Mungo wrote on 2023-11-19, 12:10:

I haven't benchmarked 8 Mhz 8086 vs. 10 MHz 8088. How close are their speeds in practice?

I haven't benchmarked them either.
But I expect the difference 8088 vs. 8086 to be similar to V20 vs. V30 - Re: Adventures of a generic Turbo XT clone [updated original post with pics]

"The IBM PC had the memory throughput of an Atari 2600."
Source: https://retrocomputing.stackexchange.com/ques … -pc-to-4-77-mhz

PS: I've deleted my previous posting. It was too long and probably of no real use, except of angering some people.
I'll leave the quote, I've posted earlier, though, but without my personal opinion.
The link is about questions why the IBM PC had certain limitations.

Jo22 wrote on 2023-11-19, 23:04:

"The IBM PC had the memory throughput of an Atari 2600." Source: https://retrocomputing.stackexchange.com/ques … -pc-to-4-77-mhz […]
Show full quote

"The IBM PC had the memory throughput of an Atari 2600."
Source: https://retrocomputing.stackexchange.com/ques … -pc-to-4-77-mhz

PS: I've deleted my previous posting. It was too long and probably of no real use, except of angering some people.
I'll leave the quote, I've posted earlier, though, but without my personal opinion.
The link is about questions why the IBM PC had certain limitations.

The article ignores that there were 5,6 and 8mhz clockings of the 8088, some old IBMs ended up with -6 parts oddly

Did quite a bit of performance investigation around 8088/8086 hardware. The topic is very interesting to me.
My signature points to a dictionary. The first few links there contain data related to the subject.
Intention was to broaden a bit the information spectrum since it is very narrow given the class of the hardware and its age.

Hope to be able to add 1-2 more boards for completeness at some point later - the Juko NEST 8086 and another assembly based on Faraday chipset.
Anyhow, figured i should drop a line here for what it is.

So if 16bit bus makes the computer about 50-60 % faster, then that would mean that 10 MHz 8088 compares to about 6.5 MHz 8086?

rmay635703 wrote on 2023-11-20, 02:35:

Jo22 wrote on 2023-11-19, 23:04:

"The IBM PC had the memory throughput of an Atari 2600." Source: https://retrocomputing.stackexchange.com/ques … -pc-to-4-77-mhz […]
Show full quote

"The IBM PC had the memory throughput of an Atari 2600."
Source: https://retrocomputing.stackexchange.com/ques … -pc-to-4-77-mhz

PS: I've deleted my previous posting. It was too long and probably of no real use, except of angering some people.
I'll leave the quote, I've posted earlier, though, but without my personal opinion.
The link is about questions why the IBM PC had certain limitations.

The article ignores that there were 5,6 and 8mhz clockings of the 8088, some old IBMs ended up with -6 parts oddly

The odd part might come down to original 808x clock timing requirements, it needed an asymmetric clock timing pulse with particular rise times, like ___/"\___/"\___/"\__ kinda thing, so driving it off a simple 50:50 mark space oscillator, or not "conditioning" the rising edge of the pulses exactly to it's liking, tended to mean you needed to use a faster part than indicated by nominal clock frequency for the best stability. See original data sheets to appreciate the full weirdness of it.

Thank you, that's interesting! 🙂👍

From what I remember (vaguely) the NEC V20/30 had a different duty cycle, too.
So strictly speaking, it was running a little bit out of spec when installed on PC/XT motherboards (as an plug in replacement).

Edit: I can be wrong, but I think that intel clock generator IC was used on PC motherboard.
So dividing that 14 MHz* clock wasn't absolutely needed to get correct timings for the 808x, the intel IC could have been taking care of proper 808x timings if a different clock was used.

(* to my understanding, the ~14 MHz frequency was traditionally convenient, because a fraction of it can be used for the, um, "recovery" cycles of processors.
It had to do with most things somehow being tied to NTSC colour burst frequency of 3,57 MHz (~14 MHz is a multiple).
A few rare systems from Europe used 4,4 MHz, the PAL equivalent.)

Jo22 wrote on 2023-11-23, 15:52:

From what I remember (vaguely) the NEC V20/30 had a different duty cycle, too.
So strictly speaking, it was running a little bit out of spec when installed on PC/XT motherboards (as an plug in replacement).

No.
8088/8086 need 33% duty cycle.
V20/V30 support both 33% and 50%.

