Something occurred to me recently. AI hardware system architectures are really applications-specific.
Computing platforms from laptops, desktops, workstations and servers follow classical architectures. Inputs, outputs, a central processing unit, busses (control, data and memory), data memory and program or instruction memory all balance in harmony in so many facets of our daily lives.
Granted, accelerators using FPGAs and GPUs are affecting system architectures. Distributed memory and and new and emerging fabrics will increase system performance and reduce latency. But core computing architectures haven’t changed much. I know this is an extreme over simplification, but it’s not untrue.
On the other hand, AI-focused system architecture from AI chipsets to the domain specific architecture are optimized specifically for the AI/ML algorithm the system is running. Maybe this is nuance for some. However for AI system architects, routing bleeding edge data rates from AI chipset to memory to I/O and more poses many challenges.
AI Hardware Summit Features Unique System-Level Solutions
Thankfully, at the AI Hardware Summit 2022, co-located with the AI Edge Summit, technical experts from Samtec will focus on systems-first AI/ML applications.
Also at the event, global AI/ML thought leaders from around the globe will collaborate and exchange ideas on next-gen AI solutions and architectures. IP providers, IC designers, interconnect solutions providers like Samtec and many more work together to create advanced AI/ML centric system solutions.
Fast, Efficient and Affordable AI. While that is the theme of this year’s event, “fast” speeds and new signal paths are always a focus as Samtec. We offer unique 112 Gbps PAM4 solutions for ever-changing AI hardware landscape. Fortunately, Samtec technical experts will discuss the latest AI interconnect technology at the event.
AI Interconnect Solutions from Samtec
Several new AI chipsets – SOCs, CPUs, GPUs, FPGAs, TPUs and others – feature 112 Gbps PAM4 performance. Samtec’s Bulls Eye® High-Performance Test System combination of compression interface, small footprint and high cycle count are ideal for AI chipset development and evaluation.
AI SoMs, CoMs and Carriers offer a complete AI subsystem on a single PCB. They also provide a path from prototype to production. Newer SoMs and CoM feature increased speed and density. Samtec’s NovaRay® Extreme Density Arrays and AcceleRate® HP High-Performance Arrays offer 112 Gbps PAM4 performance in high-speed board-to-board applications.
AI Accelerators offload repetitive processes from compute engines to optimize system performance and resource utilization. AI Accelerators utilize industry-standard form factors like PCI Express® or PECFF based on SNIA SFF-TA-1002 interconnect. Top card connectivity provides additional data paths for high-speed signals.
Lastly, AI application-specific architectures continue to emerge. From wafer scale processing to chiplets, unique substrate and package-based interconnect are key to optimized system performance. Front panel connectivity provided by solutions like Samtec’s NovaRay® I/O or Flyover® QSFP Systems offers AI system designers additional options for next-gen architectures.
Register to Attend
Samtec is proud to be an Event Partner at AI Hardware Summit. Join us at AI Hardware Summit 2022. Register here.
Leave a Reply