While there is a great amount of information available at DSL Reports SETI Forum regarding work unit processing times with PCs, there are virtually no benchmarks for the Macintosh platform. This brief report will provide information to Mac users who want some sort of "yardstick" to measure their performance. Generally, if a Mac user's individual SETI work unit times are in the range of 14 to 16 hours, they are meeting the performance of most Mac users world-wide. The following data was gathered from several sources including the SETI@home, Team MacAddict, and Apple Computer web sites.
Before reviewing and comparing any data, it is important to realize that comparing performance can be misleading, even on the same platform. Contrary to most advertising, cross-platform comparisons (i.e., PCs to Macs) are practically meaningless. It is almost impossible to compare different CPUs on any common criteria. Computer Professionals have been trying to do this for decades without much success. You might as well be discussing religion or politics.
One way to deal with this problem is to review the empirical data resulting from the millions of work units processed for SETI@home. When reviewing the average work unit times, it is important to note that SETI is not written to make the most efficient use of the platform it is compiled for. SETI is written to take advantage of large L2 cache sizes, and that is the most important issue when running SETI. Additionally, these average times probably include a large proportion of screensaver clients which are not configured to go to a blank screen and, therefore, increases the time per work unit significantly. On the other hand, there is probably an equally large proportion of users employing a RAM Disk to greatly decrease processing time. The most time-consuming part of SETI@home is the FFT routine that accesses a data set slightly larger than 512kB. Performance is much improved if this data set fits entirely in the L2 cache of the processor. Therefore, G3 and G4 PowerPCs with 1 MB or larger cache sizes will give fast performance. In addition, fast L2 cache speed (over 100 MHz) improves performance even more. Finally, some operating systems are more efficient than others. Though not as dramatic with respect to speed, a processor may be slightly more efficient under OS X or OS 9.1 than under OS 8.6 or earlier.
Data obtained from the statistics pages of SETI@home are sorted three ways, by CPU Types, by Operating Systems, and by Platforms. Here are the world-wide results for Macs as of Saturday, March 3, 2001:
Average CPU time per work unit, by CPU -- PowerPC 15 hr 43 min 40.1 sec
Average CPU time per work unit, by Operating System -- Mac OS 15 hr 48 min 33.9 sec
Average CPU time per work unit, by Platform -- Macintosh 15 hr 47 min 39.0 sec
The results for Team MacAddict and Apple Computer, Inc. are as follows:
Average CPU time per work unit, Team MacAddict -- 13 hr 53 min 40.6 sec (4753 users)
Average CPU time per work unit, Apple Computer -- 13 hr 52 min 55.8 sec (656 users)
Although SETI@home for Macintosh is designed to run on PowerPCs, including the 603 and 604 processors, it is probable that most users are running it on the 750 (G3) processor with a smaller number of users running it on the newer 7400 (G4) processor. The distribution will approximate a bell shaped curve, with slower processors (603 and 604s) on the left side, faster processors (G4s) on the right side, and a mix (mostly G3s) in the middle. With reference to this hypothetical curve, average work unit times will decrease from left to right. The differences between the overall SETI@home Macintosh data and the individual times for Team MacAddict and Apple Computer indicate different points on the curve (a different mix of processor types), with Team MacAddict and Apple Computer lying to the right of SETI@home.
The table below illustrates the effect on average WU times due to differences in specific Mac computers. Actual data from Team MacAddict is shown for five computer models, a mini-tower clone (StarMax), an iMac, an iBook, and two desktops. All computers use the same processor (G3), all use a RAM Disk, and all use the SETI screensaver set to blank the screen after one or two minutes.
Examination of the data in the table above shows the apparent effects of L2 cache size and speed, as well as bus speeds. At bus speeds of 100 MHz, doubling the cache size and increasing the processor speed by 100 MHz reduces the WU time by 1 to 3 hours. At lower bus speeds (66 MHz), the effect of doubling the cache size and increasing the processor speed by 100 MHz is even more apparent, 7 to 8 hours. Processor speed plays a very important role, of course, and very fast processors will decrease WU time dramatically. It appears that the combination of a fast processor with a large L2 cache and fast cache and bus speeds will produce the shortest WU times. As long as sufficient memory is available to run SETI@home, large amounts of memory do not seem to affect WU times appreciably. However, as stated earlier, comparing performance (even on the same platform) can lead to erroneous conclusions.
In summary, there is no simple "yardstick" to measure individual Macintosh performance. Generally, if your individual SETI work unit times are in the range of 14 to 16 hours, you are meeting the performance of most Mac users world-wide. If your times are longer, make sure you have taken advantage of all the tweaks, like using a RAM Disk and blanking the screensaver after a minute or two. You can find a link to all these tweaks in the "SETI Help Pages"at the top of the forum. Look for "OPTIMIZING SETI@HOME PROCESSING ON MACS". If you're still not satisfied, get a faster Mac. If your times are shorter than most Mac users world-wide, Congratulations!
For those who have an interest in work unit statistics for Team DSL Reports, including both PCs and Macs, clicking here will take you to Statistics on Work Unit Times for Different Processors by ScottMo.
You are visitor # to see this page.
