carlostex wrote:

> 286 is ~15% faster

It's not that the 286 is faster than the 386SX clock per clock. It really depends on the motherboard implementation.

My opinion follows from Speed Test's chart (when you run it, - it will show). If I'll get other results there, then I may change the opinion.

I do plan to take a personal look into the 286 vs 386sx story.

Would be good to have benchmark theme for 8086/88-386 systems similar to Phil's one wich is for 386 and higher. Besides Speed Test, may be used MIPS, Checkit, etc. And something taking into account video performance.

Well I can tell you my SlimsPORT 286 is faster than my T3200SX. I don't have another 8088 to compare the BT SuperSport with though.

On the other end of the scale, my 6MHz 286 is pretty slow, but it scores identically to an IBM 5170 in every test I tried, so it seems about right and that was why I built it in the first place.

The point i'm trying to make here is that even though the generational leap from the 286 to the 386 wasn't too big in terms of performance in reality the 386 CPU is always faster at the same clock than a 286. Problem is when 386SX systems started to appear chipsets were at that point very matured compared with early 386SX designs.

I'm not surprised that its easy to find 286 systems faster than 386SX systems. A good way to compare is to get a late 386SX board and downclok it accordingly to compare to late 286 systems. Then we'll get the whole picture.

In fact i do remember being very proud of my 386DX-40 back in the day because it was slightly faster than an equivalent system of a friend of mine.

Now as far as games for a good 12MHz 286, let's see:

- Apogee platformers;
- Late 80's racing simulators; (Grand Prix circuit, Indy500, hard drivin')
- Late 80's flight simulators; (F-15 SE II, F-19, LHX..)

And many more... 😵

carlostex wrote:

A good way to compare is to get a late 386SX board and downclok it accordingly to compare to late 286 systems. Then we'll get the whole picture.

Yes. To support your opinion about 386SX numbers are needed, as I've given this for my opinion.

Here's something i found, just to show the performance difference between some 286 boards:

https://translate.google.co.uk/translate?hl=e … int&prev=search

carlostex wrote:

just to show the performance difference between some 286 boards

Speed Test was made from 1994 year, so 386 MBs are late ones. If to approximate to 16 MHz results taken from 9 pcs 386SX CPUs there, and 6 pcs 286 CPUs (not PS/2), then you get averages: 29803 for 286, 25197 for 386SX. So from practical side late 386SX system is 15% slower than late 286 system.

I know "make it 486" upgrade chips existed, but what about "make it 386" ones? If so, those would be the ultimate way to compare the processors, as any 386 on 286 motherboards would be SX anyway (16-bit bus).

Tertz wrote:

Speed Test was made from 1994 year, so 386 MBs are late ones. If to approximate to 16 MHz results taken from 9 pcs 386SX CPUs there, and 6 pcs 286 CPUs (not PS/2), then you get averages: 29803 for 286, 25197 for 386SX. So from practical side late 386SX system is 15% slower than late 286 system.

You seem to have an obsession for this Speed Test utility, that benchmark alone is not gonna show the whole picture. Anyway i'll make some tests when i have some 386SX boards on my posession. But its good to see that we are talking about 286 vs 386SX systems and not 286 vs 386SX CPU's.

What i could indeed observe is that some 286 boards can be very fast while others can be much slower at the same clock speed. And i could also find as shown on that russian forum link that i'm not the only one to have observe this. All i'm trying to do is to raise awareness about not creating myths about 286 vs 386SX.

alexanrs wrote:

I know "make it 486" upgrade chips existed, but what about "make it 386" ones? If so, those would be the ultimate way to compare the processors, as any 386 on 286 motherboards would be SX anyway (16-bit bus).

Even "make it 486" upgrades often didn't show the true potential of Intel's 486 design. Despite carrying the 486 instruction set these upgrades would often have slower integrated cache and other quirks that were necessary to make it work in a previous generation socket. But i agree it would be at least interesting.

These games were what I used to play on my parent's (now mine) Unisys PW/2 Series 300.

Monkey Island 1.
Loom.
Civilization 1.
Indyana Jones And The Last Crusade.
Leisure Larry 1 + 2.
Stunts 3D (was that 4d?)
Stuntcar Racer.
Lemmings 1 and all the holliday editions.
Testdrive 1 and 2.
Ports Of Call.
Alley Cat.
Wings Of Fury.

And possible more, than these games on my list.
I used to play them on 8/10mhz with EGA.

Not my video but here's a very cool 286-16 build:

https://www.youtube.com/watch?v=S_ZEWvokp3o

BTW this is an example of a very fast 286 system. The performance figures are just a hair slower than my very similar Headland HT12 chipset based motherboard. My other 286 board the M219 is something like 40% slower. 😵

carlostex wrote:

You seem to have an obsession for this Speed Test utility

I have only this data for comparision between 286 and 386SX. Other tests would be good to see too, like the ones I've mentioned above + 3DBench.

i'll make some tests when i have some 386SX boards on my posession

As alternative - a benchmark theme for XT-386 systems. Some data may be gathered in such way.

All i'm trying to do is to raise awareness about not creating myths about 286 vs 386SX.

Based on what we have - 386SX systems generally work slower indeed. As for the reason - it's not a myth, as based on what we have there is stil a possibility for slower 386SX. And it's the most probable situation.

<--------

😁

80186 Systems are very rare, the 80186 is a MCU, not a CPU.
So there is some weird Logic required to use it in MS-Dos and IBM-PC combatible Systems.

Many 80186 are only MS-Dos Combatible but not IBM PC.