And of course, the most obvious way to produce 33% duty cycle is dividing the XTAL frequency by 3.

Grzyb wrote on 2023-11-23, 23:29:

No. 8088/8086 need 33% duty cycle. V20/V30 support both 33% and 50%. […]
Show full quote

Jo22 wrote on 2023-11-23, 15:52:

From what I remember (vaguely) the NEC V20/30 had a different duty cycle, too.
So strictly speaking, it was running a little bit out of spec when installed on PC/XT motherboards (as an plug in replacement).

No.
8088/8086 need 33% duty cycle.
V20/V30 support both 33% and 50%.

And of course, the most obvious way to produce 33% duty cycle is dividing the XTAL frequency by 3.

No. 50% is official. 33% works, but is not in spec.
According to this document, at least.

https://datasheet.octopart.com/UPD71011C-NEC- … heet-106570.pdf

Jo22 wrote on 2023-11-24, 05:21:

https://datasheet.octopart.com/UPD71011C-NEC- … heet-106570.pdf

Nothing about V20/V30 requirements there...

But - https://www.alldatasheet.com/datasheet-pdf/pd … C/UPD70108.html

Duty cycle would be tKKH/tCYK, right?

So, the minimum duty cycles - at the nominal CLK frequency - are:
- for the -5: 69/200 = 34.5%
- for the -8: 44/125 = 35.2%
- for the -10: 39/100 = 39%

Really?
V20 isn't a drop-in replacement for 8088, after all???

Edit:

Grzyb wrote on 2023-11-24, 05:59:

Jo22 wrote on 2023-11-24, 05:21:

https://datasheet.octopart.com/UPD71011C-NEC- … heet-106570.pdf

Nothing about V20/V30 requirements there...

Right. It's merely NEC's official V20/30 clock generator that features an 50% duty cycle.

Grzyb wrote on 2023-11-24, 05:59:

Really?
V20 isn't a drop-in replacement for 8088, after all???

I've never questioned that.
I merely said it's running a bit out of specification here (most things have quite some tolerances). 🤷‍♂️
Wikipedia mentions that 50% duty cycle, too.

Edit: Just double-checked.
The Olivetti M24 might need a modification, not sure.
Perhaps it will also run as is with a V30. 🤷‍♂️ M24 users may can answer that.

"The 5 MHz version of the processor can be used directly instead of the 8086.
In systems clocked at 8 megahertz, such as the Olivetti M24, the clock generator
and the crystal that determines the system clock may also need to be replaced because,
according to the specification, the V30 requires 50 percent instead of 30 percent duty cycle." (Machine translation)

Original article (1985): https://www.computerwoche.de/a/ibm-pc-tuning-von-nec,1171157

Englisch Translation

OK, note to self to examine every V20/V30 board I encounter, to see what's the duty cycle (same as CPU clock / XTAL frequency ratio, right?), and whether the CPU is underclocked.

Just noticed something.
None of the Sharp MZ series computers used an i8088.
They all used full versions of the processors, be it Z80, i8086 or iAPX 286.

https://en.wikipedia.org/wiki/Sharp_MZ

That's what I meant to say, the i8088 was sort of a mistake (technically, not economically). But it wasn't the only one.

The i8080 and i8085 are inferior to the Z80 and Z80 compatibles/derivatives/supersets.
That's why the Intel developers who worked on the 8080 had founded Zilog and created the Z80, the processor the 8080 originally was meant to be.

Edit: That's why I believe that the NEC V20 was the best that ever happened to the IBM PC.
It compares to the 8088 like an Z80 compares to the i8080.

Both the Z80 and NEC V20 are much more elegant/intelligent in their design than the predecessors.
The V20/V30 even have the ability to replace the 8080 in its functionality (8080 mode).

Edit: I forgot to mention, some of the 16-Bit MZ models ran MS-DOS, too.
There's even a video, albeit about an 80286 model.
https://www.youtube.com/watch?v=wPKchL5bKDw

Edit: Or let me put it this way, I wonder if the IBM PC was designed to be weak on purpose.

When the IBM PC 5150 was released, IBM did well with its mini computer businesses etc.

If the IBM PC was any more powerful than it was, say by using an 8086 or 68000, it would have been a threat to the other business fields.

Ok, maybe I'm just imagining things, but the development of Windows was similarly affected.

Both Windows 2.x and OS/2 1.1 shared same visuals.

Same was with Windows 3 and OS/2 1.2 and 1.3, with the difference that OS/2 had used the new GUI earlier.

So why was Windows 3.0 so crude/depressing looking in comparison to OS/2 at the time?
I think that was intentionally, due to pressure from IBM.

IBM was afraid of Windows stealing the show, so to say.
So Windows 3 was allowed to be an upgrade over Windows 2, but simultaneously had to be looking inferior to OS/2.

Then, after the split up, Windows 3.0 MME and Windows 3.1 was allowed to be looking much more friendly.

Edit: Okay, so how does all of this relate to the thread's topic?
Well, I just wonder if the IBM PC truly deserves to be considered a reference for how to build a good x86 PC.
It's a nostalgic and historical relevant piece of hardware, but was it ever "leading edge"?

Edit: I hope I'm not making people mad by thinking out loud here, but I believe without this question being answered, it's hard to think about designing an (new) 8086 PC.

Edit: I forgot to mention, I'm a young XT owner, too.

I have a Siemens Nixdorf M35 8810 PC with a very large, old mainboard and a Hitachi CRTC for its internal CGA..

I've upgraded it with a V20, a second graphics card (Hercules) and really love that old behemoth. ^^

So it's not that I hate XTs whatsoever.. It's just that I find it interesting to get rid of bottlenecks.

A new, all 16-Bit PC wouldn't loose its PC personality, whatsoever.
All the XT style components would still be there, software would still see an XT.

If done with care, the original IBM PC BIOS can run mostly on modified on same hardware (V20 patch needed, maybe).

CGA cards could still be used, too.
Even more, a new 16-Bit CGA card could be designed, but dual-ported RAM.

The timings of such a PC would be above original XT timings, of course.

But maybe a "Turbo Button" functionality could be implemented. Waistates, second xtal oscillator etc, not sure.

Mungo wrote on 2023-11-17, 21:32:

There are all kinds of authentic remakes of retro computers, like C64, ZX Spectrum, MSX, Amiga. Some of them are full remakes, some less so. This made me think: why hasn't anyone made a retro PC hardware remake? It would be really nice to have a new 8086, 8 MHz PC with 640 KB of RAM and CGA or EGA graphics output. Floppies/HDD could be emulated with SD card or such and mouse/keyboard could be handled with USB. Adlib or Sound Blaster emulation would be available on the motherboard.

Has anyone even considered such a project?
I'm pretty sure retro computing scene would love a product like that.

I absolutotally agree!
It started.
Check out the Perixx Periboard USB keyboards.
Also, the SilverStone FLP-01 and FLP-02 cases...

It's starting...!!!

Jo22 wrote on 2023-11-28, 09:57:

Just noticed something. None of the Sharp MZ series computers used an i8088. They all used full versions of the processors, be i […]
Show full quote

Just noticed something.
None of the Sharp MZ series computers used an i8088.
They all used full versions of the processors, be it Z80, i8086 or iAPX 286.

https://en.wikipedia.org/wiki/Sharp_MZ

That's what I meant to say, the i8088 was sort of a mistake (technically, not economically). But it wasn't the only one.

The i8080 and i8085 are inferior to the Z80 and Z80 compatibles/derivatives/supersets.
That's why the Intel developers who worked on the 8080 had founded Zilog and created the Z80, the processor the 8080 originally was meant to be.

Edit: That's why I believe that the NEC V20 was the best that ever happened to the IBM PC.
It compares to the 8088 like an Z80 compares to the i8080.

Both the Z80 and NEC V20 are much more elegant/intelligent in their design than the predecessors.
The V20/V30 even have the ability to replace the 8080 in its functionality (8080 mode).

Edit: I forgot to mention, some of the 16-Bit MZ models ran MS-DOS, too.
There's even a video, albeit about an 80286 model.
https://www.youtube.com/watch?v=wPKchL5bKDw

Edit: Or let me put it this way, I wonder if the IBM PC was designed to be weak on purpose.

When the IBM PC 5150 was released, IBM did well with its mini computer businesses etc.

If the IBM PC was any more powerful than it was, say by using an 8086 or 68000, it would have been a threat to the other business fields.

Ok, maybe I'm just imagining things, but the development of Windows was similarly affected.

Both Windows 2.x and OS/2 1.1 shared same visuals.

Same was with Windows 3 and OS/2 1.2 and 1.3, with the difference that OS/2 had used the new GUI earlier.

