From a strategic perspective, IBM is highly focused on using Red Hat OpenShift cloud architecture as the basis to integrate a myriad of systems and cloud architectures (public and private) into a unified hybrid cloud environment. To do this, IBM’s Power organization has ensured that its Power AIX, IBM i and Enterprise Linux operating environments and data can transparently participate in a hybrid cloud environment with other systems and other cloud architectures. As part of this strategy, containers are managed by industry-standard Kubernetes. The way IBM is containerizing its various systems/storage solutions relies on using Kubernetes (as the cloud management environment/control plane) that provides self-service capabilities while delivering additional scalability, agility and portability.
With Power already established as frictionless players in the hybrid cloud environment (per last year’s report), IBM focused its efforts this year on making it less costly to operate Red Hat software on Power systems. As part of its announcement of the Power E1080, IBM described its planned “industry-first” – the by-the-minute metering of Red Hat software, including Red Hat OpenShift and Red Hat Enterprise Linux. By metering usage, users can expect to see better utilization of their Power systems and more than 4x greater Red Hat OpenShift containerized-throughput-per core on Power10-based servers (as compared to x86-based servers).
The other “Big News”: Security progress
I’ve written in the past about how Power offers very secure workload isolation, advanced data protection and solid platform integrity. As part of this Power E1080 announcement, IBM announced transparent memory encryption with no additional management setup, four times the number of encryption engines per core (allowing for 2.5x faster AES encryption as compared to the IBM Power E980), and a re-emphasis that IBM’s Power offers security management software for every level of the system stack.
Still, at times, I write about the exciting field of Quantum Computing – and I worry that, from a security perspective, quantum computing will eventually be used to crack traditional key-encrypted computers. System makers would be wise to start planning today for quantum security invasions of the future. And this is why I was glad to see IBM work into its announcement the themes of end-to-end data protection with full-stack encryption — and especially glad to see IBM position Power to deal with future Quantum incursions with support for Quantum-safe cryptography and fully homomorphic encryption directions. It is comforting to know that a leader in the development of quantum computers is also taking steps to prevent their misuse.
For years I’ve classified IBM Power as “high-performance compute-intensive platforms.” But, I’ve also written about why Power is better suited for specific compute-intensive and general-purpose workloads than Intel-based servers (Power Systems have colossal performance and cost advantages). Still, many IT executives continue to purchase less-efficient Intel servers as a matter of course (an ingrained purchase pattern coupled with skillset concerns). I’d like to see this pattern broken – and a new emphasis put on running the right workload on the right platform.
Cloud computing makes it easier to move workloads to “the right platform.” With all of the work that IBM has been doing in DevOps to make its servers look and act like other cloud servers, and with all the work that IBM and others have done to break down infrastructure and cross-system management barriers, I’d like to see Power servers reevaluated as a matter of course in IT buying decisions. Many workloads are being run on x86-based servers today that belong on Power servers. The move to hybrid clouds has leveled the playing field. Now it’s time to deploy the right systems for the right jobs.