Today IBM announced a new AI Inference POWER9 Server. The new IBM Power System IC922 is a purpose-built inference server designed to put your AI models to work and help you unlock business insights.
The IBM Power IC922 supports up to six NVIDIA T4 GPUs and IBM intends to support up to eight NVIDIA T4 GPUs and additional accelerator options. This allows clients to leverage the inference accelerator that best suit their needs with flexibility. Power IC922 uses optimized hardware and AI software to deliver the necessary components for AI inference whether in a central data center or in a distributed data center closer to the sources of data.
Specifications:
- Fast POWER9 processors 40 CPU cores @ 3.8 GHz
- Up to 2 TB RAM
- 10 PCIe adapter slots including Gen 4 slots
- 24 Disks for 92 PB of internal storage
- Up to 6 NVIDIA GPU adapters for Inferencing AI models built on the AC922 and much more as a general fast Linux computer for 9 TB Storage, 10 PCIe adapters slots including Gen 4 and 2 TB RAM
Data, Train, Inference (DTI) workflow
According to Grace Liu from IBM, the traditional Data-Train-Inference (DTI) model for machine learning is not a linear workflow. It is instead a continuous loop consisting of three stages that interact at all times. And because the process is ongoing and continually being refined, the insights extracted are rich and valuable.
The DTI model has three main puzzle pieces: Data, Train, and Inference.
- Data – the build-a-solid-foundation piece. Not starting with a solid data foundation will send you down the wrong path before you’ve even started.
- Training – the piece where the magic of artificial intelligence occurs; where data becomes AI models.
- Inference – the piece that is really the sum of all the parts. Without proper inference, all prior efforts are for naught.
To complete your AI puzzle and reach success, it is necessary to have the right pieces of infrastructure to support each stage in the AI workflow. So how does DTI map to the IBM Power Systems hardware portfolio? For training, there’s the IBM Power System AC922 packed with up to six NVIDIA V100 GPUs and engineered to be the most powerful training platform.
Storage-rich with 24 SAS/SATA storage bays (future NVMe support intended), Power IC922 also fits neatly into the Data phase to help clients build a solid data foundation. Additionally, leveraging the advanced IO architecture and data throughput in Power IC922 can allow for rapid response requests needed in AI puzzles. Power IC922 is configurable so whether you are looking for a data server or an inference server, Power IC922 offers a needed piece to your AI puzzle.
To showcase how the IC922 fits into the AI puzzle, the U.S. Army Combat Capabilities Development Command’s Army Research Laboratory recently demonstrated how the IC922 and AC922 could be combined into a modular computing platform, creating an IBM POWER9-based supercomputer in a shipping container. This will allow them to redefine the term “edge” as they are equipping the Army to deploy an AI supercomputer anywhere in the field, including the battlefield.
Sign up for our insideHPC Newsletter