[SOLVED] Anyone with a 386 willing to satisfy a curiosity for me?

Tatwi · 2021-07-13 01:07:15

I've been puttering with QBasic 1.1 (comes with DOS 5+) over the last year and to make a long story short, I'm looking for someone to run the following program on a computer that has real 386DX 33MHz CPU, so that I can compare the results to how it runs in DOSBox.

The program is here,

https://github.com/Tatwi/QBasic/blob/ma … ENCHES.BAS

and the description is here,

https://github.com/Tatwi/QBasic/tree/master/BENCHES

Part of the long story is included in the description, if you're interested.

Other CPUs I'd like to get data for are,
- Intel 386DX 16MHz
- AMD 386 40MHz
- Intel 486SX 25MHz
- Intel 486DX 33MHz
- Any Cyrix/IBM CPU in this age/speed range.

Why?

OCD and I can't afford to buy an old PC just to answer this question, basically. But mostly OCD.

Thanks!

zapper · 2021-07-13 06:50:05

Tatwi wrote:

I've been puttering with QBasic 1.1 (comes with DOS 5+) over the last year and to make a long story short, I'm looking for someone to run the following program on a computer that has real 386DX 33MHz CPU, so that I can compare the results to how it runs in DOSBox.
The program is here,
https://github.com/Tatwi/QBasic/blob/ma … ENCHES.BAS
and the description is here,
https://github.com/Tatwi/QBasic/tree/master/BENCHES
Part of the long story is included in the description, if you're interested.
Other CPUs I'd like to get data for are,
- Intel 386DX 16MHz
- AMD 386 40MHz
- Intel 486SX 25MHz
- Intel 486DX 33MHz
- Any Cyrix/IBM CPU in this age/speed range.
Why?
OCD and I can't afford to buy an old PC just to answer this question, basically. But mostly OCD.
Thanks!

I probably could, if I had a device that old, but I don't really know if I do... though. So... yeah...

Someone may be able to answer you, but I don't know...

Although if all else fails, try dosbox-x maybe? idk...

steve_v · 2021-07-13 07:00:50

No 386 unfortunately, the oldest operational machine I have is an IBM PC 330 100-DX4 (AKA 6571-W5L).
Intel 486 DX4-100, 256k L2 (WT), 48MB 70ns FPM RAM, CL-GD5430 video (via VLB-PCI bridge).
I have a DX2-66 CPU I could drop in it for some extra slowness, and I could probably try underclocking it, if the IBM CrippleBIOS(TM) will let me of course.

It's running the latest FreeDOS, which I'm not overly keen to uproot, and the drive is a) FAT32 formatted, and b) a complete bastard to get at, so if you want such a data point you'll have to point me at a source for this QBasic whatsit. I was a Pascal nerd back when, so I never had need or desire to go near it.

Last edited by steve_v (2021-07-13 07:17:40)

Tatwi · 2021-07-14 02:09:08

zapper wrote:

Although if all else fails, try dosbox-x maybe? idk...

The idea here in this post is to get results using real hardware so there is an accurate baseline for performance characteristics.

I do have results from various emulators, but these results show that the emulators aren't entirely accurate (PCem Pentium 233MMX is 30% faster than my real Pentium 233MMX machine), so I'd like to get some accurate results from real hardware. Incidentally, DOSBox and DOSBox-X performance is identical when set to the same cycles, regardless of any core/cpu related setting available in DOSBox-x, so it's not worth the hassle of compiling/using DOSBox-x for this use case.

steve_v wrote:

No 386 unfortunately, the oldest operational machine I have is an IBM PC 330 100-DX4 (AKA 6571-W5L).
Intel 486 DX4-100, 256k L2 (WT), 48MB 70ns FPM RAM, CL-GD5430 video (via VLB-PCI bridge).
I have a DX2-66 CPU I could drop in it for some extra slowness, and I could probably try underclocking it, if the IBM CrippleBIOS(TM) will let me of course.
It's running the latest FreeDOS, which I'm not overly keen to uproot, and the drive is a) FAT32 formatted, and b) a complete bastard to get at, so if you want such a data point you'll have to point me at a source for this QBasic whatsit. I was a Pascal nerd back when, so I never had need or desire to go near it.

