Awesome experiment! Yeah bottom of the barrel 386DX -- what a waste of money it would have been back then. I would be curious to see tests with a slightly newer 386 board but with that CPU... Wondering if that board is adding wait states to the RAM slowing things down considerably? Maybe it only has a 16 bit data path to the RAM? Interesting stuff none the less!
It's like most of these things, very early adopters paid over the odds and often found the tech wasn't really worth the huge volumes of cash they paid! The 386 doesn't really come in to its own until you break past 20MHz, then things start getting useful.
My only concern would be that 90s 386 boards tend to be hardwired to either 33 or 40MHz, or only have a selection between those two, and so those would, somewhat ironically, almost certainly use wait states. This would still be present if I replaced crystals or played with the clock generator, artificially slowing things down. I do think they'd go faster otherwise, as cache was more common by then and the chipsets had gained much better features in general, especially in areas such as memory access. The system doesn't start if I try with only two RAM sticks installed, suggesting it is 32-Bit, as intended, and that they weren't doing something weird with it like those weird dual socket boards in the 90s did. 386DX chips should always have a 32-Bit memory interface really, hence needing four 8-Bit SIMMs to run. You can get away with only two on an SX as this only has a 16-Bit data bus and a 24-Bit address bus like the 286. - Imagine if they'd kept using SIMMs into the Pentium era, you'd need eight of the damn things before the system would even start up.
@@sjogosPT Most Pentiums did, but my comment was referring to the older 30-Pin types. Rather amusingly, there _is_ a Pentium motherboard which uses those, the EDOM MP066; www.win3x.org/uh19/motherboard/show/8664
@@HighTreason610 amazing. Pentium could use simms 32, simms 72 and sdram. Pentium could use alot of things from pretty old ones (like simms 32) to even a pretty new system with socket7, AGP Bus and sdram. Pentium was launched in a middle of a computer revolution. Pretty nice!
Great video dude!! I am cracking here right now when you start to get angry about the landmark benchmark 😂😂... I dont use this software as well. 😅. Cheers, Peter
It's looking to me like at the time it would have been a much better purchase to get a 286 machine and either waited that out to a 486 or Pentium machine. I used to think the 286 was useless compared to the 386 based on what I've heard people say and seen demonstrated with at least 1 game, but this has made me appreciate it a lot more.
Makes me wonder how common the 16mhz 386 was though compared to the 33 and 40mhz versions. I feel like the 33mhz DX was a much more popular 386 machine where I come from, and with external cache was a massive step up compared to the 286 that I remember.
My dad is a designer and uses ACAD to this day. When I was a kid, 87 90's etc 386 sucked compared to a 286. This was mostly due to what appeared to be the speed of the 286 vs 386. 286 did more work clock for clock than 386 did. I find your video interesting from what I remember my dad saying back then. Cool you used ACAD for the benchmark. I just remember the 386 was considered kinda worthless since the 486 blew it away. I need to get my a$$ in gear and test stuff again.
33:54 yeah including the Tandy 2500 sx/20 or sx/25 good god that is a major pain in the rear end. good job on the test's. I love it when just cut loose on the dumb crap company's do for stupid reasons I all ways need a good laugh. Thanks form the states.
Same. Also what huge slots? The slots on the board are regular ISA, the drive bays take standard 5.25" drives, but they require mounting rails that I only have four of unless I want to make more.
"Unreal or 'protected' mode" made me chuckle, along with the usual irreverence, of course. It's true that the 386 was a chip without a cause for a few years from I've read in the press of the time. OS/2 was supposed to be the white knight, released well afterwards, but Windows 3 became the standard instead. Considering most games released up until 1990 still supported the AT, it took a long time for the 386 to get traction, used mostly for file serving like you said. My first PC was a 386SX in 1993. By that time the 486 was selling well. Kudos for the inclusion of AutoCAD - never seen an old version before.
