lowey@sask.UUCP (05/26/87)
Hi, Back by Popular Demand . . . Here are the results of my more "scientific" benchmark tests. I did the tests on 13 different machines covering a wide variety of configurations. I'll be posting the program used to do these tests in another message, so that you can pick apart my methodology at your leisure. Before I get started, please, no flames about not enough information about clock speed, graphics card used, manufacturer of the disk drives, or other equipment information. Most of these were done in a computer store just before closing, with sales people who didn't even know what MHz stands for. The results are shown in two tables. The first table shows mainly the computational intensive tasks. These tests could be run on all machines tested so they form a good basis for comparison. The second table compares disk speed and graphics speed, which vary a lot between machines. For example, one machine used a Monochrome card, so the graphics tests could not be done. The second table also gives more info on the Whetstone results. Some Notes on Accuracy: ~~~~~~~~~~~~~~~~~~~~~~ The timer routine I used is only accurate to .03 seconds. Since I call it twice for each test (once at the beginning and once at the end) each result is only accurate to .06 seconds. In addition, since the total in the first table is calculated from 8 values, the overall error for the total is 8 * 0.06= 0.48 seconds. These are all worst case figures. I could have averaged 5 runs on each machine, but my beard would be quite grey by the time it was finished! The Systems Tested: ~~~~~~~~~~~~~~~~~~~ Each system is numbered for reference in the tables. Numbers with a "*" beside them in the tables indicates that the 80287 Numeric co-processor was present. The systems are numbered in order of increasing speed when the Math Coprocessor is not included. All systems have an IBM-CGA compatible card, produced by the same manufacturer as the computer. The exception is the IBM-XT computer, which has an IBM Monochrome card. ============================================================================ # | Model | CPU | NDP | Clock | Notes ============================================================================ 1 | Compaq 386 (Model 2570) | 80386 | 80287 | | 2 | IBM Model 60 (8560) | 80286 | --- | | White Mono Display 3 | IBM Model 50 (8550) | 80286 | --- | | White Mono Display 4 | IBM AT (5170) | 80286 | 80287 | 8 | 5 | DEC VAXMATE | 80286 | 80287 | 8 | 6 | TANDY 3000 HD | 80286 | --- | 8 | 7 | IBM MODEL 30 (8530) | 8086 | --- | | 8 | AT&T 6300 | 8086 | --- | | AT&T Mono Graphics 9 | TANDY 1000 SX | 8088 | --- | 7.16 | 20M Hardcard 10 | ZENITH ZM-151-52 | V20 | --- | 4.77 | Identical to unit 13 11 | IBM-XT (5160) | 8088 | --- | 4.77 | IBM Mono 12 | AMIGA 100 + Sidecar | 8088 | --- | 4.77 | Dysan 20M Filecard 13 | ZENITH ZM-151-22 | 8088 | --- | 4.77 | Identical to unit 10 ============================================================================ Notes: - There are no 80387 chips out yet, so the Compaq 386 uses the 80286 chip - Units 10 and 13 are identical except for the V20 chip - All new IBM machines used the "paper white" monitor. - I'm not sure how accurate the clock speeds are - The Amiga tests were done emulating a Colour PC with 16 colours. The speed can be improved if less colours or the mono PC emulator is used. - The AMIGA had a 20M Dysan filecard partitioned for 10M for the amiga and 10M for the sidecar. - Unit 9 had a 20M hardcard for a hard disk. - All the 286 machines had 1.2M floppy disks, but all my tests were done using a 360K floppy. Benchmark Tests: All the tests were performed by booting my test diskette. This removed the problems encountered with memory resident programs, etc. The disks had no autoexec.bat or config.sys files. I used PC-DOS 3.2 on all disks requiring a 5.25 inch disk. I used PC-DOS 3.3 on disks booting from a 3.5" disk. PC-DOS probably gives an unfair advantage to IBM machines. The test program is written in Turbo Pascal. I compiled it under Turbo version 3.01A for PC-DOS. I did two versions, one for 8087 machines and another one for normal machines. The source code used was identical. All tests on floppy diskettes were done on either a 360K formatted 5.25" or 720K formatted 3.5" diskette. I performed 11 tests. The following is a short description of each: TEXT: Tested the time required to display 100 79 column lines in text mode. MEMORY: Moved a 4000 byte array from the data segment to the heap segment and back 2000 times. Sieve: Calculated prime numbers up to 16383 ten times Adapted from C program in byte magazine, November 1985 page 322 fibonacci Calculate 24th fibonacci number, from BYTE mag, Nov. 1985, p.280 Whetstone Do one iteration of the Whetstone test. Times are recorded for the entire test in the first table. Timer for the first and second pass, as well as whetstone Instructions per Second (WIPS) are stored in the second table. Intmath integer mathematics test, adapted from BYTE mag. Nov. 1985 p. 292 Float Test multiply and divide operations, from BYTE mag, Nov. 1985, p.284 trancendental Do tan(arctan(exp(ln(sqrt(i*1.0*i))))) 2500 times Disk Test Times how long it takes to write 160K to the disk. The possibility of buffering was not taken into account. Graphic Text: Same as the TEXT test above, except done in 640 by 200 Graphics mode Points test: plot all the points on the screen, in 640 by 200 graphics mode, starting from the bottom and working towards the top. The Results: ~~~~~~~~~~~~ Main Tests (Computers with a * are the 80x87 versions) ============================================================================== # | text | mem | sieve | fib | whet | int | float | trans | total ============================================================================== 1*| 3.46 | 8.35 | 1.49 | 7.36 | 12.58 | 2.58 | 15.60 | 5.93 | 57.35 4*| 4.72 | 12.97 | 3.46 | 18.18 | 15.55 | 5.71 | 18.78 | 5.54 | 84.91 5*| 4.45 | 12.96 | 3.51 | 18.29 | 16.70 | 5.71 | 20.10 | 5.88 | 87.60 1 | 3.46 | 8.35 | 1.38 | 7.31 | 20.33 | 2.59 | 26.96 | 22.69 | 93.07 2 | 3.13 | 10.27 | 2.80 | 14.50 | 37.67 | 4.51 | 48.55 | 42.34 | 163.77 3 | 3.14 | 10.30 | 2.80 | 14.61 | 38.01 | 4.55 | 48.67 | 42.68 | 164.84 4 | 4.67 | 12.91 | 3.52 | 18.18 | 47.40 | 5.66 | 60.81 | 53.06 | 206.21 5 | 4.45 | 12.96 | 3.52 | 18.23 | 47.67 | 5.66 | 60.91 | 53.22 | 206.62 6 | 6.26 | 12.96 | 3.51 | 18.18 | 47.68 | 5.66 | 60.75 | 53.11 | 208.11 7 | 5.49 | 34.72 | 6.21 | 29.22 | 68.60 | 17.30 | 88.21 | 74.59 | 324.34 8 | 7.41 | 39.44 | 6.87 | 32.40 | 77.17 | 18.68 | 99.69 | 83.98 | 365.64 9 | 10.77 | 48.72 | 13.51 | 69.81 | 153.79 | 30.26 | 191.25 | 165.54 | 683.65 10 | 10.28 | 29.39 | 13.40 | 68.93 | 169.78 | 25.76 | 212.78 | 187.35 | 717.67 11 | 9.83 | 60.91 | 13.68 | 72.62 | 175.37 | 34.38 | 220.42 | 192.89 | 780.10 12 | 10.76 | 60.91 | 13.62 | 72.67 | 175.60 | 34.44 | 220.48 | 192.89 | 781.37 13 | 10.44 | 60.97 | 13.62 | 72.72 | 175.65 | 34.44 | 220.69 | 193.06 | 781.59 ============================================================================== Other Tests (Disk speed and Graphics): ============================================================================== | Floppy Disk | Fixed Disk | Graphics || Whetstone test # | Time | Size | Time | Size | graph | points || Pass 1 | Pass 2 | WIPS ============================================================================== 1*| 57.56 | 360K | 6.15 | 20M | 8.02 | 5.77 || 1.26 | 11.32 | 99.40 4*| 60.47 | 360K | 11.70 | 32M | 6.16 | 11.64 || 1.54 | 13.95 | 80.58 5*| 72.61 | 360K | 11.59 | 20M | 9.33 | 17.79 || 1.65 | 14.94 | 75.24 1 | 56.18 | 360K | 6.26 | 20M | 8.02 | 5.77 || 1.82 | 18.51 | 59.92 2 | 71.46 | 720K | 11.04 | 44M | 6.54 | 10.21 || 3.29 | 34.38 | 32.16 3 | 71.40 | 720K | 14.61 | 20M | 6.53 | 10.27 || 3.30 | 34.66 | 31.89 4 | 59.32 | 360K | 11.76 | 32M | 6.16 | 11.70 || 4.18 | 43.17 | 25.65 5 | 71.23 | 360K | 12.85 | 20M | 9.29 | 17.80 || 4.23 | 43.34 | 25.57 6 | 59.38 | 360K | -- | -- | 7.91 | 13.01 || 4.29 | 43.34 | 25.61 7 | 71.90 | 720K | -- | -- | 9.67 | 19.05 || 6.04 | 62.51 | 17.71 8 | 73.11 | 360K | 33.01 | 10M | 12.63 | 25.92 || 6.81 | 70.36 | 15.74 9 | 71.07 | 360K | 60.86 | 20M H| 20.11 | 62.67 || 13.51 | 140.22 | 7.89 10 | 70.86 | 360K | 45.92 | 10M | 24.00 | 66.73 || 14.78 | 154.89 | 7.14 11 | 71.07 | 360K | 49.60 | 10M | -- | -- || 15.21 | 160.06 | 6.90 12 | 74.92 | 360K | 38.39 | 20M | 18.67 | 49.44 || 15.33 | 160.22 | 6.90 13 | 70.63 | 360K | 46.96 | 10M | 17.69 | 49.59 || 15.27 | 160.27 | 6.90 ============================================================================= Conclusions: ~~~~~~~~~~~~ Compare unit 10 with unit 13. These are identical machines except for the NEC V20 chip in one and the 8088 chip in the other. The tests show how the V20 is much faster at memory moves and mathematic calculations. Compare the results for the 80287 version of a machine versus the Non 80287 machine. It looks like the 80287 greatly increases floating math, especially in Trancendental functions. It is interesting though that it consistantly slows down disk speeds. I don't know if this is the chip or the 8087 version of Turbo Pascal. It looks like IBM's new bios does speed up the machines a bit. If anyone else runs the benchmarks, I'd appreciate getting the results. Happy Trails . . . ______________________________________________________________________________ | Kevin Lowey |The above is the personal opinion of Kevin | | University of Saskatchewan |Lowey. It does not reflect the position of| | Computing Services |the University of Saskatchewan in any way. | | SaskTel: (306) 966-4826 | | | Bitnet:LOWEY@SASK. (preferred) |I am in no way affiliated with any of the | | UUCP: ihnp4!sask!lowey.uucp |above mentioned companies other than U of S| |________________________________|___________________________________________|