"Extra" Top Crunch Results - LS-DYNA on Sun Fire X2100
Thursday Apr 20, 2006
Yesterday, I posted a blog entry about a Sun Fire X2100 cluster World Record press release, now here is more data...
The Sun Fire X2100 beats all other HW vendor clusters with LS-DYNA results currently posted on the "Topcrunch" website. These server clusters were also equipped with high performance interconnects: CRAY XD1 Single & Dual Core Opteron w/Rapid Array, IBM eserver P5 575 w/ HPS, HP CP4000 Opteron DL145 w/ IB (Topspin), HP CP6000 I-2 rx2600 w/ IB (TopSpin), etc. As of 4/17/06, results are available at http://www.topcrunch.org/benchmark_results.sfe
LS-DYNA is a prominent crash code used in the automotive industry. There is a standardized benchmark test suite and website with results from different hardware vendor cluster configurations. This application is similar to other types of MCAE: cpu intensive floating-point operations, memory intensive, high I/O speed, intense higher performing interconnects performance.
Two benchmarks are measured: "Neon_Refined" and a much larger "3 Vehicle Collision." Sun Fire X2100 surpassed all published results, ranging from one to 32 processors, . The scalability of these codes is non-linear as the size of the cluster is growing, with the typical sweet spot of 16 processors, where Sun's solution maintained 80% efficiency. Additionally, on the "Neon Refined" benchmark, the 16 Sun Fire X2100 servers are 17% faster than the same size HP system, based on HP ProLiant DL145 servers; 29% quicker than the IBM system, based on eServer 326 and Myrinet interconnect; and 56% speedier than the Itanium2-powered HP system, based on Integrity rx2600 Servers.
| System/Processor | Interconnect | nodes x CPUs | Time (smaller=better) |
| Sun X2100 3.0 GHz Opteron 156 | IB(Topspin) | 31 x 1 | 5680 sec |
| HP CP4000 2.6GHz Opteron DL145 | IB (Topspin) | 16 x 2 | 6347 sec |
| IBM eserver P5 575 1.9 GHz | HPS | 4 x 8 | 6379 sec |
| CRAY XD1 2.2 GHZ Dual Core Opteron | Rapid Array | 16 x 2 x 1 | 6652 sec |
| SGI Altix3700 1.6 GHz I-2 | NUMALINK | 1 x32 | 6672 sec |
| CRAY XD1 2.4 GHZ Opteron | Rapid Array | 16 x 2 | 6878 sec |
| CRAY XD1 2.2 GHz Dual Core Opteron | Rapid Array | 8 x 2 x 2 | 7591 sec |
| HP CP4000 2.4GHz Opteron DL145 | Myrinet | 16 x 2 | 8024 sec |
| HP CP6000 1.5GHz I-2 rx2600 | IB (TopSpin) | 16 x 2 | 8189 sec |
| Galactic GT4000 3.6Ghz Xeon | IB | 16 x 2 | 8433 sec |
Footnotes:
Neon_Refined Model: 1 processor time is 9222 sec., efficiency is 100.00%; 2 processor time is 4629 sec., efficiency is 94.25%; 4 processor time is 2446 sec., efficiency is 94.25%; 8 processor time is 1283 sec., efficiency is 89.85%; 16 processor time is 703 sec., efficiency is 81.99%.
(4) Sun Fire X2100 cluster: Time - 703 sec.; HP CP4000 cluster, ProLiant DL145: Time - 822 sec.; IBM eServer 326 cluster: Time - 911 sec.; HP CP6000 cluster, Integrity rx2600 cluster: Time - 1101 sec.
|
|
|
|