So why was Windows 3.0 so crude/depressing looking in comparison to OS/2 at the time?
I think that was intentionally, due to pressure from IBM.

IBM was afraid of Windows stealing the show, so to say.
So Windows 3 was allowed to be an upgrade over Windows 2, but simultaneously had to be looking inferior to OS/2.

Then, after the split up, Windows 3.0 MME and Windows 3.1 was allowed to be looking much more friendly.

Edit: Okay, so how does all of this relate to the thread's topic?
Well, I just wonder if the IBM PC truly deserves to be considered a reference for how to build a good x86 PC.
It's a nostalgic and historical relevant piece of hardware, but was it ever "leading edge"?

Edit: I hope I'm not making people mad by thinking out loud here, but I believe without this question being answered, it's hard to think about designing an (new) 8086 PC.

Edit: I forgot to mention, I'm a young XT owner, too.

I have a Siemens Nixdorf M35 8810 PC with a very large, old mainboard and a Hitachi CRTC for its internal CGA..

I've upgraded it with a V20, a second graphics card (Hercules) and really love that old behemoth. ^^

So it's not that I hate XTs whatsoever.. It's just that I find it interesting to get rid of bottlenecks.

A new, all 16-Bit PC wouldn't loose its PC personality, whatsoever.
All the XT style components would still be there, software would still see an XT.

If done with care, the original IBM PC BIOS can run mostly on modified on same hardware (V20 patch needed, maybe).

CGA cards could still be used, too.
Even more, a new 16-Bit CGA card could be designed, but dual-ported RAM.

The timings of such a PC would be above original XT timings, of course.

But maybe a "Turbo Button" functionality could be implemented. Waistates, second xtal oscillator etc, not sure.

Intel in 1980 when the pc was being spec’d was having problems producing compliant 5mhz 8086 chips and for god knows what reason 8088’s were more likely to pass QC.

The net result was a significant price delta both on the cpu and its related components.

What’s amusing is that the price difference rapidly eroded to become irrelevant by late 83.

Next, in the real world many persist in stating an 8086 is only 5% faster than an 8088.

The reality of the performance gap is somewhat complex , many individuals wrote software as if they were writing for an 8080 using a lot of small writes and moves =<8 bit. More advanced Orthonganol /Tangential coding techniques and 16 bit writes didn’t happen as much as it should have leaving performance on the table until XT class systems were retired.

In a world of a 16 bit XT class system with all 16 bit slots / components maybe more advanced coding methods would have been used but if not the performance difference would be minimal moving nibbles on a 16 bit xt

rmay635703 wrote on 2025-06-07, 20:58:

Intel in 1980 when the pc was being spec’d was having problems producing compliant 5mhz 8086 chips and for god knows what reason 8088’s were more likely to pass QC.

Hi! In the beginning, intel was just a small semiconductor company, I think.
It's possible that manufacturing the i8086 caused a challenged to intel fabs.
Bigger or more advanced companies such as Motorola (maker of 68000) maybe didn't have so high failure rates for making 808x, not sure. 🤷‍♂️

rmay635703 wrote on 2025-06-07, 20:58:

The net result was a significant price delta both on the cpu and its related components.

What’s amusing is that the price difference rapidly eroded to become irrelevant by late 83.

I think that was thanks to second sourcing, too.
https://www.cpu-collection.de/?tn=0&l0=cl&l1=8086/88

I mean, if even Siemens and some Fench company (MHS) were able to produce 8086/8088 chips.. 😉

rmay635703 wrote on 2025-06-07, 20:58:

Next, in the real world many persist in stating an 8086 is only 5% faster than an 8088.

I think that's in parts because they both had a poor performing bus unit.

The 8086 did at least allow memory to be interfaced in full 16-Bit,
which in a 16-Bit bus system helped the other peripherals to not to have to fight a bottleneck.

While it didn't help the i8086 so much, a NEC V30 did later benefit from full 16-Bit memory path, at very least.
So many PCs got to see their true performance "post mortem", so to say.

Also, at least one of the i8086 designers called the 8088 an "castrated" chip.
So it really wasn't beloved by the iAPX design team, probably.
https://www.pcworld.com/article/146957/article.html?page=2

rmay635703 wrote on 2025-06-07, 20:58:

The reality of the performance gap is somewhat complex , many individuals wrote software as if they were writing for an 8080 using a lot of small writes and moves =<8 bit. More advanced Orthonganol /Tangential coding techniques and 16 bit writes didn’t happen as much as it should have leaving performance on the table until XT class systems were retired.

