The NA-SEL jumper is through-hole. The PCB is 2mm thick so the legs stick out less.

The NA pin (A13) for most i486 and am486 is NC or INC (internally not connected), so using no jumper or 2-3 is the solution for the CPU, however motherboards that use NA can't be used.

The NA pin is pulled up to 5v via 20k.

The default is 1-2 for TI486SXL CPUs. Motherboards may pull the pin up/down stronger. The 20k pullup attached is likely not required if the motherboard has much stronger pullups anyway. Option 2-3 is disconnected and pulled down strong to avoid hair trigger noise on the pin.

There's always a 20k pullup to NA unless you desolder resistor R1.

I don't know which boards used the pin for pipelining. The Symphony board you listed with 1k pullup may suggest it uses it(?)

The C&T F82C351/C might support 50mhz (itself, not necessarily other components). I bought a C&T 82C836B-25 (build date 2004) recently for the 386sx and it ran at 50mhz. More improved silicon process(?). A recently produced B-40 may also run at 50...

It's been through vinegar baths, fixed many lines & replaced ICS, I just need to check more in the area of the battery.

Whoops, not sure how I missed the through holes on that jumper.

The Symphony Haydn-II has a particular affinity towards the SXL, so it may very well support this pipelining.

I was repairing some traces on my Mac SE/30 the other day and noticed that even some traces which look intact can have non-visible breaks for which the multimeter sometimes catches, and sometimes not.

Feipoa, have you tested your interposer with a SIS Rabbit chipset 310/320/330? Does it work on this chipset?

I bought a new 386DX board, Acer V5 (AcerPower 386/33w PC) SIS Rabbit 310/320/330. Changed BIOS with many of the alternatives on retroweb. Only blank screen on boot with the SXL2.

What a wonderful project!

As an owner of a couple of SXL2s (unused) and of several 386 boards that are already capable of recognizing the Cyrix 486DLC, I'm wondering if a small production run will be available to those hobbyists who don't have the necessary skill to assemble something so complex by themselves.

@mikesg
Yes, it does work on sis rabbit based boards.
Successfully tested on several of them.

MSxyz wrote on 2024-07-10, 19:53:

What a wonderful project!

As an owner of a couple of SXL2s (unused) and of several 386 boards that are already capable of recognizing the Cyrix 486DLC, I'm wondering if a small production run will be available to those hobbyists who don't have the necessary skill to assemble something so complex by themselves.

If/once my version works I will most likely sell it on Ebay in order to buy more sockets to build more. Also you can build it yourself.

Mine uses the fancy sockets, so it's expensive, but it also makes making a lot of them difficult.. The Price for one to build yourself is ~$210 AUD / $140 USD. Price for two is $175 AUD. Price for three+ is $150 AUD. Plus 1-2 hours soldering. So if you just wanted one pre-built it would be $200 minimum AUD.

MikeSG wrote on 2024-07-11, 07:51:

If/once my version works I will most likely sell it on Ebay in order to buy more sockets to build more. Also you can build it yourself.

Mine uses the fancy sockets, so it's expensive, but it also makes making a lot of them difficult.. The Price for one to build yourself is ~$210 AUD / $140 USD. Price for two is $175 AUD. Price for three+ is $150 AUD. Plus 1-2 hours soldering. So if you just wanted one pre-built it would be $200 minimum AUD.

Well, building one myself is out of question as my soldering skills and equipment are nowhere near the level needed for such a project.

I would gladly pay that sum for one already assembled and ready to use. The worst it can happen is a black screen and no POST, which is something I'm already accustomed to in this hobby of mine! 😁

There should be a pretty good chance of them working on most boards. Definitely the newer boards.

Currently testing issues with it..... many of the pins on the PGA-168 pin press-fit are not contacting properly. Going to try ordering PCBs with ENIG finish next time instead of HASL, they're supposed to be much better for Plated Through Hole. The tolerances are really fine on the PGA-168 pin holes.

MikeSG wrote on 2024-07-11, 15:52:

Currently testing issues with it..... many of the pins on the PGA-168 pin press-fit are not contacting properly. Going to try ordering PCBs with ENIG finish next time instead of HASL, they're supposed to be much better for Plated Through Hole. The tolerances are really fine on the PGA-168 pin holes.

I recall there being an option with JLCPCB to upgrade to ENIG 2U, which would double the gold thickness. Costs a little extra. May help some if you end up pulling the press fit sockets in and out too much.

It may help, but I would test fit all ~5 PCBS for the best fitting (and test connections) anyway.

The unevenness from the HASL is the major thing. It's supposed to be +/-0.05mm tolerance but the PCB can't be ordered with that...

