"Home Computer" Benchmarks

How fast were those old vintage machines?

Bob Beeman

Updated 2013-07-25 @ 16:15 EDT (UT-4)

Long ago, back in the late 1970s and 1980s "Home Computers" started to become available. These were things like the Comodore 64, TI-99, etc. Naturally, as engineers my friends and I wanted to compare the performance of these machines over time and compared to IBM mainframes.

You can run your own JavaScript Based Test in your browser on the linked page. Remember, different browsers will give different results even on the same hardware.

Admittedly these tests were simple-minded, for example they all run in Basic (an interpreted language), they don't measure memory or disk performance, etc. etc. Nevertheless we did what we could with what we had. The benchmarks we settled on were run in whatever version of "Floating-Point Basic" was available on the computer in question. We had two general test programs. One, "BML", (Bench Mark Loops) which simply measured how long it took the computer to do a given number of empty "for loops". The other, "BMA", put some arithmetic calculation into the loops to see how much this slowed the process down. The tests were identified with a trailing digit indicating the base 10 logarithm of the number of repetitions in the loop. For example test BML4 would be the BML test with 10,000 repetitions in the loop. BML6 would have 1,000,000 repetitions of the loop.

In one sense these benchmarks reflect reality. On compute-intensive logic or arithmetic tests my iMac is, in fact, stupendously faster than the IBM 370/3081 using TSO. I know this because the JavaScript-driven factoring page on my site can, in a fraction of a second, tell me that the first prime larger than 999999999999999 (10¹⁵-1) is 1000000000000037. The IBM system took a noticable time for much smaller primes (within the 23-bit mantissa of its floating point number representation), for example for finding the first prime greater than 8,000,000. The algorithms used are the same.

The identity of the tester and date of the measurement are indicated. Testers were:

RHB	Robert H Beeman
TF	Tim Funderburk
THG	Todd H Gardner
RFJ	Robert F Jurewicz
RAM	Richard A Morstadt
RJS	Robert J Sabon
LAT	Lloyd A Tarr

The procedure for the tests was to load the program and time with a stop watch how long was consumed by the program. In the case of the IBM mainframe we used the internal process time duration metric provided by the operating system.

The two tests were as follows:

BML*

FOR I=1 to 10*

NEXT I

END

BMA*

FOR I=1 to 10*

J=(6*(I-1)+9)/3

NEXT I

END

The results were as follows, in order of Loops/sec on the BMA test, the one that actually does some work inside the loop. Do not confuse the date of testing with the date of availability of the product. We tested these when we had a unit available, not necessarily when it was the "latest, greatest" thing.

Machine	Program	Time(sec)	Loops/sec	µsec/loop	Name / Date	Notes
iMac (2010)	BML9 BMA9	1.13 4.17	882,612,533 239,865,675	0.001 0.004	RHB 2013-07-24 RHB 2013-07-24	1
IBM 370 / 3081	BML6 BMA6	1.93 5.14	518,000 190,000	1.93 5.14	TF 1982-10-25 TF 1982-10-25	2
Intel 386 33 MHz	BML5 BMA4	40 13	2,500 769	400 1,300	RHB 1995-12-22 RHB 1995-12-22
BTI Site	BML5 BMA4	20 20.6	5,000 385	200 2,600	RHB 1982-10-25 RHB 1982-10-25
HP 9825 Desktop	BML5 BMA4	59 33	1,695 303	590 3,300	RFJ 1982-10-26 RFJ 1982-10-26
Macintosh Plus	BML5 BMA4	52 44	1,923 227	520 4,400	BRD 1987-11-19 BRD 1987-11-19	3
IBM Personal	BML4 BMA4	12 95	833 105	1,200 9,500	RFJ 1982-10-26 RFJ 1982-10-26
Apple II+	BML4 BMA3	13 11.5	769 87	1,300 11,500	LAT 1982-10-26 LAT 1982-10-26
ZX-81	BML4 BMA3	45 14	222 71	4,500 14,000	RJS 1982-10-25 RJS 1982-10-26
Atari 400	BML4 BMA3	23 19	435 53	2,300 19,000	TF 1982-10-27 TF 1982-10-27
TI-99	BML4 BMA3	28 23	357 43	2,800 23,000	RHB 1982-10-18 RAM 1982-10-26
TRS-80 Pocket	BML2 BMA2	23 54	4.35 1.85	230,000 540,000	THG 1982-10-27 THG 1982-10-27

Notes:

iMac: I don't have a version of Basic, so I used JavaScript in the Firefox v22 browser. The iMac is a 2010 model 2.66 GHz Intel Core 2 Duo with 4 GB 1067 MHz DDR3 RAM running OSX 10.6.8 Snow Leopard. Comparing timings done with Basic and JavaScript is tricky at best.
Time durations were measured by the internal JavaScript time clock in the browser which has a resolution of 1 ms. Loops per second and µsec/loop were also calculated by the script after the looping was completed. Safari and Chrome appear to be significantly slower than Firefox v22 on this machine.
IBM 370/3081: Yes, I know this is not a "Home" computer, but we had one available to us from our employer, Automatic Electric Co. in Northlake, IL. So in the spirit that "The only difference between men and boys is the price of their toys", we measured it with the same Basic programs as the other computers. The 370/3081 was running TSO (Time Sharing Option) with about 100 users logged on. The TSO "TIM" function was used, which shows elapsed (not CPU) time. Nevertheless, the numbers are reasonable given that the 370 was specified as having a 26 nsec cycle time (38.5 MHz) , compared to the iMac's 0.376 ns cycle time based on 2.66 GHz clock rate. Both units were dual processor. 2660 MHz / 38.5 MHz = 69.09. 882,612,553 / 518,000 = 1703 times faster for the iMac. 1703/69 = 24.7, so you could imagine that the 370/3081 had a 24.7 loading factor for the 100 TSO users, most of whom would be inactive at any given time.
In any case, these report the perceived power by a single user, which is what you have for all but the 370/3081.
Macintosh Plus: This system used a Motorola 68000 16-bit processor with a 7.5 MHz clock rate. The "Plus" was the same as the original Macintosh except for a NuBus port and 1 MB of RAM memory (later upgraded to 4 MB on this particular machine) instead of 256 KB, bringing it to the maximum architectural limit. No basic compiler was installed on the machine, so a HyperCard script was used. Probably a very pessimistic comparison, but this is what was available. Operating system was 3.2.
Ω