luckybob wrote on 2021-07-12, 01:54:

I'd like to be the voice of reason here, and tell everyone; NONE OF THIS FUCKING MATTERS. […]
Show full quote

I'd like to be the voice of reason here, and tell everyone; NONE OF THIS FUCKING MATTERS.

Bickering like children about whats "faster" as if it fucking matters. *Maybe* it did in 1990, but not now when we are all spoiled for choice on hardware.

Jesus tapdancing Christ.

Just build what makes you happy.

THANK YOU.
Be but these are the systems some of us actually used back then..Have a 3 in the CPUs description doesn.t nessarily mean a far superior than a CPU with a 2 in its. There a a lot factors involved and it is well known that when 386sx cpus process 16-bit instructions they use approx twice the cpu cycles as equivalent speed 286 cpu. We wouldn't want folk new to the hobby thinking otherwise would we ? ...😉

Copmuters are not just about gamez after all...😉

Here's a snippet from the Australian Computer Magazine Tech Tips , April '93, Q & A section at the time.

It was by far the best computer related magazine I have read and pulled no punches with respect to tech performance...😉

Caluser2000 wrote on 2021-07-12, 02:07:

There a a lot factors involved and it is well known that when 386sx cpus process 16-bit instructions they use approx twice the cpu cycles as equivalent speed 286 cpu.

That's simply untrue, and my testing supports this statement.

I share your enthusiasm for the 286, it's my favorite PC platform in fact, but I can't abide blatant misinformation

maxtherabbit wrote on 2021-07-12, 02:55:

Caluser2000 wrote on 2021-07-12, 02:07:

There a a lot factors involved and it is well known that when 386sx cpus process 16-bit instructions they use approx twice the cpu cycles as equivalent speed 286 cpu.

That's simply untrue, and my testing supports this statement.

I share your enthusiasm for the 286, it's my favorite PC platform in fact, but I can't abide blatant misinformation

That's fine. I've in ssue with your view point at all. It would apeair some in the industry back then supported my view point..😉

Seemed to me that in those days computer journos were jobless regular journos who if they'd used wordstar or something had enough PC experience to get hired. At least a bunch of them would spell wait states, waite states, bus, buss, and cache with an accent like they were trying to say cafe, or cachet. Worst of all, they'd believe the press releases.

Edit: BTW back in the day (94ish) I had to read this through about 10 times when I first found it, to undo all the "kept up with the PC press" smarts I thought I had... http://www.faqs.org/faqs/pc-hardware-faq/part1/

Another snippet to look at...😉

http://www.redhill.net.au/c/c-2.html

Well what do you know. A vogons member has done 28/256 vs 386SX25 vs 386DX25 comparison Performance comparison of CPU: 286-25 vs 386DX-25 vs 386SX-25

From an 2017 USENET post:

"The only time an i386SX really could outperform an equivalent clocked
i286 was running 16-bit code that did a lot of multiword (32-bit or
wider) computation, or code that did a whole lot of segment switching
[which was significantly faster on the i386 than on the i286].

But most people would not have discovered this - apart from some
enormous WYSIWYG page design apps, there just weren't many 16-bit
programs that would run faster on a i386SX unless the clock rate also
was significantly faster.

And the i386SX was abysmal when running 32-bit code ... on average
only ~40% the performance of a DX at the same clock speed.

George"

Actually the whole thread is worth a read= https://www.electronics-related.com/showthrea … 1-chronological

null

Anders- wrote on 2021-07-12, 11:54:

For all practical matters a 386SX is faster or equal to a 286.

I'd say maybe equal but not faster, especially with decent mobo. The 386 core does regress with some - rather useful - instructions, possibly to simplify the silicon during transition to 32-bit and higher clocks. But then there were some very common instruction cases like PUSH, POP, or any ALU op with GPR,imm arguments and some memory ones too that were faster on 386. Depending on the actual code it can swing both ways, usually in 286 favour with 8088/8086 code, but nothing even close to 2x.

In the Performance comparison of CPU: 286-25 vs 386DX-25 vs 386SX-25 thread linked above, unless I somehow misunderstood it, the 386SX system at 25MHz ever so slightly edged over the 286 one in actual applications. Pure synthetic benchmarks on memory (and possibly I/O) the 286 handily won due to regression of anything string-related (LODS, STOS, MOVS, including REP), and that can manifest in games too but not to such extent. 16-bit '86 code is badly register starved and the common use of the PUSH/POP and immediate arguments in instructions does help the 386 to regain some ground.
Not sure what the deal is with "386SX-25 + overclocked" setup, clearly it scores about half of the non-overclocked one and is unstable. That is what I said about badly misconfigured 386SX systems, and/or really poor mobo performance.

