WRF performance
Dec. 14th, 2007 06:21 pmThis is interesting: a bunch of researchers have broken various performance records using WRF. They've managed to get 8.8 TFLOPS out of a Cray XT4 with a 100 TFLOPS peak and 3.4 out of a Blue Gene/L with a peak of 103 TFLOPS. Very impressive.
Unfortunately, the Japanese got there first. They managed to get a sustained 36 TFLOPS out of a climate model — OK, a bogus one using an Eulerian dynamical core to artificially bump up the flop counts — but they managed it. And they managed it back in 2002.
Unfortunately, the Japanese got there first. They managed to get a sustained 36 TFLOPS out of a climate model — OK, a bogus one using an Eulerian dynamical core to artificially bump up the flop counts — but they managed it. And they managed it back in 2002.
no subject
Date: 2007-12-14 11:37 pm (UTC)no subject
Date: 2007-12-15 11:22 am (UTC)