On Next Platform TV for today, July 29 we delve into a congressional perspective on AI investments; discuss a new path to low precision in ASICs; talk about the 8-socket server market; and also touch on object storage insights.
On today’s show we discuss the future of AI investments from a government perspective with U.S. House of Representative, Jerry McNerney, who servers as co-chair of the Congressional Artificial Intelligence Caucus.
We also spoke to James Garland of Trinity College Dublin about how an old bitslicing technique from the dawn of time is being recast to bring tunable mixed precision to the vector units in Arm and Xeon processors, allowing them to do inference precisely and with a lot more throughput than using FP32.
Supporting the S3 object protocol is not the same thing as being able to run the same object storage on premises and across various public clouds. And we talked Gary Ogasawara, chief technology officer at Cloudian, one of the few free-standing, neutral object storage players left, about how to create a single substrate for object storage.
And finally, we spoke to Jay Shenoy of Hyve Solutions, one of the big suppliers of iron to hyperscalers and an early adopter of Open Compute designs, about its entry into the eight-socket server market and why hyperscalers are choosing big iron for certain workloads — particularly machine learning and analytics — instead of conventional and less costly two-socket servers.
Thanks as always for tuning in. Timestamps for specific interviews below.
2:04 – Congressional insights on AI investments locally, nationally
11:26 – Bitslicing could disrupt everything in the FPGA for DL market
22:28 – The 8-socket server market; trends and momentum
33:25 – The future of object storage, especially in hyperscale