IBM POWER10 PROCESSORS
IBM POWER10 processoris designed to deliver significantly higher encryption performance with four times the number of AES encryption units per core compared to the IBM POWER9 processor for today’s most demanding standards.
The new CPU core architecture in the IBM POWER10 processor has a built-in matrix accelerator that is extrapolated to provide 10x, 15x and 20x faster AI inference for FP32, BFloat16 and INT8 calculations per socket, respectively, than the IBM POWER9 processor. This allows you to bring artificial intelligence into enterprise applications and gain greater insight.
IBM POWER10 PROCESSOR BENEFITS
7nm processor technology delivers up to a threefold increase in capacity and energy efficiency over IBM POWER9
Memory Inception technologysupports multi-petabyte memory clusters and improves cloud capacity and cost-effectiveness for memory-intensive ISVworkloads as well as for inferring large AI models
hardware-enabled security features including transparent memory encryption designed to support end-to-end security
architecture achieves significantly higher encryption performance with four times the number of AES encryption units per core compared to the IBM POWER9 processor
CPU core architecture has a matrix accelerator that provides up to 20x faster per socket derivation of FP32, BFloat16 and INT8 AI computations than previous generation CPUs
IBM POWER E1080 SERVERS
The IBM Power10 generation of E1080 servers is designed for seamless use of technology. With Hybrid Cloud Credits, enterprises can deploy capacity across Power Private Cloud and Power Virtual Server. Architectural consistency across these environments provides deployment flexibility without the need for additional middleware or application reformatting.
With IBM Power10 technology, the IBM Power E1080 server can deliver up to 30 % more performance per core and more than 50 % more overall capacity at the socket and system level compared to previous-generation servers.
The IBM Power E1080 server and Power10 processor bring new capabilities to enterprise AI. Four IBM Power10 Matrix Math Accelerator (MMA) cores that can deliver up to a five-fold improvement in AI inference compared to the IBM Power E980v server. While the new increased MMA count provides hardware-focused improvements, the IBM Power E1080 also supports bring-your-own-model capabilities with IBM Auto-AI and “no-code” tools. The IBM Power E1080 also supports Open Neural Network Exchange, or ONNX. Trained AI models available on ONNX, using some of the most popular frameworks such as TensorFlow, PyTorch and others, can be deployed on the IBM Power E1080 from x86-based servers without requiring code changes.