Both unstructured and structured data collected continue to grow. By 2020, we anticipate that the world’s volume of digital data will exceed 44 zettabytes, an astounding number.
The need to drive insights from this growing data has become ever more critical to differentiate in the market, and the need to support compute-intensive workloads continues to rise in order to remain competitive in the market.
Organizations are developing AI models to drive insights and enable the inferences to improve customer experiences. Machine learning models are becoming more and more complex with the increase in the amount of data, and deep learning models continue to gain traction as the need for more precise algorithms to differentiate in the market grows.
Every AI model is different and needs different precision and training levels. Furthermore, High Performance Computing (HPC) continues to grow in importance in many industries to drive these models and compute-intensive workloads.
IBM Cloud offers different performance-level GPU cards
To support customers’ HPC and AI workloads efficiently, IBM Cloud GPU offerings provide 17.3% better performance per dollar compared to AWS (Performance Report). We at IBM realize these varied needs, so we have enabled different performance-level GPU cards with bare metal servers and virtual instances:
- NVIDIA Tesla M60: Fundamental enterprise performance for virtualization and professional graphics.
- NVIDIA Tesla K80: Reliable enterprise performance for introductory AI computing.
- NVIDIA Tesla P100: Essential performance for growing advanced AI and HPC capabilities.
- NVIDIA Tesla V100: Maximum performance for progressive deep learning workloads.
Two additional GPU capabilities
IBM Cloud GPU offerings support many industries, such as financial, healthcare, and industrial sectors. As we continue to support the growing needs of these industries, we have updated our current GPU capabilities with two additional capabilities:
- 32GB V100 GPU Cards: Now available 32GB V100 GPU Cards for AI and HPC workloads in DAL13, WDC07, LO04, and FRA04. With 640 Tensor Cores, Tesla V100 is the world’s first GPU to break the 100 teraFLOPS (TFLOPS) barrier of deep learning performance. Compared to 16GB V100 GPU Cards, V100 GPU Cards with 32GB configuration take advantage of these high-performing cards to drive down time to train deep learning or machine learning models, whether its floating point precision or fixed point. To order please click here, or to learn more click here.
- GPU cards with Virtual Servers in Tokyo, Frankfurt, and Sydney: Now available P100 and V100 GPU Cards in additional data centers in FRA02, SYD04, and TOK02. Customers in these new data centers can now take advantage of multi-tenancy to reduce their cost of using GPU cards with virtual server offerings. Customers can use the GPU cards’ application on-demand without large compute bare metal servers if they need rapid deployment for a PoC and future scaling of their applications. To order please click here, or to learn more click here.
Benefits of GPU cards with bare metal vs. virtual instances
IBM Cloud Bare Metal Servers with GPU: High level of customization to fit high-performance workloads that can offload highly complex compute algorithms to the parallel processing of GPU without the noisy neighbors. Bare metal servers with GPUs can support larger workloads with more users, considerably reducing the cost of running applications. It also helps support use cases that could not be supported in the past as the number of applications for GPU cards continue to rise.
IBM Cloud Virtual Servers: Low provisioning time with pre-configured virtual servers with GPU cards allows customers to deploy high-performance workloads on their virtual machines at a very low price. Customers can run their PoC’s on the virtual machines to run experiments. Once mature, the workloads can be efficiently scaled or burst with hourly or monthly options.