Event Details
Per a recent (https://www.forbes.com/sites/bethkindig/2024/06/20/ai-power-consumption-rapidly-becoming-mission-critical/), "The rise of generative AI and surging GPU shipments is causing data centers to scale from tens of thousands to 100,000-plus accelerators, shifting the emphasis to power as a mission-critical problem to solve." making this talk particularly relevant.
The US Department of Energy projects that by the year 2055, given current projected growth, electrical demand for devices including the internet will exceed the capacity of planet Earth to provide that power. The electrical demands on data processing centers in a world of AI and cryptocurrency on top of standard workloads lead the industry down an unsustainable path. Hyperscale systems are feverishly adding processing, memory, storage, connectivity fabric, and communications at an alarming rate. Software engineers need to be part of the total cost of ownership calculation to address this energy crisis.
This talk details some key aspects of the system architectures that are being deployed in data centers, with focus on data movement between resources. Every time data is moved there is inherent inefficiency; we will calculate the wasted data movement in key parts of the system architecture. With this analysis, it is clear that the expansion of the data center architecture has exacerbated the variations in access latency time to each resource as well.
Software developers can leverage knowledge of the energy flow through the system to improve efficiency. Gains can be made on every level from the choice of machine level data types – to choice of high level languages – to various programmer interfaces between the application and the hardware resources. The variations in media access latency are only getting wider and more complex, so software developers are tasked with comprehending not only what data to process but where to place it in the system hierarchy.
The call to action will summarize some of the specific ways careful software engineering can help us avert the looming power crisis.
About the Speaker Bill Gervasi.
Bill Gervasi is Principal Systems Architect for Wolley Inc, designing advanced memory and storage solutions for data
NOTE: This is an IN PERSON meeting. To receive a link to the event's recording, complete the survey referenced in the header or footer.
Co-sponsored by: OC ACM and Knobbe Martens, an Intellectual Property & Technology law firm
Agenda:
6:30 PM Networking at physical meeting location
7:00 PM Announcements and Presentation with Q&A
8:00 PM Follow-up quesitons for presenter and networking
8:30 PM Meeting Adjourned
2040 Main St , Ste 14, Irvine, California, United States