I have one of these with a piss-yellow PCB. I've had it in my parts box for more than 20 years, and only powered it on to see if it was working.
Your particular card might have been made in 1991, but I'm pretty sure the design is from 1989 or earlier.
Definitely would have been found in a fast 286 or 386SX. It doesn't offer much beyond basic VGA. 640x480 with 256 colours, and probably 800x600 but very slow since it's not accelerated.

Is this not one of the few cards that has SVGA drivers for early versions of Windows?

Actually this is a great card, but for 8088/8086 machines. I would even state "the best graphics card for XTs".
Here is mine: image

If you test it on 486 or later hardware it is unimpressive.
Evident from this benchmark with bunch of cards split by time period - early 16-bit ISA, late 16-bit ISA, VLB, and PCI. It checks all major players/chipsets at the time. Tests were performed on Asus PVI-486SP3 at 133/33MHz, so all cards can run.

Later on, during some Turbo XT fact finding activities, which resulted in building a 8086 hotrod rig, i realized that the story is slightly different from above.
This thread grew too big, it is hard to keep track of the different topics discussed, so here is a convenient link with details (somewhere in the middle there).

It was established early in the thread which are the best 72-pin FPM/EDO RAM modules for peak 486 overclocking. Results were validated by other forum members to hold true.
At the same time, despite the plenty of noise around 286 and 386 hardware, the RAM related details were mostly left untouched.
Was not able to find much in the pile of posts while looking for some specific information.
Running a wider search in the forum and online didn't go well either. So, decided to add some notes for the record.

This is roughly two-thirds of the 30-pin memory modules here. The other third is scattered across several motherboards. Didn't bother collecting it for the photo.
The picture is basically clarifying that a lot of RAM sticks were tested on a lot of different 286/386 motherboards over a lot of years before any conclusions were drawn.
The first 4 columns on the left are all 1Mb modules. The column on the right is 4Mb ones.

While 386 and later hardware is somewhat tolerant to RAM chips quality, the fast 0-wait state 286 systems are brutal towards them, especially when overclocking is in order, so will start with that.

--- 286

The best chips and modules are the ones in zone 1.
All of them are 27.5MHz 0-WS capable in 4x1Mb configuration.
Roughly half of them are 30MHz 0-WS capable, but in 2x1Mb configuration only.
Hynix HY514400J-70 + parity chips Hynix HYB511000BJ-60 (relabeled by Siemens) or Mosel-Vitelic VT531000 J-60.
There are no modules with VT531000 J-60 parity chips in the photos because i use them in the 286 hotrod PC. Didn't feel like pulling them out.
The last two sticks are with two chips only - they are problematic for 286 machines because of parity. A transplant will be needed to enable them.

Equally good are these 1Mb modules from MiGron (a forum user), based on 45ns rated Mosel-Vitelic V533C806HK45 chips.
Their packaging is a problem for all 286 chipsets i touched so far. It takes at least 286/386 hybrid silicon like Headland HT-16 or HT-18 to see them.
Motherboards based on them are impressive and can tick fine well beyond the 30MHz, but have internal wait states which reduce strain on the memory but lower the overall performance, so i am not sure if these chips are in fact better than the ones in group 1. Still, they are great for the right type of 286 boards.

The next best chips are in groups 3 and 3A, where group 3 is a curated subset of 3A.
IBM branded modules. Some of them work really well at up to 25MHz 0-WS, including in 4x1Mb configurations.
A hodge-podge of different chip brands/models.
OKI and Siemens using Hynix HYB511000.
Also, clumsy relabeling - multiple serial numbers and other stuff printed on top of each other.

Group 3A.
These are good for 16-20MHz 0-WS in 4x1Mb configurations.

The rest of the chips in the big pile are ok for systems running within specification, at low frequencies (12-16MHz) with 0-wait states, or at higher frequencies (above 16Mhz) with inflated wait states.

--- 386

Based on my arbitrary standards, 386 class hardware requires no less than 16Mb of system memory. This can often be a problem for peak overclocking.
In such cases reducing the RAM size can enable tighter timings, but that's below 16Mb, so it is discarded. Ok, back to the subject.