2.0mm PCB adds $62AUD/$40USD
ENIG adds $40AUD/$25USD
ENIG 2U adds another $50AUD/$30USD
Press-Fit Holes (+/-0.05mm) adds $270AUD/$170USD

Everything except 2.0mm PCB and ENIG costs too much.

isnt all of it self inflicted? $5 pcb + individual pins seem just fine

Initially i had only one SXL2-66 CPU that didn't go much further than 70MHz. Regardless if air cooling or Peltier.
An important detail is that i have an ATX PSU with AT adapter on the test bench. It outputs 4.8V instead of 5V. After few more cascades down it hits the adapter with mere ~4.6-4.7V.

2 weeks ago i obtained 6 SXL2-66 CPUs. Neither of them can do stable 80MHz under the above conditions.
Bumping the Peltier to 12V didn't do anything. This is highly unusual because 12V Peltier can strap any CPU to toe the line on overclocking (assuming decent silicon die).

Discarded the direct 5V approach outlined here because it implies PSU with good 5V lines.
Went for the 12V option instead.

Inserted a PGA132 socket between the adapter and the motherboard. Removed all the 5VCC pins from it. Connected the input pin of the adapter's MIC VRM to the +12V line.
Noticed that once the VRM starts heating up the output voltage gets a bit unstable. Heatsink and fan resolved the issue.

With 5.2-5.3 V to CPU and 5V Peltier two of the processors are mostly stable at 90MHz. DOS interactive graphics tests pass, but anything more challenging than that is still unstable.

DTK PEM-0030Y which has the same performance and overclocking characteristics of 4036Y.
256Kb L2 cache
16Mb RAM
STB Nitro 2Mb (CL GD-5434)
Standard IDE + CF card.

All BIOS settings on max except:
TIMING PARAMETER SELECTION = EXTENDED (best is NORMAL)
DRAM WAIT STATE SELECTION = 2 W/S (best is 0)
BUS CLOCK SELECTOR = CLCK2/2 <--- that is 22.5MHz ISA bus

cyrix.exe -cd -e -i1 -i2

Some early results:
Superscape - 50 fps
PC Player Benchmark - 13.2 fps <--- impressive. best result to date.
Doom - 31.9 fps <--- even more impressive. not sure what to think about it yet.

Stupid good results. Much better than BL3 at 110MHz (2x55). Deep in 486 territory.

And couple of pictures from the quick Saturday morning workshop:

The less than 2 hours effort is inconclusive about long-term stability and peak performance.
Will see how it goes in the coming days, but the prospect of complete stability is not looking great in this first round.
Don't see how Peltier can be integrated with the SXL2-66 contraption to avoid water condensation.
Looks like the more likely path to stable system is to go through more SXL2 CPUs and find the magical one that can do 90+MHz on air cooling.

Two more notes:
100MHz is possible, but highly unstable.
SXL2-66 CPUs can survive 6.8V but don't ask how i found out.

It is possible to get DRAM down to 1 W/S. I know this is stable in DOS from previous testing of these DTK motherboards.
DRAM WAIT STATE SELECTION = 1 W/S (best is 0)

Doom is now at 34 fps.

Quake 1 is at 10.5 fps. This is silly.
First i thought i am getting some skewed clock or something, but the observed framerate is really "high".

EDIT:
The Quake 1 result above is incorrect.
It was a freak moment where the clock got skewed somehow.
The actual frame rate is 3.9 fps.

Wow, you used the heat pipes! Play DOOM and Quake some to see if the games really feel like 33 fps and 10 fps, respectively. I will be curious to know if more testing concludes if it is necessary, or if a standard 45 mm heatsink with 10 cfm fan is sufficient.

Your benchmark results are better than good. I bet if you take the speed down to 87 MHz, you can achieve stability with the existing CPU at 5.25 V. At what voltage did you determine you could achieve stable 80 MHz operation?

Very interesting to note that peltier did not illicit any positive improvement.

One note that I probably should have brought up a few more times, was that the MIC prefixed VRM tended to need 0.15 - 0.2 V more voltage compared to the LP prefixed VRM for stability. This was consistent across the CPUs. I recommend using the LP regulator for this reason, even if you are only pushing 80 MHz. It lets you run the CPU at a lower voltage.

If interested, you can get the heat generation down some on the VRM side if you use an in-line switching regulator in series with the linear regulator. I'd say this is more of a finishing touch though.

I was surprised to see you running DRAM wait state at 2 ws. My setup, with the same motherboard, runs at 1 WS. But later you provide 1 ws results. Was your setup initially not stable at 1 ws?

Did taking the CPU up to 6.8 V improve anything stability wise? The most I tried was 5.35 V.

Shot a video with my phone at 30 fps while running the DOOM test. Flipped it frame by frame after that. Used the known framerate of the video as a strob. The ~30 fps result holds true.

I am starting to wonder if we crossed into 486 territory here and violated our own rule to stick with 386 class CPUs.
According to the diagram in this document, page 2 - the SXL2 is a 486 class design. The rest of the notes are along these lines as well.

While exploring this hardware path and the performance advantages it brings, i am starting to wonder if is should go straight for a 386-486 interposer and strap an Am-5x86-133 running at 4x50MHz. : )

Didn't spend time on 2x40MHz. Went for 90+.

