The IBM Power9 CPU is specifically designed for today's dataintensive and AI workloads. With the processor and the IBM Power 9 System AC922, training times in the areas of machine learning and deep learning can be accelerated fourfold. Get to know the power of the latest IBM Power9 chip generation: Test your AI applications in the MEGWARE Benchmark Center on our IBM Power 9 System AC922 - free of charge!
- The bandwidth offered to transfer the data into the V100 Tensor Core GPUs is 150GB/s per GPU while the bandwith on the V100 Tensor Core GPUs is even 7 TB per second.
- Power9 uses the latest generation of the PCI Express standard, PCIe Gen4. This connects peripheral devices with the chipset of the main processor. Compared to the previous version, PCIe Gen4 allows approximately twice the data bandwidth.
- In addition to PCIe 4, the Power9 chip has other advanced I/O connections such as CAPI 2.0, OpenCAPI and above all NVLink. The latter is a high-speed bus from Nvidia, which was developed for data transport between GPU nodes and CPUs. The combination of Power9 and NVLink offers a 5.6 times higher data throughput than comparable PCIe Gen3 systems.
- There are 8 billion transistors on the Power9 processor. This transistor form factor density, which IBM has developed in its Power9 cores, results in even faster processing speeds and higher performance.
- Finally, Power9 supports complex model calculations with up to 2TB of RAM memory, simplifying the execution of enterprise AI applications.
The AC922 is designed to develop precise AI applications with high data throughput and very short training time.
Flexible application possibilities
Create and train AI models with flexible deployment options - on-premises or in the cloud
Innovation without borders
Use popular open source frameworks, software and tools within a closed ecosystem to deliver the entire training.
Integrated end-to-end security
The reliability of power systems and IBM-secured open source frameworks leads to a maximum of security.
MEGWARE enables you to test the IBM Power 9 System AC922 in our in-house Benchmark Center free of charge. You will receive a test account for this purpose, which you can access remotely. Our HPC and AI engineers will support you in setting up the account.
The MEGWARE Benchmark Center has many years of experience in practical performance measurement. Our methodical competence ensures meaningful comparative figures. Thus, we support you in selecting the suitable components for your AI applications.
The test option can be used for an unlimited time. We will agree with all interested parties on an individually suitable time window for the test. Thus the test can be interrupted for a few days and resumed at a later date.
At the Max Planck Institute for Dynamics of Complex Systems, around 230 employees are researching new methods for the synthesis, analysis, dynamics and control of highly complex processes for chemical production, biotechnology and energy conversion. The research concept uses model-based methods from systems process engineering, systems theory and scientific computing. This is where Power9 comes into play.
The direct NVLink connection between the POWER9 CPU and GPUs allows efficient use of GPUs even for small problems. While conventional systems are still busy transferring data between memory and graphics memory, the GPUs on the POWER systems are already starting to work. From experience, I can say that the IBM POWER9 / NVIDIA Tesla V100 is more likely to solve problems with GPU support than comparable x86 systems.
Artificial intelligence is conquering everyday business life. Experts estimate that in 2019 AI technologies will have an impact on sales of 220.6 billion euros in Germany. The front-runner is the automotive industry, followed by consumer goods production and mechanical engineering.
A recent market survey commissioned by Tech Data also shows that German companies are not lagging behind the trend technology AI. According to the survey, almost a third of German companies (250 employees and more) are currently in the evaluation and planning phase. 21 percent are working on a proof of concept or developing prototypes and 16 percent are already busy with the introduction.
What is the reason for the steadily increasing use of AI technologies? In the past, AI projects have failed due to technical hurdles of various kinds. However, what was often missing in the past is now available: in addition to attractively priced storage solutions, the necessary computer resources. With the high performance of CPUs such as Power9, machine learning and deep learning projects can be successfully implemented.