The best type of modules are in group 4.
Both of them are performing equally well in 2-3 hotrod 386 motherboards ticking at 50 and 55 MHz base frequency and 2-WS.
The first 4 sticks are based on LGS GM71C4100CJ60.

The second 4 are from MiGron and using Siemens relabeled Hynix HYB6117400BJ-50 chips.

The Micron based modules in group 5 are also very good, but roughly half of them are unreliable in fast 386 motherboards at 50MHz or more.

Everex STEP NS EV-1605B - an impressive 8086 / Turbo XT motherboard.

Assembly is in pristine condition - crisp.
Build quality is impressive. That statement is especially true in the context of XT class hardware.
It is either a very late product - won’t be surprised if it is from year 1993-ish, or if it was released before then - it was some high-end piece of equipment, and hats down to Everex designers/managers for that.
I am totally guessing here since cannot find any information about release date, product line, etc.
The board is not a standard AT form factor. This almost made me skip on it since i keep healthy distance from custom/proprietary contraptions.
The standard AT power connector and included ISA slots connector tipped the weights in favor of buying it. Glad i did it.

The board has plenty of gear integrated - FDD controller, COM/LPT controller, CGA/EGA graphics adapter.
Not sure which ICs implement the on-board video card. Cannot find any information about the two main chips. Must be the one in the upper left corner but who knows.

Included were 10MHz rated Siemens 8086 CPU and 640Kb of 120ns system/video RAM.
The Siemens CPU was substituted by 16MHz rated NEC V30HL.
Populated the empty FPU socket with my best 10MHz rated Intel 8087 chip. These get hot, so heatsink+fan are mandatory or overclocking is not happening.

Another important feature of the board is the small number of RAM chips.
Fewer RAM chips equates to better chance of higher overclock since overclocking of 808x hardware is usually limited by RAM (and FPU).
Replaced the 640Kb of 120ns chips with trusted 70ns rated Hynix and 80ns Micron ones.
The socketed system/video RAM is a time saver otherwise i would have to add the sockets. This is necessary to rotate chips around until find a set that allows stable overclock.

There are plenty of crystal oscillators present.
The two in the upper left corner are related to the on-board graphics adapter.
Then there is four in the middle.
The upper one is for the floppy (i think).
The 14.3181 does not need explaining.
The full-can is for the system clock.
The fourth one - not sure what its purpose is.

There is a BIOS menu that allows changing some options like floppy drive(s), display type, etc.

--- performance

This motherboard is slightly slower than Tandy 1000SL on a clock-to-clock basis, but overclocks like crazy. Will use the word "impressive" again.
Didn't think 808x hardware is 20MHz capable, but was proven wrong, with one small caveat - no FPU.
The best 8087 FPUs are up to 10MHz rated. Some rare chips can go up to ~17MHz, but no further. 20MHz is just too much for them.

For the testing i used Lo-tech XT-CF-lite rev.2 adapter and Western Digital WD90C00-JK 512Kb which is the fastest video card for XT machines.

Here is how some benchmarks look at 20MHz:

The board was tested at 10MHz (for clock-to-clock comparison with the rest), 16.5MHz with full gear (including FPU), and peak overclock of 20MHz (without FPU).
It is completely stable in all 3 configurations.
And the best part - it just does it !
... which i found ... impressive.
: )

benchmark results in the following post

Consolidated benchmark results for 808x hardware.

Everex EV-1605 at 20MHz cannot run with FPU.

What's the key to getting an 8087 to overclock to 17MHz? I once tried to overclock my 8087-1 to 12MHz and it failed miserably.

So does the Everex run the ISA slots at 20MHz too?

I randomly found this circuit which is a bit strangely named "CPU gearbox" Page 86 of Elektor Electronics 303 Projects book project number 060
https://archive.org/details/Elektor_Electroni … ge/n85/mode/2up

It's for a 6502 system, but what I'm thinking is that it might be useful for getting high clock on x86/88 systems where you can get a high mem/cpu clock and maybe bus or FPU clock is giving you problems, so it "downshifts" to a speed to cope with that i/o when required then "shifts up" to max speed again for other stuff.

