It’s just simpler and more efficient! Working in collaboration with GigaIO, SourceCode offers the SuperNODE for the growing set of applications that need many GPUs or FPGAs. Near-linear scalability is achieved through a fast interconnect and fabric-centric software.
GigaIO SuperNODE harnesses the combined computational power of multiple GPUs at ultra-low latency, providing the horsepower necessary for a robust AI infrastructure. It can use various accelerator technologies (like GPUs and FPGAs) at scale and with efficiency, reducing component and energy costs associated with multi-CPU systems.
Scaling GPUs over multiple nodes, SuperNODE demonstrates close to perfect scale factors. The Hashcat graph below shows GPUs scaling perfectly linearly all the way to 32. The ResNet50 scale factor drops to 99% for a 1% degradation per GPU, or an overall scale factor of 70%.
The GigaIO SuperNODE connects up to 32 AMD Instinct™ MI210 GPUs or 24 NVIDIA A100s to a single node at the same latency and performance as if they were physically located inside the server box.
GigaIO FabreX is a pioneering composable architecture that integrates computing, storage, and other communication I/O into a single-system cluster fabric, using PCIe technology to deliver the industry’s lowest latency and highest effective bandwidth. This seamless connectivity allows workloads to run more efficiently, utilizing the combined power of all accelerators in a single server, rather than spread across multiple nodes with complex networking.
