bobm@acsdev.uucp (Bob Marrow) (12/19/89)
I'm thinking of putting together a Unix system around one of the new cheap 386SX boards. Does anyone have actual measurements of the speed of these boards running 32-bit protected mode software, as compared with the 386DX? I have seen a lot of MS-DOS timings (semi-worthless, since they tend to be of 16-bit 286 programs) and speculative flames that the SX is sure to be much slower because of its 16-bit memory bus (also worthless, I'm looking for hard numbers). I remember seeing a reference to an article on this subject in MIPS magazine a few months ago but have not been able to get my hands on a copy. If someone has it, can s/he e-mail me a summary of the results or post them to the net? Thanks much.
mike@antel.uucp (Michael Borza) (12/21/89)
In article <BOBM.89Dec18222655@acsdev.uucp> bobm@acsdev.uucp (Bob Marrow) writes: >I'm thinking of putting together a Unix system around one >of the new cheap 386SX boards. Does anyone have actual >measurements of the speed of these boards running 32-bit >protected mode software, as compared with the 386DX? Noooooo.... don't do it!! Spend the extra couple'a'hundred bucks for a real 386. I thought about doing the same thing, so I borrowed a 386SX with 387SX from my hardware vendor and tried out ISC 386/ix 2.0.1 dev. system with X Windows on it and compared it against my 386 systems. The 386SX had 2 MB of memory on the mother board and 2 MB on a 16-bit expansion slot board, and had a WD-1003 controller running a 28 ms ST-506 disk. I compared it against two 32-bit 386 systems: the first a 25 MHz 386 with 387, 10 MB of 32-bit memory, and ESDI disk running ISC 386/ix 2.0.1; the second is a 20 MHz 386 with 387, 4 MB of 32-bit memory, and ESDI disk running 386/ix 1.0.6. You can logically argue that this comparison are unfair, given the differences in disk subsystems and memory architecture; however, the test was valid for me, since I was trying to decide whether I could live with the performance of the SX, compared to the machines I use every day at work. The results below are for two tests: the first is a make of `flex', which I consider representative of the kinds of software engineering stuff I do. The second was for a small floating point test (lots of exps, ln's,...). Times are given in minutes:seconds, rounded to the nearest second. 16 MHz 386SX 25 MHz 386 20 MHz 386 ISC 2.0.1 ISC 2.0.1 ISC 1.0.6 real 11:37 1:56 2:25 user 10:07 1:20 1:18 sys 0:37 0:11 0:13 real 0:50 0:08 0:15 user 0:43 0:07 0:08 sys 0:02 0:00 0:01 Just based on memory bus bandwidth and clock speed considerations we expect the SX to be 3.125 times slower than the 25 MHz 386, and 2.5 times slower than the 20 MHz 386. If you multiply out the sys times quoted above, you'll get roughly these ratios. The difference in the remaining times is due to disk performance and the effects of the expansion bus memory card. Subjectively, the system felt about the same speed as my old 10 MHz AT running Microport SV/AT UNIX. X-Windows performance was terrible. I originally tried it out without a 387SX, but watching the screen repaint was like watching paint dry. With the 387SX, performance was much better, but in the long run, it would have slowly driven my nuts to keep using this turkey. I strongly suggest you try to borrow one of these to try out before you buy. For me, the performance hit just wasn't worth the paltry savings. mike borza. -- Michael Borza Antel Optronics Inc. (416)335-5507 3325B Mainway, Burlington, Ont., Canada L7M 1A6 work: mike@antel.UUCP or uunet!utai!utgpu!maccs!antel!mike home: mike@boopsy.UUCP or uunet!utai!utgpu!maccs!boopsy!mike