BitWrangler wrote on 2024-01-24, 18:39:

I randomly found this circuit which is a bit strangely named "CPU gearbox" Page 86 of Elektor Electronics 303 Projects book project number 060
https://archive.org/details/Elektor_Electroni … ge/n85/mode/2up

It's for a 6502 system, but what I'm thinking is that it might be useful for getting high clock on x86/88 systems where you can get a high mem/cpu clock and maybe bus or FPU clock is giving you problems, so it "downshifts" to a speed to cope with that i/o when required then "shifts up" to max speed again for other stuff.

Pretty cool! But it requires that the system is stable when changing speed on the fly ..

BitWrangler wrote on 2024-01-24, 18:39:

https://archive.org/details/Elektor_Electroni … ge/n85/mode/2up
"downshifts" to a speed to cope with that i/o when required then "shifts up" to max speed again for other stuff.

Thats more or less how cheapskate Amiga 14MHz turbo modification worked https://aminet.net/package/docs/hard/14MhzA500
68000 has a lot if instructions burning high number of cycles between memory accesses, idea was to run CPU at 14MHz until it requested IO, then slow down to 7MHz, then back to 14Hz again. In practice just the clock trick without real fast memory results in ~50% speed bump, great considering whole thing only needed three 10 cent chips.

Diagram showing the logic https://github.com/Mathesar/14mhz-accelerator … _schematics.pdf of course this one has additional fast ram, but clock switching logic is afaik the same.
14MHz CPU only 1.52 x Amiga600 https://github.com/jbilander/SF500/blob/main/ … d_A590_pic5.jpg
14MHz CPU with additional fast ram also running at 14MHz 2.74 x Amiga600 https://github.com/jbilander/SF500/blob/main/ … _rev1b_pic7.jpg
14MHz 68010 (only difference is small optimization allowing for some loop instruction to reuse two word prefetch queue instead of refetching from ram https://pasti.fxatari.com/68kdocs/68kPrefetch.html) 2.93 x Amiga600 https://youtu.be/P_PzyiNm_DI?feature=shared&t=305

Btw I recently watched "Live with Dave Haynie - Commodore Business Machines C128, Plus4, Amiga" - BilHerd https://www.youtube.com/watch?v=6ZT209i-3Lo where Dave revealed Commodore was paying $2.5 per 68000 Hitachi CPU while Apple was paying $8 for Motorola ones.

Sorry for further derailing PC Battle station thread 🙁

@Anonymous Coward
The key to get 8087 FPUs is to have enough of them at hand.
Unlike the NEC V30HL these FPUs don't scale very well.
ISA bus runs synchronously. Plenty of ISA cards handle 20MHz just fine. They start to get unstable in the 25-30MHz range.

@BitWrangler
Nice find, but as @H3nrik V! said - in this particular case the system never lights-up if 8087 chip is present and the base frequency is above 17MHz, so this approach won't help.

@rasz_pl
Good info and a spicy story about 68000 pricing. Linked video is too long. Did they clarify why the difference ?
Never tried real Amiga. Only in emulation so far.
I get the itch sometimes, but the whole thing feels like another $$$$ black hole, so hesitation is a factor.

---

So, based on what BitWrangler and rasz_pl pointed at - can 8087 FPUs run asynchronously from the base clock ?
There are some vague hints about that, but i don't think it is possible. Anyone ?
Asking this because if possible, this will be the best option - keep the rest of the system at full throttle and only modulate the FPU.
This does not even need to be automatic. It can be manually controlled by user, or be a hardcoded FPU downclocker.
Otherwise, if the entire system has to be trotted down - that's basically a turbo switch, or a simple contraption to switch between 2 crystal oscillators if the motherboard does not offer turbo switch.

pshipkov wrote on 2024-01-28, 04:21:

Good info and a spicy story about 68000 pricing. Linked video is too long. Did they clarify why the difference ?

Yes, Commodore had its own fab (MOS) so every time they needed chip from third party a sample was decapped and fab people analyzed how much would it cost to build in house. Much easier to negotiate with such report in hand 😀 If negotiations failed they would just reverse engineer or copy the chip, thats what they did for C64 PLA after Signetics didnt want to play ball.

pshipkov wrote on 2024-01-28, 04:21:

So, based on what BitWrangler and rasz_pl pointed at - can 8087 FPUs run asynchronously from the base clock ?

x86 FPUs use some really weird and unintuitive communication mechanisms. Let me google 8087 FPU... there it is https://www.os2museum.com/wp/learn-something- … 87-intricacies/

"The FPU snooped the CPU’s bus transactions and observed all instruction fetches. The CPU used the QS0/QS1 pins to inform the 8087 about the state of its prefetch queue. The FPU essentially executed instructions in lockstep with the CPU."