I also contest the performance loss of 386SX vs 386DX. We are talking 16-bit code here, and in that scenario it's only some 20% loss, not 60%. In pure 32-bit code, with tons of optimized block memory operations, yeah, it would be way worse. But as we are discussing 286 vs 386SX lets not brind 386DX and 32-bit code into that.

As for various articles back in the day, there were a lot of distributors with 286 machines that wanted to sell them before the 386 fully took over. I wouldn't be very surprised if the arguments in favour of 286 speed vere, shall we say, a bit cherry picked. And possibly paid for as well.

null

And I did misunderstand the posted benchmarks, the first case is 386DX vs 286. So I went over it again in disbelief, how the hell is 286 matching cached 386DX and even being way faster in synthethics. Well, there is a BIOS screenshot attached fortunately. What can we see there?

slowest 3-x-x-x cache read timings
+1WS on cache write for good measure
+1WS on DRAM, at merely 25MHz...

Sure, why don't we tie one competitor hands behind his back and call it an even match. I'm not sure how bad is that 386SX mobo but "Memory performance - Fastest" does not sound like it's "overclocked" and the 286 having heatsinks on DRAMs says a lot about that mobo performance.

There's something to keep in mind when comparying 286 vs 386.
The bus interface.

The ISA bus, and the earlier AT bus, were based off the 80286's front-side bus.
A 286 CPU thus can theoretically interface directly with ISA.

A 386 system may or may not require some additional glue logic to be able to do so.
This can cause a small performance drop.

Jo22 wrote on 2021-07-12, 14:15:

The ISA bus, and the earlier AT bus, were based off the 80286's front-side bus.
A 286 CPU thus can theoretically interface directly with ISA.

There must be something of the sort going on with that 286 mobo, Landmark 6.0 CPU results are actually pretty sane for the 386SX vs 286, same for NSSI. Not grat but not 2x as I've said. I have ALI chipset mobos like the one used in the test and I know it can go down to 0WS on RAM (with fast enough sticks) and also supports memory banking if all 4 slots are populated. It can do better.

But VGA (or perhaps ISA in general) performance on 386SX mobo is merely half of what 286 can do. That will obviously completly destroy any game or graphical benchmark performance. Frankly it's also not great (slower than 286) on the DX system so apparently that particular 286 mobo is just very fast in I/O. It has little to do with the CPU performance itself, it's all down to that particular motherboard.

Since they used totally different hardware in those tests I disregard the results. Again I personally tested 386sx vs 286 at identical clocks on identical configurations. The 286 IS faster but only by 3-5%

maxtherabbit wrote on 2021-07-12, 14:46:

Since they used totally different hardware in those tests I disregard the results. Again I personally tested 386sx vs 286 at identical clocks on identical configurations. The 286 IS faster but only by 3-5%

Good point. Some instructions were performed with a different number of cycles on each CPU (386/286).
I think I've got a list, even..

maxtherabbit wrote on 2021-07-12, 14:46:

Since they used totally different hardware in those tests I disregard the results. Again I personally tested 386sx vs 286 at identical clocks on identical configurations. The 286 IS faster but only by 3-5%

Perhaps an aside but I had a late 92/93 ERA small factor desktop Harris 286-25mhz, 4mb ram (72pin Simms), 100mb hd, IDE, floppy, serial, SVGA onboard
From a Kwik Trip used by the manager for record keeping

By my testing (2004 era) it was far above your example systems in Phil’s benchmark at its stock settings and I have the strong belief that there were likely no 386sx-25s available in that configuration or even similar , sadly I only ran tests out of curiosity so it’s not photo doc’d and I didn’t even have a video recording device then but I would argue that the few late 286 systems likely are uniquely above the sx25’s even made at the same point in time .

It’s unfortunate that system was tossed when it quit posting, the sx40 it’s “guts” went into was far slower at simply booting into Windows and was quite possibly just an overclocked sx16 but that system was scrapped too so I will never know why it was such a turd at stock settings.

My try with homemade 386SX upgrade with reference from VTI /VLSI Databook 1989

It show lifesign on IBM 5170 board. Check on other motherboard come soon.

Awesome. Do you have a separate thread for the development of this?

Well done! ^^