Cache/Memory Performance Benchmarks of
Shared-Memory Multiprocessor Machines

Main plots

Elapsed time - Linear scale

These plots show the elapsed time for the main section of the test-threads benchmark, as a function of the number of parallel threads.

These are the three most significant tests, with the following configurations:

• dec.a8400c: DEC (Compaq) server, 8 CPUs alpha 21264 (EV6) 524 MHz, 8 GB main memory, 4 MB on-board cache plus 64+64 KB on-chip cache.
   
• sgi.momentum: SGI Origin, 32 CPU MIPS R12000 300 MHz, 32 GB of main memory, 8MB on-board cache, 32+32 KB on-chip cache.
   
• sun.buddy: Sun SPARC Ultra-Enterprise HPC6500, 32 CPUs sparcv9 400 MHz, 8MB cache (configured as a 20 CPU system by software).

For each machine, the program was executed with working sets of two different sizes---6.7 KB and 160 KB---and two different distributions in memory---compact, and scattered over 6 GB. The green line shows the number of processors available to the benchmark.

Elapsed time - Log scale

These plots show the same data in a log-log scale, so that the "compact" tests can be compared. The gray lines show the slope of the ideal parallel speedup, namely Time(N threads) = Time(1 thread)/N.

More plots

Alternative tests

Various IC machines