Sign up for our newsletter and get the latest HPC news and analysis.
Send me information from insideHPC:


Podcast: China Tianhe-3 Exascale machine is coming


 
In this podcast, the Radio Free HPC team takes a second look at China’s plans for the Tianhe-3 exascale supercomputer.

After a short talk about the weather in Henry’s basement (it had just reached 60 F by the time we recorded the show), we got right down to business with an important announcement: our pal Rich Brueckner is leaving the show. He just has too much on his plate and something had to give. While we’re worried about the impact Rich’s departure might have on our listenership, we did take note of and welcome listeners 13, 14, and 15, who made themselves known to Henry on one of his recent business trips. Yay us.

Our first topic is China rolling out a successor to Tianhe-1, dubbed Tianhe-3. According to news articles, Tianhe-3 will be 200 times faster than Tianhe-1, with 100x more storage. What we don’t know is if these comparisons are relative to Tianhe-1 or Tianhe 1A. The later machine weighs in at 2.256 PFlop/s which means that Tianhe-3 might be as fast as 450 PFlop/s when complete. We also made a reference to a past episode, which we know you remember vividly, where we discussed China’s three-pronged strategy for exascale.
As we’re moving into our popular “Catch of the Week” segment, Shahin hijacks the conversation by questioning if anyone knows the real-world utilization rates of non-commodity configurations in public clouds. This leads to this bold estimate from Dan “I’ll bet that there isn’t a public cloud out there that has a higher than 60-65% utilization rate.” We have a spirited discussion about this pseudo-metric and how infrastructures are sized to handle peaks. We also brought up a story that malware can bring down public clouds, although someone would have to own your system before doing it.

Catch of the Week:

Download the MP3 * Subscribe on iTunes * RSS Feed

Sign up for our insideHPC Newsletter

Leave a Comment

*

Resource Links: