Generated by GPT-5-mini| POWER9 | |
|---|---|
![]() IBM España · Public domain · source | |
| Name | POWER9 |
| Developer | International Business Machines Corporation |
| Family | POWER |
| Released | 2017 |
| Architecture | POWER ISA |
| Cores | up to 24 |
| Process | 14 nm, 10 nm |
| Successor | POWER10 |
POWER9
POWER9 is a microprocessor family developed by International Business Machines Corporation and introduced in 2017 as part of the POWER architecture lineage. It served as a focal point for high-performance computing projects and enterprise servers, appearing in systems from vendors including Oracle Corporation, Dell Technologies, and Supermicro. POWER9 was positioned to compete with offerings from Intel Corporation and Advanced Micro Devices, and it played a role in initiatives involving Lawrence Livermore National Laboratory, Oak Ridge National Laboratory, and cloud providers.
POWER9 continued the evolution of the POWER architecture with emphasis on scale-out and scale-up deployments in IBM Power Systems servers and partner platforms. The design targeted workloads spanning high-performance computing, artificial intelligence, and data analytics used by institutions such as Argonne National Laboratory and companies like NVIDIA. Development efforts intersected with collaborations involving OpenPOWER Foundation, which included members such as Google LLC, Mellanox Technologies, and Tyan. Manufacturing partners included GlobalFoundries and Samsung Electronics.
The POWER9 family implements the POWER ISA with enhancements focused on throughput, memory bandwidth, and I/O. Microarchitectural features include out-of-order execution, multi-threading, and large caches, building on lessons from predecessors like microprocessors produced by IBM Research groups. POWER9 introduced designs supporting coherent accelerator interfaces such as NVLink developed by NVIDIA and high-speed fabrics like PCI Express generations used by vendors such as Broadcom Inc. The processor included support for coherent memory sharing with accelerators used in projects at organizations like Lawrence Berkeley National Laboratory.
Power9 chips were fabricated on processes provided by GlobalFoundries and benefitted from package technologies influenced by collaborations with NVIDIA and system integrators like Supermicro. The cores were organized in chip layouts optimized for symmetric multiprocessing and NUMA topologies favored by installations at Los Alamos National Laboratory and research centers. Instruction set extensions aided workloads deployed by companies such as Fujitsu and governments with computational requirements analogous to systems in use at NASA centers.
POWER9 shipped in multiple variants aimed at different market segments. Implementations included scale-out versions for cloud and hyperscale partners like Google Cloud and scale-up designs for enterprise customers such as Oracle Corporation's systems. Integrations were performed by system vendors including IBM, Dell Technologies, HPE, and ecosystem contributors listed in OpenPOWER Foundation membership. Accelerator-focused cards paired POWER9 hosts with NVIDIA Tesla GPUs and interconnects supplied by Mellanox Technologies for supercomputing clusters at institutions like Oak Ridge National Laboratory.
Commercial products based on POWER9 included offerings from IBM Power Systems, server boards from Tyan, and turnkey appliances from vendors like Lenovo. Custom efforts by research labs produced prototypes and reference designs used in projects at CERN and national supercomputing facilities.
POWER9 performance claims were evaluated in a range of benchmarks and real-world workloads. Comparative studies often involved processors from Intel Xeon line and offerings from Advanced Micro Devices in tests conducted by organizations such as SPEC benchmarking groups and academic institutions like University of California, Berkeley. POWER9 demonstrated strong memory bandwidth and I/O throughput in measurements against systems deployed at Argonne National Laboratory and within CORAL procurement projects for national centers.
In AI and machine learning contexts, POWER9 paired with NVIDIA DGX systems produced competitive throughput on frameworks used by teams at Stanford University and companies like Facebook, Inc., benefiting from NVLink interconnects. Benchmarks from independent labs and vendors highlighted strengths in multi-threaded server workloads and database processing tasks common in deployments at enterprises such as Deutsche Bank.
POWER9 was integrated into heterogeneous computing platforms emphasizing accelerator co-processing, coherent memory models, and high-speed interconnects. System integrators included IBM Power Systems, Supermicro, Lenovo, and hyperscalers like Google LLC. Platforms targeted by POWER9 encompassed on-premises enterprise servers, cloud infrastructure, and national laboratory supercomputers such as systems at Oak Ridge National Laboratory and Lawrence Livermore National Laboratory.
Integration efforts often referenced standards and consortia like OpenPOWER Foundation to enable ecosystem interoperability with vendors including Mellanox Technologies for networking and NVIDIA for GPU acceleration. Data center deployments used management tools and middleware provided by vendors such as Red Hat, Inc. and Canonical Ltd..
The software ecosystem around POWER9 included operating systems and toolchains ported by organizations such as Red Hat, Inc., SUSE, and Canonical Ltd. for distributions like Red Hat Enterprise Linux and Ubuntu. Compiler and performance tool support came from projects including GCC and commercial toolchains from IBM and partners. Machine learning frameworks like TensorFlow and PyTorch were adapted by research groups and companies for POWER9 plus NVIDIA GPU configurations in deployments at institutions such as Lawrence Berkeley National Laboratory.
Open-source communities and consortia such as OpenPOWER Foundation and projects hosted on platforms like GitHub played roles in driver, firmware, and system management development. Middleware and database vendors including Oracle Corporation and MongoDB, Inc. provided tuned builds for POWER architectures.
POWER9 included hardware mechanisms for reliability, availability, and serviceability used by enterprises such as Financial Times subscribers and government data centers. Features encompassed error-correcting codes for caches, memory protection implemented in silicon used by systems at NASA research centers, and secure boot and firmware chains supported by vendors like IBM and standards bodies referenced by Trusted Computing Group. Security mitigations and microcode updates were coordinated with vendors including Red Hat, Inc. and integrated into enterprise lifecycle processes at organizations like Deutsche Bank.
Category:IBM microprocessors