San Diego Supercomputer Center and GigaIO on Composable Infrastructure
EVENT TOPIC
Heterogeneous Computing and Composable Architectures with GigaIO’s Next-Gen Interconnect
Has your computing infrastructure moved to heterogeneous computing, where GPUs, FPGAs and other accelerators are used in ever larger numbers and types?
Are you also trying to manage the need for multiple types of accelerators, each optimized for a particular step in the workflow – especially for artificial intelligence workloads?
If so you are not alone – most accelerated workflows also often require data to move to and from storage with each processing step, which leads to both slower application performance, and overburdened networks.
Faster interconnects are not sufficient to maximize performance in a heterogeneous environment. Instead, what is needed is a more efficient and dynamic interconnect fabric, enabling data to pass from one processing step to the next seamlessly, while reducing trips to and from storage.
Composable infrastructure is the key for effective heterogeneous computing, delivering the ability to balance CPU-to-GPU compute ratios, to create systems with different types and numbers of GPUs, as well as operate with optimal communications GPU-to-GPU and GPU-to-storage. The ability to scale solutions spanning multiple GPU appliances has also emerged as a key driver of composable infrastructure adoption.
While the complexity of the task has been increasing exponentially, with new accelerator technologies appearing weekly and algorithms ever more changing, sending many to the public cloud with its hefty price tag, GigaIO has developed a simple and elegant solution: a next generation interconnect fabric based on PCIe and CXL. Called FabreX™, it enables your on-premise resources to behave with the agility of the cloud, with the security and cost control of your own infrastructure. Our testing with SDSC described in this video demonstrated the promise, and led to a National Science Foundation award to SDSC with FabreX™ as the fabric of choice for the new Prototype National Research Platform – details here.
FEATURED SPEAKER
Alan Benjamin CEO
As a co-founder Alan leads the team at GigaIO bringing a universal, dynamic, and composable fabric platform to data centers, to enable widespread AI adoption by significantly reducing risk and TCO. Launched in 2017 by industry veterans in high performance interconnects, GigaIO has gained traction in high performance and AI applications with government labs, enterprise and cloud providers by finally breaking compute out of the server box to invent true rack-scale computing. Prior to GigaIO Alan spend 30 years in the communication industry, starting with Hewlett-Packard, and most recently as COO of Pulse Electronics, an $800M communication supplier, and CEO of Excelsus Technologies. He holds a BS Electrical Engineering from Duke University and completed the Advanced Management Program (AMP) at Harvard Business school.
Watch Now
Heterogeneous Computing and Composable Architectures with GigaIO’s Next-Gen Interconnect