As the demand for multi-GPU compute continues to carve its place in the high performance computing (HPC) market, leading ODM’s like EchoStreams focus on innovative ways to achieve critical reliability in its unique designs.
"Rather than focusing strictly on performance density like many other vendors, a key benefit to EchoStreams' GPU server is its focus on reliability,” CTO of EchoStreams, Gene Lee said. “And since most of the deep learning algorithm occurs at the GPU’s core, it was absolutely critical we designed a GPU server that was able to deliver both performance but also sustain reliable GPU longevity; giving our customers a peace of mind.”
One real-world use case that shows great promise is a combination of central processing unit (CPU) and GPU acceleration in medical imaging and pattern recognition in brain scans. EchoStreams, AMD (NASDAQ: AMD) and the University of Southern California (USC) have collaborated to showcase the benefits of deep learning and machine intelligence in a medical application.
"Computational power required for processing patient data has been constantly growing with larger datasets and complex machine learning algorithms like deep neural networks," USC Research Fellow, Amir Jaberzadeh said. "It is shown that medical image computing algorithms scale very well when implemented on GPUs due to their parallel data structure.”
Together with Echostream’s 2RU / 6x GPU solution platform and AMD’s ROCm open software platform for GPU computing, USC researchers plan to publish a state-of-the-art Deep Learning algorithm for multiple sclerosis brain lesion segmentation by November 2018.
The collaboration with AMD and USC is a joint effort to bring a well-designed GPU platform to the open community, ultimately enabling greater adoption of this technology for academic and research institutions. The GridStreams GS206G-UN 2U 6x GPU server can support both the AMD Radeon Instinct™ MI25 accelerator and other Radeon™ Pro GPUs (Radeon Pro SSG & Radeon Pro Duo Graphics) with the Radeon Instinct™ MI25 compute GPU providing up to ~600 (TFLOPS) of raw GPU accelerated power for machine intelligence.
"By combining AMD EPYC™ processors and Radeon Instinct™ accelerators into unique HPC solutions, ODM partners like EchoStreams are able to leverage a robust portfolio of superior compute density and deep-learning products that are ideal for a new wave of machine intelligence technology fit for AI,” said Gregory Stoner, CTO of Machine Learning, Radeon Technologies Group, AMD. “New levels of innovation in machine learning are enabled through AMD’s introduction of