Short: Benchmark utility for all Amigas. Architecture: m68k-amigaos SSSpeed ========= Supervisor Software Benchmark program for all Amiga computers version 0.56 This software is still under development but the benchmark results are believed to be reliable and accurate. Any comments and suggestions welcome. SSSpeed is designed to measure the system performance under normal conditions. That is, it uses the default task priority and does not disable any interrupts. All other tasks and processes are allowed to run as usually. SSSpeed contains several benchmarks which are all allowed to run for five seconds. The results are reported separately so that the user can see what parts of his/her system perform well and which do not. All the results are proportional to that of a plain A1000 (with one disk drive and 512K of CHIP RAM). So, 100% means that the test runs exactly as fast as on a plain A1000. Some accelerator boards without fast 32-bit RAM may even run slower than a plain 68000-based Amiga unde r certain conditions. That's partly because the 68020 and 68030 processors fetch a long word (32 bits) from the memory even when the program only requests for a byte. One 32-bit access means two separate 16-bit accesses on the 16-bit Amiga data bus. Additional FAST RAM improves the benchmark results, even if it was a 16-bit RAM board. Using heavy display DMA will slow down the CHIP memory tests (and the whole machine if there is no FAST RAM available). All hard disk partitions, external disk drives and other devices may slow the machine a bit because their processes need some CPU time even when there is no hard disk activity. ----- Usage: This version of SSSpeed contains three different benchmarks which can all be run separately. The benchmarks are enabled individually using options Benchmarks, IOaccess, and Memory. The first executes several different benchmarks and outputs the results relative to a plain Amiga A1000. The second options tests the access speed of CIA and Custom chip registers and ExecBase (memory location $4.L). The results are given in nanoseconds (1E-9 seconds). The last option runs a memory access test and outputs the access times in nanoseconds. The access time tests are not very accurate in this version of SSSpeed because the operating speed of different CPUs varies. However, the results are somewhat interesting when different Amiga models are being compared. For example, A3000 seems to have faster access to CIA chips than the older Amigas (1317 ns on A3000/25 vs. 2727 ns on B2000/25). The access time test tries to determine the minimum and maximum values of the access times. This is because CHIP RAM access times seem to vary depending on the other loading of the CHIP bus and synchronizing of CPU/CHIP buses. If the minimum and maximum times differ more than 10 percent, they will both be printed. Otherwise the average access time is given. All results will vary depending on the loading on your machine. Each disk drive, hard disk partition etc. has a process of its own which takes up some CPU time even when the drive is not in use. Moving the mouse, pressing any keys, swapping disks etc. also consumes some CPU time. Leave the machine alone while running SSSpeed to get the most accurate results. To see more information about the actual benchmarks performed use SSSpeed ? or SSSpeed ! ############################################################################## Finally, some benchmark results from an A3000 with 8MB of FAST Page Mode RAM and several hard disk partitions installed: ------ Fast1:Asm/Sources/SSSpeed 9> ssspeed b m io NOTE: The following data is somewhat unreliable. Memory access times (in nanoseconds): byte word long word odd long word CHIP 741-938 743-937 744-937 1409 FAST 79 79 80 120 ROM 79 79 80 119 NOTE: The following data is somewhat unreliable. I/O hardware access speed (in nanoseconds): byte word long word CIA 1317 ----- ----- Custom ----- 716-932 ----- ExecBase ----- ----- 629-938 CPU/FPU performance results: Integer String Float(FFP) Float(ieee) Trig(ieee) CPU 955% 684% 490% 1331% 15102% FPU ----- ----- ----- 2998% 24242% Program execution speed: Memory copy speed: CHIP RAM FAST RAM CHIP RAM FAST RAM CPU 204% 408% 399% 741% Test completed. ############################################################################## GVP B2000-Combo 68040@33MHz, 16MB FAST RAM, FASTROM, DATA CACHE, NO BURST CPU/FPU performance results: Integer String Float(FFP) Float(ieee) Trig(ieee) CPU 2516% 2774% 1710% 8553% 17844% FPU ----- ----- ----- 24041% 18909% Program execution speed: Memory copy speed: CHIP RAM FAST RAM CHIP RAM FAST RAM CPU 134% 1894% 144% 1738% Test completed. ############################################################################## ------ NOTES: ~~~~~~ Memory access times are affected by caches. To see the actual RAM speeds disable data cache using Dave Haynie's SetCpu or a similar prg. Performance values may be slightly different from those measured with SSSpeed 0.25 because the benchmark timing has been changed to use the timer.device. The results should be more precise now. ------ Future: I'm going to build a small database in SSSpeed. It will contain the performance results of the most popular accelerator boards and all basic Amiga configurations. It will be possible to see all the results compared to any known system by using a command line option. For now, all results are proportional to that of an A1000. New benchmarks will probably be added. Please send any comments, bug reports and benchmark results to jmarin@messi.uku.fi or Jukka Marin Metsurintie 17 B 8 70150 Kuopio FINLAND Revision history 0.52 - no longer allocates hardware timer unless it is really needed 0.53t - execution tests now run a 8KB loop to give correct results on 68040 machines using 4KB caches 0.54 - uses CacheClearU() to flush CPU caches after copying code. Should now be fully '040 compatible. 0.56 - Now runs the FPUieee test on 68040 FPU and the FPUtrig test if 68040.library is present.