|
|
 Sree Ganesan presented at AIFD3 |
Affordable Deep Learning Training with Intel Habana Gaudi
Watch on YouTube
Watch on Vimeo
Sree Ganesan, Head of Software Products for Habana Labs, introduces Habana Gaudi, an Intel company building purpose-built AI accelerators. Habana’s Gaudi is an affordable deep learning training accelerator available in the cloud with AWS EC2 DL1 and on-premises in a solution with Supermicro X12 Gaudi Server and DDN AI 400X2 storage. Since cost is the most significant challenge to implementing AI and ML solutions, Habana Labs focused on an optimized AI training solution when developing Gaudi, focused on homogeneous compute, software-managed HBM2 memory, and integrated 100Gb Ethernet RoCE.
Saudi-based Amazon AWS EC2 DL1 training instances offer 60% lower cost than EC2 P4d instances with ECS and EKS container support. Saudi’s SynapseAI software suite optimizes performance and usability of the Gaudi for PyTorch and TensorFlow users. Habana is also working with partners like Hugging Face, PyTorch Lightning, and cnvrg.io to maximize developer efficiency.
Habana’s Gaudi 2 was recently announced at Intel Vision and will be available in Q3 of 2022. Habana promises that Gaudi 2 offers 2x better throughput vs. NVIDIA’s A100 for popular vision and language models.
Personnel: Sree Ganesan
Accelerating Customer Applications With Intel Habana Gaudi
Watch on YouTube
Watch on Vimeo
Sree Ganesan, Head of Software Products for Habana Labs, discusses the applications and use cases for Habana Gaudi accelerators. Most customers use Gaudi for object detection and segmentation, image classification, and natural language processing. Specific examples include Leidos in health care, who are accelerating medical imaging to detect Covid in x-ray images using AWS DL1, as well as accelerating medical benefit applications using NLP-based deep learning. Riskfuel is accelerating FinTech workloads with ResNet inspired topologies. MobileEye is accelerating 2D and 3D object detection training with Gaudi and achieving better time to market as well as cost savings.
Turning to on-premises solutions, SDSC is using Supermicro X12 Gaudi AI training servers to accelerating science and engineering workloads. For more information, visit https://Habana.io or http://www.intel.lv/HabanaGaudi
Personnel: Sree Ganesan