Monthly Archives: March 2015

U of Witwatersrand Strikes again…

Often I get an idea and it seems the fact that others get it too kind of confirms somewhat that it was a pretty good idea. Ever mindful of a Friedrich Nietzsche quote, however…

The surest way to corrupt a youth is to instruct him to hold in higher esteem those who think alike than those who think differently.

At least keeps me from thinking its a great idea. Great ideas would have to be unique, or even revolutionary. It also makes me mindful that, perhaps… just perhaps… maybe it wasn’t such a good idea after all. Anyhow, as always, history will be the judge.

To wit, a few blog posts ago (on another blog) I mentioned I bought a Wandboard for one of my projects. I have to say, the thing has grown on me; I love it. Its become my little workhorse.

While shopping around for it, I thought, given its specs, it might make for a good cluster node. Being a bit more powerful then the Udoo Quads I own, which would be another good choice for a low-rent cluster as well. Anyhow, I stumbled across the University of Witwatersrand‘s High Energy Physics Group’s Wandboard cluster and thought, “…well ok then… someone else thought so too”.

So, yay, my nVidia Jetson TK1 arrived today, woohoo… 192 CUDA cores, it has most of the same important specs of the Wandboard, plus… 192 CUDA GPU cores! Given the state of computational science these days and everyone running to GPU’s for math or computation heavy tasks, I thought… hmm… might make a nice cluster node.

nVidia's Jetson TK1 Board

nVidia’s Jetson TK1 Dev board

Turns out, darn it… U of Witwatersrand thought so too… they just installed an 11 node Jetson cluster and are currently running performance tests on it. They are speculating that they are going to get some 3850 GFLOPS out of the cluster.

Hummpff… go figure… wish I had 11 Jetsons. 🙂

Hmm… ok then…

If anyone is reading my blog and recalls, I once opined on the state of semi-conductor technology, most notably how physicists and engineers extended the life of silicon transistors instead of having to resort to much more expensive gallium arsenide. I also mentioned that its likely that the performance envelope of silicon has probably been reached unless someone comes up with a good idea like the last “doping” process.

Well, Stanford beat everyone to the punch with a way to make gallium arsenide cheaper.

According to Aneesh Nainani at Stanford, the average gallium arsenide 8in disc, cost about $5000 while the average silicon 8in disc costs about $5, you can see the problem reaching an on par economy of scale using gallium.

However, Stanford found a solution that will dramatically reduce the cost… see here… for details.