In an interesting twist, SuperMicro just won the bid to supply NASA with a major compute upgrade. SemiAccurate has been saying such rack-scale integration projects are the future, and it looks like NASA agrees.
The NASA Center for Climate Simulation (NCCS) at Goddard Space Flight Center has just announce their SCU14 upgrade. It consists of SuperMicro Fat Twin servers in three configurations, compute nodes, service nodes, and I/O nodes, all of which do what they sound like. All three nodes use the base 4U 8-node Fat Twin chassis and have 2x 20C Xeons in them with room for 12 DIMMs each. The total compute power for this addition is 1.56PF.
As you might expect from a showcase win, this one features most of the SuperMicro goodie catalog from the Motivair chilled doors to the Omnipath ToR switches. If you are thinking this reads like a showcase for Intel technologies too, that pretty much bang on too, the two companies are pretty close you know.
On the storage front, if you guessed Intel SSDs, you would be right too. The compute nodes are have persistent flash storage, a must for stateful HPC machines, but the large storage needs are on a separate Fiber Channel SAN. If you haven’t figured it out by now, that is what the I/O nodes do, like we said, the names are pretty self-explanatory.
So with this win, NASA gets their Goddard HPC cluster bumped up from 3.5PF to just over 5PF, a pretty substantial jump. More important is that SuperMicro gets a big win for their integration arm, NASA is about as high up on the image front as you can ask for. Why is this important? Before you read this, did you know SuperMicro did rack-scale integration projects like this at their factory?S|A