AutoCAD, can't even begin to understand how people use things like that. I don't think this machine was ever used for anything like it either, the original hard drive lasted just long enough for me to see that it had WordPerfect, Windows/286 and not a lot else on there. Someone 30 years ago got gypped. Wouldn't have minded digging up whatever accounting software my old workplace used and testing that - no idea what it was called, they'd moved to Sage for everything else on later systems, but the company's own accounts were managed on their first system, a 286 (think it was 8MHz) from 1986 with no hard drive. Y'know, leaves a great impression of them, _given they were a fucking accountants company._ Was slow as hell. Guess who got the job of manually migrating that data when they decided it needed moving up to Sage on a then current machine? Long gone were the days when you could pay for a data conversion service out of the back of a newspaper or magazine, no, nobody would touch it, it had to be done by hand. Perhaps it's best I _don't_ remember the name of the software, because if I see that _stupid_ dark blue (on black) and bright green interface again, my brain is gonna go and I'm not going to be held responsible for my actions.
Great video, thanks for the analysis work with this one! Also, I appreciate the dry sarcasm and color commentary. I don't suppose you are from Yorkshire, by any chance ?
Using 86box these two are pretty much a tie as well. No good reason to emulate the 286. Makes more sense to emulate the 086, 386sx, 486dx, and the 586 skipping the 286 altogether. Awesome video by the way, very cool.
Hey, you should update the title of this video so that it gets more eyeballs -- the fact that you're comparing at same MHz is compelling. Also, clarify if it's a 386DX vs. 386SX (to be fair, 386sx is a more appropriate comparison). So, maybe something like "Who wins at 16 MHz: 80286 or 80386-DX?" Yes, clickbaity, but could definitely get the channel more traffic.
Could do, but I'm not _that_ bothered about more traffic, though I'll probably update the title to be more descriptive as it was done in a hurry as I was tired. The SX would have been brand new at this time, even evidenced by this DX lacking said DX moniker, and doesn't seem as though it had gained any traction yet. It would begin to displace the 286 in time of course. In what limited tests I and some others I know have done, they seem to average out marginally slower than the 286, albeit this being rapidly offset by the higher clocks they were available at. I think the 286 video showed the T3200SX to be slower than the 286, the only other SXs I have are double the clock speed, guess the crystal could be replaced to slow them down but I'd worry about hardwired wait states skewing the test as SX boards were usually cheaply made and didn't have many options to change things like that. Perhaps an earlier board would be a better candidate if I ever happen across one.
New to this channel so not quite sure what to make of it yet. Retro computer videos filmed in old dirty murder scenes? Strange combination. Anyway. My first pc was a 386 sx-20 I think (this was well into the 90s). Very happy with it, because I was a kid and was lucky to get any pc 😄
Yeah, I suppose it makes sense that a 16-bit system would be better at running 8-bit software than a 32-bit system... sort of like how 32-bit systems were a bit better at running 16-bit software. That might have been the 286's biggest advantage... it was closer to just being a souped-up 8088 that did more work clock-for-clock on the same software. The 386 was a brand-new architecture, it would be like buying an Athlon 64 and wondering where all the 64-bit software is, or why early dual or quad-core CPUs weren't able to take full advantage of all their cores, etc. Plus, if I recall my history correctly, the 8088 was really a gimped 8086 anyway, meaning that a lot of the software written for the 8088 could have potentially taken advantage of a 16-bit architecture, but would have had no clue how to take advantage of a 32-bit architecture. So even though the 16-bit architecture might seem like a pointless dead-end between 8-bit and 32-bit by today's standards, the key to understanding why they bothered with the 286 is realizing they started with what was really more like a 16-bit chip anyway, meaning the 286 (16-bit chip) is just a faster 8086 (16-bit chip), which is in turn just a faster 8088 (16-bit address space, limited to an 8-bit data bus). Though honestly, for my money? Back in those days, I think I would have probably been quite happy with a Compaq Deskpro running an 8086 at 8 MHz... that would have run pretty much all available software at an acceptable speed until we started seeing stuff like Windows and SimCity near the end of the 1980s.
Was this the same Tandon who Western Digital bought for their hard drive manufacturing facilities, what with WD originally doing controller cards rather than drives?
