NERSCPowering Scientific Discovery Since 1974

Performance comparison between Edison and Hopper

Edison used the same benchmark suite as Hopper, the NERSC-6 benchmark suite,  to measure system performance throughout its procurement process. Instead of peak flops, NERSC uses the sustained system performance (SSP) to measure the system computational capability. Edison is 2-3 times faster than Hopper with the seven applications used for SSP computation.

 

Application CAM GAMESS GTC IMPACT-T MAESTRO MILC PARATEC

Concurrency

240

1024

2048

1024

2048

8192

1024

Streams/Core

2

2

2

2

1

1

1

Edison Time (s)

273.08

1,125.80

863.88

579.78

935.45

446.36

173.51

Hopper Time(s)

348

1389

1338

618

1901

921

353

 Speedup1)

2.6

2.4

3.0

2.2

2.0

2.1

2.0

 

SSP

Edison

258

Hopper

144

 Note:

1) Speedup=[Time(Hopper)/Time(Edison)]*Streams/core, this is per node speedup.  Note the first four applications used Hyper-Threading on Edison, so only half as many nodes as Hopper were used.

2) SSP (sustained system performance) =(total # of cores of the system)x(Geometric mean of the flops/s/core of the seven application codes)

3) The performance numbers in the table above were obtained in the dedicated mode, i.e., there was no other users on the system.

 

These are the productin runs. Since many users do not use hyper-threading, we included Edison single stram runs as well.

Application CAM GAMESS GTC IMPACT-T MAESTRO MILC PARATEC
Hopper 347.13 1,215.63 1,417.84 642.78 2,298.37 1,175.44 399.67
Edison dual stream (s) 276.2 1,230.50 864.6 587.94 995.18 478.99 197.95
Edison single stream (s) 158.56 618.1 457.95 318.64 1,090.51 448.82 175.3
Speedup dual stream 2.5 2.0 3.3 2.2 2.3 2.5 2.0
speedup single stream 2.2 2.0 3.1 2.0 2.1 2.6 2.3