On Monday, IBM announced that it collaborated with Nvidia to provide a complete package for customers wanting to jump right into the deep learning market without all the hassles of determining and setting up the perfect combination of hardware and software. The company also revealed that a cloud-based model is available as well that eliminates the need to install local hardware and software.
To trace this project, we have to jump back to September when IBM launched a new series of “OpenPower” servers that rely on the company’s Power8 processor. The launch was notable because this chip features integrated NVLink technology, a proprietary communications link created by Nvidia that directly connects the central processor to a Nvidia-based graphics processor, namely the Tesla P100 in this case. Server-focused x86 processors provided by Intel and AMD don’t have this type of integrated connectivity between the CPU and GPU.
According to IBM, the data thrashed between an IBM-based NVLink-embedded CPU and a Nvdia-based GPU moves 2.5 times faster than the movement of data between an AMD/Intel processor and a graphics chip using a PCI Express connection. More specifically, the NVLink communication can shove data at 80GB per second while x86 servers relying on PCI Express only moves data at 32GB per second.
The server IBM is supplying in its new bundle is the Power S822LC designed for high performance computing, which is the foundation for deep learning, machine learning, and artificial intelligence services. This is a 2U rack-mounted server, which can measure up to 3.5 inches in height. This unit consists of two eight-core or ten-core Power8 processors, up to 1TB of system memory, and up to four Tesla P100 graphics chips.
In a pre-release phone call with IBM, the company indicated that one Power8 CPU in the server is assigned two P100 GPUs. There are two NVLink lanes between a single CPU and a single GPU, and two NVLink lanes between the paired GPUs. Thus, the two connected GPUs have an incredibly fast access to large data sets residing in the system memory due to the NVLink connections, and the 115GB-per-second link between the CPU and the system memory. Additionally, the link between both sets of the CPU/GPU trio provides speeds of 38.4GB per second.
Outside the hardware, IBM’s new bundle consists of a deep learning software toolkit called PowerAI. This combines pre-compiled deep learning frameworks (Caffe, Torch, Theano, and OpenBLAS) into one package aimed at Ubuntu 16.04 running on an IBM Power processor, Nvidia CUDA v8.0, and Nvidia cuDNN 5.1.
That’s a lot of server talk we know, so just rest assured that IBM has provided a hardware/software solution that can be used to quickly set up a scalable platform for creating AI-reliant services.
“The new solution supports emerging computing methods of artificial intelligence, particularly deep learning,” the company stated on Monday. “IBM PowerAI also provides a continued path for Watson, IBM’s cognitive solutions platform, to extend its artificial intelligence expertise in the enterprise by using several deep learning methods to train Watson.”
In the pre-announcement briefing, IBM stressed that the PowerAI toolkit will make setting up a platform for AI quick and easy. The bundle of deep learning software is optimized and tested on IBM’s hardware for “ease of implementation.” The company also provides documents via IBM OpenPower to make the installation process a breeze, and promises to continue optimizing the software for future Power-based hardware.
But what does all of this mean for the general consumer? IBM is setting service developers up for a more personal experience. For instance, the hardware/software bundle will likely be used to create an intelligent system for detecting retail bank fraud. This service would compare the face provided in physical IDs that are used across multiple bank branches against the face of the actual customer standing in front of the counter.
Another scenario would be chat bots and call center automation. In both situations, artificial intelligence backed by IBM’s new bundle would quickly understand speech and language, and then answer questions in a more human-like manner. Services would also benefit from a smarter search system thanks to deep learning powered by the hardware/software combo.
“PowerAI democratizes deep learning and other advanced analytic technologies by giving enterprise data scientists and research scientists alike an easy-to-deploy platform to rapidly advance their journey on AI,” said Ken King, General Manager, OpenPower. “Coupled with our high-performance computing servers built for AI, IBM provides what we believe is the best platform for enterprises building AI-based software, whether it’s chatbots for customer engagement, or real-time analysis of social media data.”
Finally, the second component of IBM’s announcement focused on Nimbix, a high performing computing (HPC) cloud platform provider. Nimbix expanded its cloud-based services by installing IBM’s Power S822LC server and the PowerAI suite. Thanks to this offering, customers not wanting to purchase and install the hardware/software combo into a data center can create services that rely on computing in the cloud.
IBM indicated in the briefing that the new Nimbix solution is ideal for high performance computing and deep learning, and won’t require that customers create a virtual machine to provide their services. One of the current compute applications listed on the Nimbix website is machine learning inference demonstration software with a starting price of $5 per hour.