1:29 - That case brings back some memories, for sure. Around 1992 or 1993, I bought a 386 from a friend's cousin, and it used that exact same case. (I'm 95% sure it wasn't a 286 in there.) With a Sound Blaster 1, and "4D Sports Boxing" on it. lol It had this version of the theme tune, which was far better than the old theme, IMO... ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-PL4Rls0lT7M.html I also had a Tandon 386 for a while. Very similar case, but not quite the same. I don't think I had a CD-ROM drive until about 1994, which was originally a Dual-Speed Aztech drive. But it wasn't fast enough to play the FMV on games like Megarace smoothly, so I eventually got the Quad Speed version. I think the Quad Speed cost around £199 back then, which was quite a chunk. (I got it as a combined birthday / Christmas present, ofc, as I was about 13 then.) Great memories of playing Alone In The Dark, and the first time having proper Red Book audio on a PC.
I'm starting to relive some of those old memories with some retro PC projects. Mainly involving the MiSTer FPGA board though (yeah, yeah, I know. lol) I designed a PCI adapter for it, which is very close to working with a Voodoo 1 card. But then I discovered that the Voodoo 1 had apparently died since the last time I tested it about 5 years ago. Would love to get hold of an AWE32 again, but I don't have a machine with ISA atm, only much newer stuff like a Q6600. I bought a Pentium board about six months ago, and had a P200 and P233 MMX chip which I had for years. Not sure if those chips are supported on the board, though, and it's hard to find the proper manual for the jumper settings.
Suppose the generic case the 286 is in probably got used for all sorts of things, really, had reason to suspect this one originally had a 386SX or 486SX in it but don't know, only got the case and the rest of the hardware was long gone. It gave me 80s vibes, so seemed a good home for this system. The Mitsumi is a single speed drive, they wouldn't have been common at that time and it doesn't have much use in such a slow system, but at least nothing on said system generally cares for higher speed drives. Cutscene heavy games on later systems seem to always want a quad speed. Do you have a photo of this Pentium board? I might be able to find the jumper settings for it, or else they might be generic enough that I can figure them out by looking at it.
@@HighTreason610 Hi. Thanks for offering to have a look. When I first looked months ago, I found some manuals and jumper setttings that were close to the model number, but not quite close enough. The main difference were the jumper settings for the VRM I think? I tried setting it to the lowest value, by calculating it using the VRM datasheet, then gradually increasing it until it matched the voltage for the 233MMX. It still didn't boot (well, do the POST beep), and I'm not 100% sure about my PSU, nor whether the board definitely works or not, as it was an eBay purchase. If it's the CPU, then I have a 200MMX instead that I could try. I only recently got hold of a PCI gfx card, too, so I was relying on the PC speaker for diag before. I have an ISA diag card somewhere, but it's probably in the loft. lol The manual for the AB-PR5 is online, but it has a DIP switch block instead of jumpers. Here are some photos of my board... i.imgur.com/gCri2JC.jpg i.imgur.com/nL4duex.jpg i.imgur.com/ftNAbI7.jpg i.imgur.com/rMp2tAk.jpg Same issue as the poster here... www.vogons.org/viewtopic.php?t=52640
Actually, I have the SL27J / 2.8V P200 MMX in the socket atm. www.cpu-world.com/sspec/SL/SL27J.html And there are some jumper settings here that might get me closer... stason.org/TULARC/pc/motherboards/A/ABIT-COMPUTER-CORPORATION-Pentium-PR5-R2-PCI-REV-2.html
OK, I stopped being lazy, and did a few more measurements. lol... JP11 (VIO voltage) Pins 1+2 = 3.24V? (measured 3.240V on vreg Q5) Pins 3+4 = 3.38V? (measured 3.375V on vreg Q5) Pins 5+6 = 3.54V? (measured 3.540V on vreg Q5) JP12 (Vcore voltage) Pins 1+2 = 2.5V (measured 2.496V on vreg Q1) Pins 3+4 = 2.7V (measured 2.689V on vreg Q1) Pins 5+6 = 2.9V (measured 2.919V on vreg Q1) Pins 7+8 = 3.3V (measured 3.366V on vreg Q1) Pins 9+10 = 3.5V (measured 3.525V on vreg Q1) (vreg Q7 seems to just be for 5 Volts.) I haven't confirmed the clock speed from the DIP switch block yet, but it probably matches the manual for the other rev. I tested those voltages with the CPU removed from the socket, then set it back to 2.7V Vcore, and 3.38V VIO (probably OK if that's slightly higher than 3.3V.) But I was tested the board with 5V from a bench PSU, and it was current limiting at the max 3 Amps, so I'm gonna need more juice. And I know this gfx card (ATI Rage Pro 128) also needs the 12V rail to work. I'll probably modify an ATX PSU, as I think my AT one is faulty.