That's a nice machine! Technically my first PC was a hand-me-down 486DX4 100 clone that I used for a bit in 1998. Getting it running taught me how to remove a Master Boot Sector virus, shortly after learning what a boot sector was!

If you're able to down clock you 486 to 33MHz that would be great, but the results at 66MHz or 100MHz would be excellent as well, because I could compare them to my results from emulated systems (using PCem version 17). Also, from what I have seen through emulated system, a 486DX2/66 ends up being exactly twice as fast in the benchmark as a 486DX33, so halving the results on a real 486DX2/66 should be an accurate representation of a 486DX33.

As far as running the software goes, Qbasic is included in the DOS 6.22 boot disk available here,

https://www.allbootdisks.com/download/dos.html

So it's possible to you write that image to a floppy and boot your computer using it rather than FreeDOS. If you copy my BENCHES.BAS file onto the floppy disk, you can run it like so

qbasic /run benches.bas

The results can be saved to a text file, results.text. To exit Qbasic, press Alt to open the menu, then arrow down the File menu and select Exit.

Alternately, you could download QBasic 1.1 from qbasic.net here,

https://www.qbasic.net/en/qbasic-downlo … preter.htm

and add it to your FreeDOS installation, as it's 100% compatible with FreeDOS.

I sincerely appreciate your time and consideration! Please don't feel obligated to help, especially if it means futzing with your hardware to do so. My burning curiosity is not that important.

steve_v · 2021-07-14 05:33:04

Tatwi wrote:

results show that the emulators aren't entirely accurate (PCem Pentium 233MMX is 30% faster than my real Pentium 233MMX machine), so I'd like to get some accurate results from real hardware.

Those old boxes (particularly 486 and early Pentium) were extremely sensitive to motherboard chipset, bus speed, wait states, and cache and memory timings. IMO 30% variation isn't outside the realm of possibility, not knowing the exact system PCEM is targeting or the details of your benchmark.

Tatwi wrote:

That's a nice machine!

It sure is cute, and like most IBM hardware of the time it's built like a tank. Wish the BIOS had more (or more like _any_) configuration options though.
Documentation is pretty scarce too, and original software packages are apparently unobtainium as only the unicorn "no hard file" (HDD-less) models came with any install media.
Still, it can Doom, and it can Descent. Hell, it can even Fallout at a playable framerate, so that's a win.
One day I might even get around to seeing how well it can Warp, that being the only truly appropriate OS for an IBM 486 box and all.

Tatwi wrote:

Technically my first PC was a hand-me-down 486DX4 100 clone that I used for a bit in 1998. Getting it running taught me how to remove a Master Boot Sector virus, shortly after learning what a boot sector was!

Mine was a hand-me-down as well, a mighty 386 sx-16, with 4MB RAM, an 80MB HDD (what luxury!) and a hercules mono display.
It came with a copy of Borland Turbo Pascal, a couple of books... and an admonition to "not waste time with that awful language" (BASIC)
As there was exactly zero entertainment available that would run on a Hercules MGA, the only option was writing my own, which I did with rather... mixed results. I still kinda miss the awesome compilation speed of turbo pascal, it put my current 5GHz Gentoo desktop to shame.

Tatwi wrote:

If you're able to down clock you 486 to 33MHz that would be great, but the results at 66MHz or 100MHz would be excellent as well, because I could compare them to my results from emulated systems (using PCem version 17)...

I can certainly get 66 and 100, 33 will be a bit more interesting. TBH I suspect many of the jumpers on this board are just for show, and the BIOS secretly does whatever it feels like on any given day.

Tatwi wrote:

Please don't feel obligated to help, especially if it means futzing with your hardware to do so. My burning curiosity is not that important.

