DARPA’s “Exascale Challenge” asks vendors to design a supercomputer than can cram a petaflop of power into a single rack. Several vendors have begun on the effort, and SGI’s entry was codenamed ‘Project Mojo’. Folks at The Register got some details on the system, and… Well, first read the brief description:
As it turns out, the stick of the Project Mojo system is a computing element that is nearly as long as the rack is deep – three feet – with the width and a little more than the height of a double-wide PCI-Express peripheral card. Mannel wouldn’t say what processor is implemented on the stick, but it is possible that SGI has variants with both Intel Xeon and AMD Opteron processors. Considering that Project Mojo is an experimental system with limited sales on the front end, it is reasonable to conjecture that SGI will start with Xeons and expand into Opterons if there is customer demand.
So an element is 3-feet deep, and roughly the size of a double-wide PCIe card, and (if you read a bit further) you can cram 80 of them into a full-size rack.
This sounds mighty familiar. In fact, over at my “other job”, I have one of the BOXX “RenderFarm on Wheels” systems containing their renderBOXX module. The renderBOXX module is 32-inches deep (4 inches shy of 3 feet), 7 inches tall and under 4 inches wide, but each module actually contains 2 nodes. In a full-size rack, you can pack in 80 nodes.
So basically, it sounds like SGI is taking BOXX’s RenderBOXX system and splitting the nodes in half, and found a way to pack in a passively-cooled GPU. Nothing too earth shattering there.
I hope they have some bigger plans coming down the pipe.