Example: Siemens PC-D, Custom Bus System, BIOS loaded from Floppy Disk, IBM PC Incombatible, runs Custom DOS 3.20/2.1 or 5.0.

Thanks for your recommendations and keep em' coming!

Games I tried so far which work great are:

- Bio Menace
- Commander Keen 1 trought 6
- Civiliations 1
- Dyna Blaster (played a little /w my fiancee - had a blast 😁)
- Dune 1
- Duke Nukem 1 and 2 (2 slows down a little sometimes)
- Drakkhen
- Gobliins 2 and Goblins 3
- Golden Axe
- Jetpack
- Kings Quest 4
- Lotus
- Lemmings 1 trough 3
- Monkey Island
- Prehistorik 1
- Prince of Persia
- Rampart
- SimCity (even at high res!)
- Space Quest 3 and 5
- Stunts
- Supaplex
- Volfield (you can kind of cheat if you push the turbo button at the right time 😁 )
- Wolf 3D (runs fine even with sound)

Games that run but are too slow to play:
- Dune 2 - too slow even at 16MHz. In my opinion this game needs a 40MHz 386DX to play well with sound enabled. Runs alright on my 33MHz 386DX with PC Speaker sounds.
- Jazz
- Lost Vikings - gets pretty slow in some stages while others run perfectly...
- Mario
- Prehistorik 2

The WD Paradise 1MB video card helps A LOT. I tried a slower OAK card and I had trouble with more graphically intensive games in VGA mode.

Quick question - is there a MemMaker - like program that will run on a 286? Memmaker for DOS 6.22 requires at least a 386 class CPU to run 🙁.

I'm trying to load CD-ROM drivers on the machine (I found a really small one - like 10-15kb) but with it loaded and SB drivers loaded some games don't have enough conventional memory to run (Space Quest series). So far I got around the problem by making a boot menu, but I'd like to optimize how drivers are loaded even further. I don't use EMM386 (obviously), only HIMEM.SYS, and I set all software and drivers with LOADHIGH and DEVICEHIGH so they will load in the "generous 😁" 384K of upper memory but I still need more conventional memory (at least 50kb more).

Tertz wrote:

Based on what we have - 386SX systems generally work slower indeed. As for the reason - it's not a myth, as based on what we have there is stil a possibility for slower 386SX. And it's the most probable situation.

That's actually interesting. I have a couple of 386SX boards with soldered 33Mhz AMD CPUs. These can be made to run at 16MHz and 8MHz by various jumper settings and use of the turbo button. Benching one against my 12MHz 286 overclocked to 16MHz might be interesting. If the 286 is faster clock per clock, we have our answer. So far all I know from testing my 386SX is that at 8MHz it is SLOW - so slow that PCConfig rates it very close to a 8MHz XT. Then again, pushing turbo might also enable wait states alongside the clock speed reduction.

Gods might run fast enough.

VIDE-CDD.SYS should take way less than 10kB, MSCDEX is the issue (though you can always try using SHSUCDX instead). Anyway... MEMMAKER isn't magic - it just rearranges the TSRs in memory (with the DEVICEHIGH /L:X and LH /L:X parameter), it does nothing you can't do on your own. Are your UMBs fragmented?

Secret Agent, Crystal Caves, Cosmo's cosmic adventure, Spear of Destiny, Battle Chess, The Blues Brothers, Haloween Harry in Alien Carnage, Catacomb 3D, Dangerous Dave, Skunny, Jill from the jungle, Chicago 95, Paganitzu, Settlers 2, just pick what you want, these are my most played games + that ones you writed.

alexanrs wrote:

Are your UMBs fragmented?

Don't even know how to check... most likely they are and that might be part of the problem. Memmaker would have sorted that out.

MEM /F /P if memory doesn't fail me.
This will give you an overview of available memory regions. The procedure then is as follows:

• By default LOADHIGH and DEVICEHIGH load stuff to the UMB block with the largest quantity of available memory, you need to override that
• Run MEM /C and take note of how large each driver/TSR is
• Edit AUTOEXEC.BAT and CONFIG.SYS and specify the region manually with "DEVICEHIGH /L:?=..." and "LH /L:? ...", where "?" is the region number. Beware some odd drivers like KEYB.COM and UNIVBE: they require large amounts of free memory when loading (UNIVBE is, in particular, horrendous about it - in some machines I can't load it high no matter what), and then they will relinquish a bunch of memory back to the OS, so one of these should be one of the first things you load on a smaller block.
• Reboot and use "MEM /C /P" and "MEM /F /P" to see how are the drivers distributed now. If a driver fails to load at the specified region it will fall back to the default behaviour.
• Change the specified block and order of loading drivers until you are satisfied

kanecvr wrote:

with it loaded and SB drivers loaded some games don't have enough conventional memory

SB .sys drivers may be not needed for many games.

I have a couple of 386SX boards with soldered 33Mhz AMD CPUs. These can be made to run at 16MHz and 8MHz by various jumper settings and use of the turbo button. Benching one against my 12MHz 286 overclocked to 16MHz might be interesting.

Concrete bench test may be important. Results in for example CheckIt may differ from Speed Test. Memory should be clean without TSR and drivers as they may affect benchs. Quantity of free RAM may affect. I'd also compare archiving times by rar/zip on some big text like Bible or several such files.
Systems may to have different speeds per 1 MHz: in CheckIt dhrystones 286 had range from 150 to 250, - 60%.

If the 286 is faster clock per clock, we have our answer.

The more systems and tests - the better. Benchmark theme would be good for XT-386.
To set systems to lower clock is not needed as what is interesting is to compare speed per 1 MHz in typical working mode. But as additional measurement may be done on lower clocks.