oracle engineer wrote:ServeTheHome asked Oracle why it chose to create a cluster of Raspberry Pis instead of using a virtualized Arm server and one company rep said simply that "...a big cluster is cool."
It won't be so cool when they upgrade it to use Pi4Bs.
It'll be cooler but too hot for youdavidcoton wrote: ↑Sat Sep 21, 2019 8:50 pmIt won't be so cool when they upgrade it to use Pi4Bs.
Except it's not, its an example of what you can do with $37,100 worth of computers plus many thousands of dollars in additional equipment.We don't expect this product to go commercial, but it is a really neat example of just how much you can do with a $35 computer.
The built in network socket on the 3B+ IS a USB network adaptor (as were the ones on all B models prior to that).
In a way I'm disappointed Oracle didn't build their cluster using SPARC-based single board computers. Maybe the Raspberry SPARC was hastily cancelled after the Pi 4B released.
Completely wrong. Simply because the definition of supercomputer change every year. The Cray 1 was a supercomputer. The Pi4 is faster than a
The fact that you need to explain this proves that most folks don't have a clue about how supercomputers work.jamesh wrote: ↑Sun Sep 22, 2019 9:43 amAnd most certainly not a publicity stunt. This is a means of testing software destined for top of the performance league computers wich cost many millions, without having to waste development time on those very devices. Debug your distributed code on a Pi cluster, very cheaply, then move it to the real workhorse. That would pay for this cluster in a matter of days. Modern supercomputer time is expensive.
Is that where the term "hot seat" came from?
Now that the Pi 4B has true  gigabit Ethernet, it makes a much more reasonable machine to build clusters out of. It would be amusing to see a Raspberry Pi 4B entry in the student cluster competition at ISC 2020. In the past the goal has been to run a number of computational codes as fast as possible using less than 3kW. As each Pi uses about 5W, that would be about 600 Raspberry Pi computers.jcyr wrote: ↑Sun Sep 22, 2019 2:16 pmThey are not calling a 50 year old design a supercomputer, they are calling a current design a supercomputer. Furthermore, the super slow interconnect matrix using USB 2.0 constrained EThernet I would consider a disqualifier.
I think you missed out a "Gigabit" in there
Good catch. I've fixed it.
The intuitive definition of a supercomputer is that at the time of delivery it is among the fastest machines in the world. Not so intuitively, the performance cutoff is to be one of the 500 best performing machines in the world when solving huge systems of linear equations. Everything else is better called a cluster.
That page is a bit out of date. Fortunately, the department cluster was dismantled to make room for office space and observe fire codes that weren't properly implemented in the new building.Heater wrote: ↑Sun Sep 22, 2019 4:55 pmYou need to get some students on the case. Your PDEWulf cluster seems to be a bit neglected: http://fractal.math.unr.edu/~ejolson/ap ... owulf.html
I'm wondering what one can do with 25000 GPIOs ?