That's why the Z80 was > than i8080, I think.
It had an 4-Bit ALU, could merge two 8-Bit registers into an 16-Bit register. So crazy.
Turbo Pascal for CP/M-80 took advantage of this, I think.

If the i8086 had been focusing on building upon Z80 architecture instead of i8080, then the i8086 would have seen more 16-Bit use early on.
Simply because source code (or object code) from Z80 instead i8080 would have been the role model for porting to x86 in first place.

rmay635703 wrote on 2025-06-07, 20:58:

In a world of a 16 bit XT class system with all 16 bit slots / components maybe more advanced coding methods would have been used but if not the performance difference would be minimal moving nibbles on a 16 bit xt

The other 16-Bit PCs such as Olivetti M24/AT&T 63000 and PC1512/PC1640 were much quicker with an 16-Bit architecture (and twice the clock rate).
Merely video speed was poor, thanks to 8-Bit MC6845 (that slug of a CGA chip).
Substituting it through a modern microcontroller would have been better, maybe.

Some random thoughts that come to mind:

The 80186 was used in Tandy 2000, which was one of the fastest x86 PCs of its time.
The Tandy 2000 was notable for being used as a development computer for MS Windows 1.x..

The Siemens PC-X running Unix/Sinix (and PC-D running DOS) was an 80186 system, too.

For embedded purpose, the i80186 (SB60168?) was so popular that production lasted until mid-late 2000s (1982 to 2007).
In retrospect, the 80188/80186 was perhaps most popular x86 chip ever.

The use of normal MS-DOS compilers and development systems made them so popular, I think.
It's really good that the extended 80286 Real-Mode instructions set had made it into 8018x and NEC chips, too.

Edit: The Am186 ED-33KC is interesting. Supports up to 40 MHz and both 8/16-Bit bus operation.
https://www.cpu-galerie.de/html/amdam186-186ed33kc.html

Jo22 wrote:

That's why the Z80 was > than i8080, I think.
It had an 4-Bit ALU, could merge two 8-Bit registers into an 16-Bit register. So crazy.

And then there's the Zilog Z8000, which supported combining all of the general-purpose registers into a set of four 64-bit registers... in 1979. x86-64 wouldn't come for another 20 years!

Merely video speed was poor, thanks to 8-Bit MC6845 (that slug of a CGA chip)

That's the CRT controller, which generates sync pulses and pixel addresses for the display. It has no impact on a video card's performance.

The Am186 ED-33KC is interesting. Supports up to 40 MHz and both 8/16-Bit bus operation.
https://www.cpu-galerie.de/html/amdam186-186ed33kc.html

Indeed! I didn't think AMD made an 80186 clone.

GemCookie wrote on 2025-06-08, 09:30:

Jo22 wrote:

That's why the Z80 was > than i8080, I think.
It had an 4-Bit ALU, could merge two 8-Bit registers into an 16-Bit register. So crazy.

And then there's the Zilog Z8000, which supported combining all of the general-purpose registers into a set of four 64-bit registers... in 1979. x86-64 wouldn't come for another 20 years!

The Z8000! Yay! ^^ Yes, it was remarkable. But the Z80 with its popularity was perhaps its greatest rival, I guess.

GemCookie wrote on 2025-06-08, 09:30:

Merely video speed was poor, thanks to 8-Bit MC6845 (that slug of a CGA chip)

That's the CRT controller, which generates sync pulses and pixel addresses for the display. It has no impact on a video card's performance.

You're right, my bad. I was more thinking of it being an 8-Bit device here.
If it was an 16-Bit device, then the bus wouldn't need to switch back and forth between 8/16-Bit mode so often.
Both the address generation part and pixel writing could have been done in "one rush" in 16-Bit, maybe.
Things like wait states, recovery times etc. wouldn't have caused a slowdown, maybe.

GemCookie wrote on 2025-06-08, 09:30:

The Am186 ED-33KC is interesting. Supports up to 40 MHz and both 8/16-Bit bus operation.
https://www.cpu-galerie.de/html/amdam186-186ed33kc.html

Indeed! I didn't think AMD made an 80186 clone.

Hi, it was used in some network routers, too, I think.