Futzing with the hardware isn't a problem, that's kinda what the box is for. Just not today. Today is for installing Debian Etch, and yes, that will indeed take most of the day.

Now that I have a fully functional network stack I'll probably just FTP QBasic onto it, assuming running in FreeDOS won't bugger up the results.
I'll have a go sometime in the next couple of days, time permitting.

Last edited by steve_v (2021-07-14 06:08:41)

zapper · 2021-07-18 02:42:01

Tatwi wrote:

zapper wrote:
Although if all else fails, try dosbox-x maybe? idk...
The idea here in this post is to get results using real hardware so there is an accurate baseline for performance characteristics.
I do have results from various emulators, but these results show that the emulators aren't entirely accurate (PCem Pentium 233MMX is 30% faster than my real Pentium 233MMX machine), so I'd like to get some accurate results from real hardware. Incidentally, DOSBox and DOSBox-X performance is identical when set to the same cycles, regardless of any core/cpu related setting available in DOSBox-x, so it's not worth the hassle of compiling/using DOSBox-x for this use case.
steve_v wrote:
No 386 unfortunately, the oldest operational machine I have is an IBM PC 330 100-DX4 (AKA 6571-W5L).
Intel 486 DX4-100, 256k L2 (WT), 48MB 70ns FPM RAM, CL-GD5430 video (via VLB-PCI bridge).
I have a DX2-66 CPU I could drop in it for some extra slowness, and I could probably try underclocking it, if the IBM CrippleBIOS(TM) will let me of course.
It's running the latest FreeDOS, which I'm not overly keen to uproot, and the drive is a) FAT32 formatted, and b) a complete bastard to get at, so if you want such a data point you'll have to point me at a source for this QBasic whatsit. I was a Pascal nerd back when, so I never had need or desire to go near it.
That's a nice machine! Technically my first PC was a hand-me-down 486DX4 100 clone that I used for a bit in 1998. Getting it running taught me how to remove a Master Boot Sector virus, shortly after learning what a boot sector was!
If you're able to down clock you 486 to 33MHz that would be great, but the results at 66MHz or 100MHz would be excellent as well, because I could compare them to my results from emulated systems (using PCem version 17). Also, from what I have seen through emulated system, a 486DX2/66 ends up being exactly twice as fast in the benchmark as a 486DX33, so halving the results on a real 486DX2/66 should be an accurate representation of a 486DX33.
As far as running the software goes, Qbasic is included in the DOS 6.22 boot disk available here,
https://www.allbootdisks.com/download/dos.html
So it's possible to you write that image to a floppy and boot your computer using it rather than FreeDOS. If you copy my BENCHES.BAS file onto the floppy disk, you can run it like so
qbasic /run benches.bas
The results can be saved to a text file, results.text. To exit Qbasic, press Alt to open the menu, then arrow down the File menu and select Exit.
Alternately, you could download QBasic 1.1 from qbasic.net here,
https://www.qbasic.net/en/qbasic-downlo … preter.htm
and add it to your FreeDOS installation, as it's 100% compatible with FreeDOS.
I sincerely appreciate your time and consideration! Please don't feel obligated to help, especially if it means futzing with your hardware to do so. My burning curiosity is not that important.

Point taken, you raised a good point, emulation would ruin the effect. heh...

I would expect dosbox to be faster though.

Last edited by zapper (2021-07-18 02:42:45)

steve_v · 2021-07-18 07:09:53

I haven't forgotten about you Tatwi, I've just been a little distracted fighting the newfangled udev initrd madness in Debian Etch.
Now that I've given up and moved on to something far more appropriate (Slackware ), yours is back on the agenda. Restoring DOS partition over FTP now, so it might even be... Soon(TM).

Last edited by steve_v (2021-07-18 07:11:07)

steve_v · 2021-07-19 07:46:55

Here ya go. As I suspected the multiplier jumpers on this weird-ass IBM board do nothing at all, so 66 & 100MHz (in that order below) it will have to be.