At 45MHz base frequency and higher some of the heavy tests may fail from time to time with DRAM WAIT STATE SELECTION = 1. All is fine in DOS but i know the system will eventually flake when pushed hard.

Thanks for the hint about the VRM models. I am very far from there. I hope to get to a point where i will want to think about it.

The system finished booting to DOS when i noticed the 6.8 volts to CPU. Didn't test anything. I am just proud with my response time powering it down.

That article mentions that the IBM 486SLC2 is faster than the TI SXLC2, but that the IBM unit couldn't be sold except as part of a system or subsystem - but this was also true of the TI SXL2 and SXLC2 series. I guess this article came out before the pushback or lawsuit from Cyrix?

The article also talks about how Intel didn't make a 486SX2, but they did [eventually].

Says how the SXL is the same as the DLC, but with more L1 cache, and how the SXL is slower than Intel's 486 core. The SXL "implements a 386DX-like bus interface in a 486DX pinout". The article writer calls it a "near-486-class CPU". To me, the SXL2 feels more 386-specific and also came in a PGA132 package, while the IBM DLC3/BL3 did not. I think all IBM SLC2 and DLC3/BL3 systems came on VLB systems. If I were to rate which CPU, the SXL2 or the BL3, was inching closer to 486-class, I would pick the BL3.

There's a table in the article from PC Mag 6.0 claiming the SXL2-50 is about 50% faster than the i486SX-33. I thought the SXL2-50 was closer to an i486-25. Has anyone run this comparison using non-synthetic tests?
.
.
.
If you take your ISA down to 11.25 Mhz (CLK2/4), what DOOM score are you getting? I'd like to compare our numbers on the same board (PEM-0030Y) to ensure funny business going on with your system. Something certainly feels off with the clock.

At 90 MHz and 11.25 Mhz, I receive a DOOM score of 23.3 fps (DRAM 1 ws) or 23.1 (DRAM 2 ws). Your system is obtaining 33.9 fps in DOOM. Is the increase in ISA speed from 11.25 MHz to 22.5 MHz enough to account for such an increase in DOOM frame rate? I suspect not.

Judging by https://thandor.net/benchmarks/vga/vga?a=69&c=0&o=0&s=Submit this fps jump after removing VGA bottleneck seems plausible
ET4000AX 1MB ISA(P100) 27.10
Stealth 64 PCI 4MB PCI(P100) 66.27

My opinion about 386 upgrades varies. On different days the answer is different.

DOOM at 90/11.25/1ws = 24 fps (23.97 fps, 3116 ticks)

I am still not 100% sure about the timings and test results i report. Something feels off.
Will have to spend more time on that. Good chance i have to edit the posts.

@rasz_pl
Cool link. Very interesting. Thanks.

Hmmmm.... Your system gets 0.7 fps faster than mine, although the MB and graphics card are the same. I probably have some DOS drivers or cyrix settings slowing it down. I don't cache the first 64 MB afer the 1 MB boundary, whereas you do.

At any rate, what does your BL3 get at 2x45 and 2x55 on the PEM-0030Y? From 24 to 34 fps from 11.25 to 22.5 MHz feels erreneous, thus was wanting to see the BL3 differential at the same ISA rates.

Most likely is the CYRIX.EXE flags. I have everything on. It matters.

Spent some time today to more carefully examine the reported performance results.
Wolf3D - varies from 100-130 fps. This test is slightly volatile, but always within 0.1-2 fps range. 20-30 fps difference between remote test runs (several restarts apart) gives me a pause.
EDIT: Yet, the 100+ fps is clearly noticeable on screen. Frames just fly. It is 100+ indeed. Saying this very confidently. This is incorrect. Proved wrong in following post.
Superscape - 52.5 fps. No volatility.
PC Player Benchmark - 13.2 fps. No volatility.
Doom - 33.9 - 34.5 fps. Varies within that range. This test is fairly consistent between runs, so not sure what is going on.
Quake 1 - 3.9 fps with one exception very early on while getting SXL2-90 running. Reported 10.5 fps which never occurred again.

All CPUs are quite unstable, so a lot of attempts to test anything go nowhere. I am very far from getting to the interesting (to me) tests if cannot handle the simple DOS interactive graphics.

One thing i realized is that i actually don't need Peltier at all. I cannot see a difference right now with or without Peltier in the loop. As long as voltage is above 4.8-4.9 volts the system behaves similarly. This is inconclusive of course, because the CPUs are so unstable as they are, but it is clear to me that there is no significant thermal factor with them, even when running them with 5.3-5.5 volts.

The best i was able to get in DOOM with BL3 on these Symphony boards was 28.33 fps 2x55MHz.

The best stable number in Doom is with MSI MS-3124 at 2x55 with BL3 as listed here.

So, if i am not fooling myself somehow with the collected performance numbers, we can say that Symphony Haydn + SXL2 + fast ISA bus is emerging as the ultimate 386 DOS graphics blaster. It is just significantly better than anything we have seen so far.