Impressive. I must add, for those here who give a damn, that I’ve spent years in the high performance computing (HPC) field. Supercomputing, in other words; primarily Linux clusters. If you look at the Top 500 list, published twice a year, of the most powerful systems on the planet....you’ll see many of the systems I used to “own” for Big Blue (in a worldwide marketing sense).
In that field, nVidia’s GPU technology is a serious game changer. In HPC, clusters are used for serious modeling/simulation, and that means serious number crunching. GPU’s are perfect for this. Think of them as math co-processors on steroids.
Traditionally, we all used generic Intel or AMD CPU’s (almost always dual proc) “pizza box” servers lashed together for clusters. Blades came later, but the idea was still the same: lots and lots of “cheap” servers running off-the-shelf CPU’s lashed together to run massive problems.
The core issue there (no pun intended) is that CPU’s do certain things very well; they aren’t dedicated math co-processors. What nVidia and their brilliant Tesla architecture brought to the table is highly significant.
Now....and I’m pulling these numbers out of my, um, ear......rather than 20 racks of Intel or AMD CPU-based servers in a cluster, you may have five racks: 2 with CPU-based servers (or maybe 3), 3 racks of GPU’s (Tesla’s...maybe 2 racks). You get the idea. FAR smaller node count, and offloading the serious number crunching (utilizing CUDA) to the GPU’s vs. more generic CPU’s.
Huge cost savings, huge power savings, huge cooling savings, far more efficiency.
That, folks, in a nutshell, is the future (frankly, the real present) in supercomputing.....and nVidia deserves one hell of a lot of credit for it.
Gonna rip the Intel Atom...LOL!