                                    Results
--------------------------------------------------------------------------------
       Test                                Seconds
--------------------------------------------------------------------------------
 Floating Point Math                 4.015625 

 Integer Math                        4.1171875 

 Math IF/ELSE Logic                  3.8984375 

 String IF/ELSE Logic                3.3046875 

 String Select Case Logic            3.234375 

 Integer Array Sort                  3.1328125 

 Seq. Print Rate                     .8828125 

 Seq. Character Fill Rate            .65625 

 Rnd. Character Fill Rate            1.1015625 

 Rnd. Character FR, Rnd. Color       1.53125 

 Seq. Pixel Fill Rate                1.2109375 

 Seq. Pixel FR, Rnd. Color           6.703125 
--------------------------------------------------------------------------------

Total Time: 33.7890625

                                    Results
--------------------------------------------------------------------------------
       Test                                Seconds
--------------------------------------------------------------------------------
 Floating Point Math                 2.90625 

 Integer Math                        2.96875 

 Math IF/ELSE Logic                  2.6953125 

 String IF/ELSE Logic                2.3046875 

 String Select Case Logic            2.3046875 

 Integer Array Sort                  2.140625 

 Seq. Print Rate                     .765625 

 Seq. Character Fill Rate            .5 

 Rnd. Character Fill Rate            .7734375 

 Rnd. Character FR, Rnd. Color       .9765625 

 Seq. Pixel Fill Rate                .875 

 Seq. Pixel FR, Rnd. Color           4.671875 
--------------------------------------------------------------------------------

Total Time: 23.8828125

Not quite what I'd expect given the difference in clock, but it is what it is.

Tatwi · 2021-07-27 15:04:55

Oh you rock, steve_v! Thank you so much for taking the time to run those test. I sincerely appreciate it.

I added the results to the spreadsheet and I used it to make this comparison between real and virtual CPUs. I think it's kind of interesting, especially the math and logic related test that aren't potentially affected by a difference in graphics cards.

The virtual CPUs are faster than their real counterparts, so that's a good thing to know as it becomes harder to find functioning old hardware. Personally I have three old machines of the era that are just e-waste now, broken and dead for various reason despite my best efforts to revive them.

I just don't have the money to throw at a replacement, given the crazy prices for this stuff now. Back in 2003 I paid a whole $25 for my Deskpro 4000 desktop and something around $20 for my now-dead Compaq LTE/25 laptop (486SX 25MHz). My other two dead computers were free from the computer shop I worked at in 2005. Back then this stuff was worthless junk - my boss had a literal barn full of old computers and parts! Now though, you're looking at $500+ CAD for a verified working system that isn't beat to hell or otherwise abused. Heck, people want $150+ CAD for untested junk. It's kind of insane.

So yeah, I sincerely appreciate your time and effort!

Last edited by Tatwi (2021-07-27 16:21:17)

The officially official Devuan Forum!

#1 2021-07-13 01:07:15

[SOLVED] Anyone with a 386 willing to satisfy a curiosity for me?

#2 2021-07-13 06:50:05

Re: [SOLVED] Anyone with a 386 willing to satisfy a curiosity for me?

#3 2021-07-13 07:00:50

Re: [SOLVED] Anyone with a 386 willing to satisfy a curiosity for me?

#4 2021-07-14 02:09:08

Re: [SOLVED] Anyone with a 386 willing to satisfy a curiosity for me?

#5 2021-07-14 05:33:04

Re: [SOLVED] Anyone with a 386 willing to satisfy a curiosity for me?

#6 2021-07-18 02:42:01

Re: [SOLVED] Anyone with a 386 willing to satisfy a curiosity for me?

#7 2021-07-18 07:09:53

Re: [SOLVED] Anyone with a 386 willing to satisfy a curiosity for me?

#8 2021-07-19 07:46:55

Re: [SOLVED] Anyone with a 386 willing to satisfy a curiosity for me?

#9 2021-07-27 15:04:55

Re: [SOLVED] Anyone with a 386 willing to satisfy a curiosity for me?

Board footer