Core Microarchitecture Performance: Woodcrest Preview

Pages: 1 2 3 4 5 6 7 8 9 10


Our last and unarguably most significant test is SPECjbb2005, which we have previously described. Out of our suite of benchmarks, SPECjbb2005 is the most directly correlated with commercial server performance. It is also an industry standard benchmark that is widely used by every vendor. Submissions for SPECjbb2005 range from small 1P systems, to massive 128P Itanium2 systems from SGI. The command line options used were:

-server –XXaggresive -XXtlasize64k –Xms2040M –Xmx2040M

Each warehouse independently spawns a thread in the benchmark, thus determining the level of concurrency in the benchmark. Each system has an expected peak number of warehouses (N), which corresponds to the total number of hardware threads supported. The score, measured in Business OPerations per second (BOP/s) is actually an average of the throughputs for N, N+1, N+2…2N warehouses. Figure 8 below shows the results from SPECjbb2005, with the relevant scores circled in red.

Figure 8 – SPECjbb2005 Performance

The Woodcrest system scored 56,804, 56,835 and 56,678 BOP/s for our three runs, while the Dempsey box managed 47,930, 47,449 and 47,707 BOP/s. In comparison, the best published two socket Opteron system, scores only 51,106 BOP/s, and that is with extensive tuning efforts.

Considering that our JVM is tuned for Dempsey, but not for Woodcrest, a 20% performance lead is very respectable. Moreover, our test uses a 32b JVM, which does not perform as well as a fully configured and tweaked 64b JVM. It is quite likely that a fully tuned Woodcrest system would score in the 70-90K range, which would easily put Woodcrest ahead of all other competitors in the two socket space.

Pages: « Prev   1 2 3 4 5 6 7 8 9 10   Next »

Discuss (22 comments)