Could you use ELKS Linux on the 286 and and gray386linux so you can take two modern linuxes for these machines, and put them against each other? Would need to find something that would build with both the ia16-gcc and the GCC version for the 386… and would be able to equally benchmark them. But also using them for remote X if you can to access a modern computer would be cool, or tty is easier. Over tty chat with a local llama ai on your other new machine? Neato. I’m sure there are many other use cases here and I’m just brain malfunctioning
Eh, I prefer having an actual software library and boot times that are shorter than one week. It wouldn't really change anything benchmark-wise, either, nor would it really tell us anything if it somehow did. About the only Unix type thing I could envision running on something like this at the time would be Xenix. For networking this was really the era of NetWare (80-90% of the market by then) and occasionally, LAN Manager. As far as use cases, the 386 here was dismantled to keep other machines alive (which is why it was bought) while the 286 is largely used for entertainment & testing things.
I like this but isn't the whole explanation for all the benchmark results 32bit computing, as you suggest, right at the end? You may struggle to find an example of a 32bit app but, then again, back then, any 32bit app was likely to be hugely niche and probably industrial and for anybody who used, whatever apps fit the description, then it may be worthwhile for them? If you are using an 8088 compatible AutoCAD then it probably isn't using the 386 to any advantage, rather than there being no advantage to the 386. Testing again with any 32bit apps you could find might make a more revealing comparison.
I did state as much, as far as I remember. There just aren't really any 32-Bit applications to test, given they all came way later and usually require far more capable machines than these. They also can't be compared as they tend to use a radically different code base and, of course, won't run on 16-Bit processors. If we take AutoCAD as an example, support for the 386 didn't arrive until at least Q2 1990, where they _claim_ it was 50-62% faster, but then they would because they want to sell copies. They never once mention anywhere what spec this 386 was, either, so it could well be at some high clock with a bunch of cache added on. A top of the line 25MHz machine with 64K L1 cache will get you something like that, given the crippling memory performance detracts from what it might have otherwise achieved. The cost added by having that 64K SRAM is still gonna burn a hole in your pocket, though, and the motherboard itself will, as there aren't many QFPs that can handle those kind of clocks yet, meaning it's going to be built out of glue logic. NetWare also lacked 386 support until 1989/1990. Lotus didn't attain 32-Bit support until 1997. In the time of these 16MHz machines, it seems short of writing your own code (which some people certainly did) and running some obscure operating system, there's still no incentive to own a 386 over a 286 99% of the time. This particular machine was used as an office computer, with DOS 3 and WordPerfect. Someone definitely didn't get their money's worth. At best, software might have treated it as 'an 8088 with EMS', or a kind of 'weird 8088 that somehow has XMS', but that's about it. With so few 386s in the wild, nobody wanted to spend the time and money writing separate versions of their software for it yet. Too many 8088s left out there and still selling well. Short of getting into Unix/Xenix, which I can't stand, and writing something myself, there's just nothing to demonstrate or test.
Ah yes, those things. Don't think there were ever many of them out there as the defect seemed to be quickly corrected and 80s 386s are already quite limited in number. Somebody, somewhere, was almost certainly furious, however.
Ah yes, the "16 BIT S/W ONLY" defects for one thing. Suppose this would have been even more of a deal breaker earlier on, though this seems to have been ironed out by 1988.
I'm led to believe there are several such projects out there, even for other platforms - I know someone who is currently building an ATX C64. They're kinda cool, but not my cup of tea. I much prefer to have an ancient piece of trash in front of me with no documentation.
