Deep-learning suite speeds deployments on intelligent edge devices, data centers

A new suite of products and services help drive enterprise adoption of deep learning, at scale, on intelligent edge devices and in data centers.

The offerings build on DeepCube’s patented platform, which is the industry’s first software-based deep learning accelerator that drastically improves performance on any existing hardware. Now, DeepCube will offer solutions for neural network training and an inference engine respectively, allowing users to leverage DeepCube’s technology to address challenges in their deep learning pipeline. Additionally, a new service offering will make available DeepCube’s team of leading AI experts to support deep learning projects.

DeepCube’s new offerings include:
• CubeIQ: the first fully automated training framework that can take a model and surgically eliminate unnecessary parameters to ensure that physical, real-world constraints and profiles are met. CubeIQ trains models with a significant reduction in size, and with the pre-knowledge of the end target and environment. This leads to a drastic speed increase, minimized compute footprint, and efficient edge deployment – all while maintaining accuracy.
• CubeEngine: an inference engine designed to run next-generation deep learning models for optimal performance. CubeEngine is designed to accelerate CubeIQ generated models, by dynamically assigning the optimal kernels suitable for the specific hardware and model execution. CubeEngine is architected as a composable inference engine, unlike prior generation monolithic inference engines.
• CubeAdvisor: an expert-level service offered to leverage DeepCube’s wide-ranging ML experience, with guidance from some of the world’s leading AI experts and PhDs. It helps customers design, optimize and deploy deep learning models, ensuring customers accomplish the best performing model that fits the strict cost, performance, power, latency and other metrics of the chosen hardware.

To trial the new suite of products, DeepCube utilized 2nd Gen AMD EPYC-based cloud instances and the new DeepCube solutions to showcase high levels of inference performance on a multitude of popular neural networks, including ResNet-50, BERT-Large, and DLRM. To read more about this sweep of AI benchmarking on 2nd Gen AMD EPYC based cloud instances, stay tuned for an upcoming blog post.

“The offerings announced by DeepCube today are the culmination of decades of work and research by some of the world’s leading experts in deep learning,” said Michael Zimmerman, CEO at DeepCube. “We have long been focused on solving the technical challenges of training and inference for next-generation deep learning models, which is no easy feat – this is proven by the fact that so many enterprises are still unable to take their AI models out of the research stage. But we’re confident in the power of our patented technology, and by commercializing it through CubeIQ, CubeEngine and CubeAdvisor, we’re taking steps toward democratizing deep learning across industries.”

“AMD worked with DeepCube to preview the new solutions on cloud instances using 2nd Gen AMD EPYC processors, and saw fantastic performance for deep learning workloads,” said Kumaran Siva, corporate vice president, EPYC Cloud Business, AMD. “We believe that DeepCube’s innovative CubeIQ and CubeEngine products, coupled with optimizations specific to current and future generation AMD EPYC CPUs, will set a new bar for performance and business metrics, such as inference throughput, latency, and performance/dollar.”

For qualifying parties, DeepCube is offering a free trial license to CubeIQ and CubeEngine, with CubeAdvisor as an option.

For more information on DeepCube or the free trial, visit www.deepcube.com or reach out to info@deepcube.com.

For more information on 2nd Gen AMD EPYC processors, visit https://www.amd.com/en/processors/epyc-7002-series.