Next Platform Radio The Next Platform

The Next Platform for March 16

For those who haven’t followed these devices closely, you heard it here, FPGAs are changing, they are truly moving toward what we would consider a Next Platform. We explore this in the context of their functions as hybrid compute and network devices first.

Later in the program we talk about storage that’s memory and while also storage, is also software. How’s that for a teaser? And later we’ll touch on some serverless and containerization trends at the end of the program.

A lot to cover, stay tuned.

On today's program we’ll take a dive into the number of two of the leading server makers to get a sense of future infrastructure spending and deployment trends. After that we’ll talk to Dr. Jeremy C. Smith of Oak Ridge National Lab about his team’s work developing a treatment for the cornovirus using the Summit supercomputer at ORNL and molecular dynamics modeling at massive scale. We’ll end with insights about being remote…more on that to come.

We are dedicating a good bit of the show today to a detailed wrapup of AMD’s financial analyst day because folks, this goes quite a bit beyond financial projections and into roadmap territory. whether you’re someone who buys compute or gets paid to think about the market for it, you’ll want to stay tuned. AMD is definitely back and this vision, well, let’s just say it’s bold.

After our analysis of that with our own Timothy Prickett Morgan we’ll talk about the near future of investments and capital for large-scale compute infrastructure and we’ll close the program with some Titanic-class autonomous news from today that should float your boat.

A quieter day than usual on the news front save one big item, the announcement of the much anticipated El Capitan supercomputer coming to Lawrence Livermore National Laboratory. We have some analysis on that coming up in a moment.

Later in the program we’ll be talking about CPU-based deep learning training. It’s been an uphill battle for chipmakers like Intel to make the case for CPUs as viable alternatives for training at scale but more work is showing how various optimizations, at least using TensorFlow, can take advantage of higher memory as training systems scale.

Other news commentary and analysis to boot. Thanks for listening.

On today’s edition, among many other things, we’ll be talking about Honeywell’s announcement that they will be bringing the quote “most powerful quantum computer” end to market. Now, that’s a tricky title to proclaim since it’s not necessarily based on the mainstream metric of quantum achievement, the connected. qubit count. TO get to the heart of this we’ll talk with Tony Uttly, Honeywell’s president of quantum solutions.

But first, we’ll take a look at a few key news items from the last twenty four hours and talk to some of those involved, including Mellanox about today’s acquisition of Titan-IC, Ayar Labs about its future in military and government following further strategic funding, this time from Lockheed Martin Ventures, the head of Darwin AI about the state of the black box problem in neural networks and where explanability will continue to matter, especially as enterprise AI grows.

We’ll offer some closing analysis on a tech topic that is not at all tehncial. Yet another event is out of the running for time on our schedules, the GPU Technology Conference has announced that it will be moving to an on-line only format. Some thoughts about that later.

We’ll start today with some insight around another item of news, the unveiling of startup Ampere’s 80-core server processor with our own Timothy Prickett Morgan, who just wrote an incredibly detailed piece after an in-deoth conversation with Ampere’s Renee James. That’s live on TNP now, go take a look.