This would actually make sense as even the fence sitters were starting to move away from their aged minicomputers rapidly by this time. The Tandon here, though, was running DOS 3, WordPerfect and Windows/286 in its former role.
You are comparing in reality almost the same procesor 386 SX 16 MHz was in reality 16bit processor. :) Reason to buy one was the price and later to upgrade to 386 DX. However, the speed could be improved with more cache. If you want to compare 386 vs 286, get 386 DX 40 MHz. :)
Unfortunately you are misinformed. The 386SX is a 32-Bit processor. The 386DX-16 tested here certainly is. Cache was generally unavailable in this era and the cost of SRAM was comparable to an entire computer system, as I said. Even when the option was available, almost nobody ever installed it. There was also a small gap where Intel didn't make an 82385 at higher clock speeds and the then new integrated chipsets hadn't figured out how to clone it yet, rendering it completely unavailable on 20 and 25MHz systems for some time unless the board maker got creative - read; made things even more expensive and less reliable. Not sure why I would want to directly test any 16MHz machine against 386DX-40, an off-brand chip that released over half a decade later when both platforms were just about dead and buried. I don't understand what this would prove, especially as there was no 40MHz 286, barring perhaps an obscure SoC. By the time AMD released their 386-40, the 486 was around and would almost invariably flatten it. Hopefully this clears things up a little for you.
@@HighTreason610 hm, I was working that time as PC technician, I was informed by Intel guys, that 386SX is 16 bit one. Maybe they were wrong? From personal experiences, it was really slow CPU. Sometimes even the 286 on 20Mhz was faster. However, looking on wikipedia article now for 386 and subpart 386SX confirms information partialy- "Intel introduced the 80386SX, most often referred to as the 386SX, a cut-down version of the 80386 with a 16-bit data bus" . Internaly it was 32bit, however it was working with 16bit bus.
It's certainly a pretty common misunderstanding and I used to think the same thing. It's a weird thing, a 32-Bit CPU, 16-Bit data bus, 24-Bit address bus. Indeed, when running 16-Bit code it is usually outpaced by a 286, even on the same motherboards. When running 32-Bit code it suffers pretty badly, presumably in part having to resort to things like double transfers. Another thing worth keeping in mind from this era is RAM. It had lagged behind somewhat and was still stuck at around 120-100ns at best, meaning it couldn't handle anything past 8-10MHz. On 386 machines it was usually sitting behind 3-5 cycle Wait States, which only worsened as clocks went up until people found ways around it, namely faster DRAM, improved access patterns, the introduction of interleaving and a gradual drop in the price of SRAM for cache. That 1990s 12MHz 286 could often outrun this 1980s 16MHz one solely because it had 80ns RAM and thus, could run with no wait states enabled. You'd need 60ns RAM to do this at 16MHz and that wasn't gonna happen any time soon. On the 386SX, the RAM speed problem is exacerbated somewhat, especially if the CPU enters 32-Bit protected mode. Every dword transfer will functionally be two separate word transfers on the data bus, meaning you hit the wait state wall twice. Of course 99% of everything was still running in 16-Bit mode anyway, so the problem likely rarely came up until Windows/386 started gaining popularity.
I've also heard it called "Virtual Real Mode" before now. If you've ever opened a DOS window under Windows 3.1 or 9x, then the CPU has been running in that mode. If you start Win3.1 in "286 Mode" then you cannot use a DOS Window, or else not in that way and you'll know because it will instantly go full screen. Technically in "Unreal Mode" it's almost like a virtual machine, whereby the CPU is running in Protected Mode, but presents to the application as a CPU in real mode, allocates memory that can be accessed likewise and implements a full copy of all BIOS routines for the application to use. This mode wasn't available on the 286.
All the cursing feels out of place tbh, and a bit like pretending to be mad on purpose, just for the video. Come on, it's good even without those parts.
I'm guessing you've never been to Hull, huh? Every other word is a cuss word here. That and I feel bad for the networked channels who can't cuss in case they upset an advertiser and figured I'd better compensate for them. Nah, seriously though, I just say it however it comes out on the day. If you'd had to put up with the Tandon's shenanigans you'd be yelling too.