That means its impossible.

387 directly supports Async mode because FPU-CPU communication is even weirder. Afaik its something like CPU performing an IO operation to special magic ports 8000F8h/8000FCh/8000FEh 😮 a combination of pins A23 and M/IO are used to trigger FPU to wake up, slow down CPU and snoop its bus.

pshipkov wrote on 2024-01-28, 04:21:

The key to get 8087 FPUs is to have enough of them at hand.
Unlike the NEC V30HL these FPUs don't scale very well.
ISA bus runs synchronously. Plenty of ISA cards handle 20MHz just fine. They start to get unstable in the 25-30MHz range.

Do you think it would help to have a military spec 8087? I've been keeping my eyes open for one for quite a while, but I only ever see the 5 and 8MHz versions. Was a 10Mhz part ever made?

I dunno if milspec will be all that different clock frequency wise, most stuff was just extended temperature range and better radiation screening.

The military graded 8087 chips, prefixed with MD/ML, are not great. But that opinion is based on a single chip i have.
The error can be large, so keep that in mind.
8087 FPUs get very hot, but simple heatsink+fan take care of that to a large extent. Speaking about room temperatures here of course.
If not active cooled they don't go much higher than 10MHz, especially noticeable in long-running tests and use cases.
I use these 10MHz rated chips.
Some of them can do 17MHz, which is impressive 70% overclock, but still ...

Is the production date 1983 or 1993?

83

It's really too bad intel never bothered with an 80C87. It seems they were still hawking the same old 8087 in the early 90s using the same ancient die process.

Paktron Electronics Ltd. PX-3000 Turbo - an 8088 Turbo XT motherboard based on the PX3000 (?) chipset.

An interesting looking package.
Highly integrated, very compact, still in great condition.
First time seeing this "chipset". Not sure if it is an unique design, or something relabeled.
All good and everything, but what actually caught my attention was the supported memory configurations.
Only 4 chips are required for 512/640 Kb - an important factor for XT overclocking activities.
In the majority of the cases RAM is the limiting factor, since these old memory chips don't handle very well higher frequencies.
The strategy is to curate a set of them that can do it. Fewer chips = faster curation process and improved reliability.

Base frequency is 4.5MHz.
Turbo mode is 9MHz.

Desoldered the 27MHz crystal oscillator and put female pins in the corresponding through-holes.
The board actually functions just fine without the TURBO-XTAL oscillator.
Either the base frequency is derived from the 14.3818Mhz one, or the main IC has integrated clock generator.
But if TURBO-XTAL oscillator is present, it is considered.

The Retro Web has documentation for jumpers and switches. Won't go through any of that here.

Tested with the next components:
16MHz rated NEC V20 CPU
10MHz rated Intel 8087 FPU,
70ns rated 512Kb RAM, Hynix
Western Digital WD90C00-JK 512Kb video card (seems to be the fastest thing for 808# machines)
Lo-tech XT-CF-lite rev.2 XT-IDE controller with Transcend 2Gb CF card
DOS 4.01

Had big hopes about this assembly, but they didn't materialize.
9MHz is the maximum the system can take. That was a bummer.
The rest of the used components can take much higher frequencies, as shown in previous posts.
So, the limitations seem to be in the motherboard itself.

benchmark results

Neat looking motherboard.
I can see how it can be a great fit for someone's compact Turbo XT build.
Unfortunately does not overclock at all, which limits potential fun activities and tinkering around it.

Cute board, shame about the lack of headroom. Don't let it near me, with most of the important stuff up one end I'd be reaching for the saw and thinking "XT ITX" 🤣

Let's first check that mythical "garage" of yours.
Good chance there is a PX-3000 laying in that dark corner.
But the healthy choice will be to stay out of XT-ITX related activities entirely.
Which reminds me, how is the BL3 rig coming along ? ;P