IBM is the longest-standing and most technically adept server supplier. However it’s a long time since it was the market leader. This month it has added new processors which will help it increase the installed base (see my chart above), revenues (see the Figure below) and profits. Let’s look at the new processors.
IBM Power10
On September 8th IBM launched its new general purpose Power10 processors, putting it for the first time into its Power E1080 server. Designed by IBM Research, it is an 8-layer chip manufactured by Samsung – the first processor launched based on 7nm Extreme UltraViolet (EUV) process technology. Its aim is to help customers to run ‘a secured, frictionless hybrid cloud experience’. Its features include:
- Better integration with Red Hat software and throughput – the new machine is designed to meter Red Hat Enterprise Linux and OpenShift workloads, something it introduced with its Power Virtual Server (PowerVS) back in June 2019. Its throughput for these containerized applications is over four times greater per core than the best x86 processors available at launch.
- Improvement on its previous best server – 30% more performance and 50% better capacity per core than its predecessor, the E980, due in part to the reduced process size. At the socket and system level it has 50% better total capacity.
- Designed for AI – the E1080 also includes four Matrix Math Accelerators (MMAs) per core, multiplying the previous machine’s AI inference performance by five in comparison with the E980 server. Users who have built AI applications on x86 servers using Open Neural Network Exchange (ONNX) imbedded in application frameworks from TensorFlow and PyTorch can move them to the E1080 without having to recode them.
- Greener – servers based on the new chip offer users a saving of 33% on energy consumption for the same workload running on its older system. IBM claims that its new server will give customers even greater savings in comparison with x86 systems. To get the same performance a user could consolidate 126 x86 servers to just 2 E1080s, giving them a reduction in energy usage of 80% and of per-core software licenses of 70%.
- Better security tools – including ‘transparent memory encryption’, four times more encryption engines per core, delivering 2.5 times faster AES encryption than before on the E980.
At launch it published a new high-end record of 174,000 2-tier users running SAP’s Application Performance Standard benchmark on an 8-core system – 40% greater than the fastest x86-based platform. It also launched its new tiered Power Expert Care service for protecting customers from the latest cybersecurity threats and providing hardware and software coherence and higher system availability.
Financial institutions in particular have had a hard time using public cloud services for various reasons. IBM’s strategy of supporting ‘hybrid multicloud environments’ includes its Power Virtual Server application which provides architectural consistency between on-premise and cloud-based servers, which should see more companies running their high-security custom applications as cloud workloads. Its reference customers at launch included Finanz Informatik, a German IT services and outsourcing supplier for savings and state banks.
Its Power Private Cloud for Dynamic Capacity software also allows users to scale their usage up and down more easily, supporting more ‘pay-as-you-go’ payments for both cloud and on-premise server customers.
Perhaps because it has so many customers in the Finance and Central Government sectors, IBM has always offered extreme security on all of its systems. Its new Power system has security features everywhere across the system stack, including the processor, its memory, operating system, hypervisor and applications. The new system uses PowerVM as its built-in hypervisor, which has fewer Common Vulnerabilies and Exposures (CVEs) than its competitors as assessed by NIST’s National Vulnerabilities Database (NVD).
Telum as well
IBM also announced its Telum processor in September. Developed by IBM research over the last 3 years, it’s designed to support AI workloads such as real-time fraud detection and will be included in its future systems. It describes the chip as ‘on-chip hardware acceleration’ – something akin to a Field Programmable Gate Array (FPGAs) perhaps.
The Future
IBM Power is impressive as a processor – giving the systems in which they are used constant performance advantages over other RISC (such as Oracle’s Sparc) and x86 chips from Intel and, to a lesser extent, AMD. The challenge for IBM has shifted over time from competing with other proprietary servers, to competing with commercial x86 ones, to competing with Amazon’s – and similar – self-built x86 ones. Despite its strategic transformation to become a cloud supplier and a few attempts, it has not been successful in selling its processors to cloud suppliers’ – a market that is making Intel rich.
However it now has a major opportunity. Its acquisition of Red Hat, the adoption of OpenShift across the board and its detailed hybrid multi-cloud work gives it a technical lead over almost all other vendors. Given the difficulty chip fabricators such as GlobalFoundries and Intel have had in moving to 7nm processes its lead will probably be longer-lived than often before. Perhaps the biggest test of the success of its new Power chips will be in the proportion of them it deploys in its own Cloud data centers over time.