are these actual BL3 chips or are you quasi making them somehow out of Intel chips? And they just become BL3s because of the interposer? What makes a bl3 a bl3 and not just another intel chip? I think I recall that these were intel produced chips yeah?

Is there a source for them somewhere? (The bl3 chips themselves)

@Feipoa
Yes, these wires must be removed before eventual unsoldering of the processor.
The rest will be fine.
You already used soldering paste with low temperature melting point.
You blow around clear areas for some time and stay away from the elements at risk. This will heat-up the PCB and by itself may be enough to separate the CPU. In the worst case a quick pass or two in risky areas will be enough for unproblematic separation.

@Sphere478
They are more advanced silicon than Intel/Amd 386 chips.
Quick search online will reveal the specs (and some misleading youtube videos).
BL2 BL3 CPUs from IBM are OEM only parts that are pin compatible with 386SX and 386DX, respectively.
16Kb level 1 cache. 1x, 2x, 3x internal multipliers, software controllable.
33/66 and 25/75MHz rated chips available. Most of of them can operate in the 80-90MHz range.
Selected few can hit 100MHz (2x50).
Slightly clock-to-clock faster than SXL2 processors. Feipoa saw on one of his motherboards the opposite.

Few links with more concrete data for your reference:
BL2 in Alaris Leopard
Bl3 in Alaris Cougar II @75MHz and unstable 100MHz
Bl3 in Alaris Cougar II @100MHz
Bl3 in DTK PEM-4036Y @100MHz - the more relevant info for the topic here.

You probably better at hot air than me. It took me several hours to place all these components and I cannot do what you are suggesting without a near redo of everything.

The trimmer is nearly touching the leads and its height makes leaving it in place unrealistic. The casing on the trimmer will melt. It might survive, but will look horrible. There are another four wires with thin cladding that melts easily. These four wires are all running through and under the corner of the CPU (not really visible in photo). The cladding will melt and the wires will short. In the first mage, you can see how easily the cladding melts. That melt is just from a very quick desolder from the AMS1117 unit.

There are an additional four wires holding on the de-coupling caps. They all have to be removed. The tantalum can probably stay, but the casing will turn brown from the heat. It will look bad. I've seen this happen with the SXL2 work. The tantalum is fairly easy to install, so I'd remove it anyway.

From my experience, the only realistic way to swap the CPU is to replace 9 carefully placed wires, break the epoxy on the trimmer, pull it, and redo. I'd rather source another 386 interposer, ideally the same model for easily replication.

The CPU which is on the existing interposer worked up to high frequencies on the IODATA module. I think 110 MHz, but essentially without L1 (flush cache every I/O). I have another IODATA which couldn't go past 75 MHz under the same test conditions. I plan to solder the NOS BL3 onto the IODATA donor. If it also works at high frequencies, then my second NOS BL3 may be a suitable candidate for another interposer.

EDIT: I should clarify - the issue with replacing the BL3 on the existing module is the removal, not installation. I can do hot air w/paste to get a new chip on, but removing the chip with those components will create a lot of re-work of components.

pshipkov,

Another idea I had was to do a straight swap on my virgin JATON module and run the BL3 at 5 V. If it cannot handle 100 MHz at 5 V, it likely cannot handle 100 MHz at lower frequencies. When you finally got your BL3 working at 100 MHz, what voltage did it work at?

Could you also remind me what the fastest stable BL3 frequency you achieved on the hack module at 5 V?

pshipkov wrote on 2023-01-21, 03:03:

@Feipoa Yes, these wires must be removed before eventual unsoldering of the processor. The rest will be fine. You already used s […]
Show full quote

@Feipoa
Yes, these wires must be removed before eventual unsoldering of the processor.
The rest will be fine.
You already used soldering paste with low temperature melting point.
You blow around clear areas for some time and stay away from the elements at risk. This will heat-up the PCB and by itself may be enough to separate the CPU. In the worst case a quick pass or two in risky areas will be enough for unproblematic separation.

@Sphere478
They are more advanced silicon than Intel/Amd 386 chips.
Quick search online will reveal the specs (and some misleading youtube videos).
BL2 BL3 CPUs from IBM are OEM only parts that are pin compatible with 386SX and 386DX, respectively.
16Kb level 1 cache. 1x, 2x, 3x internal multipliers, software controllable.
33/66 and 25/75MHz rated chips available. Most of of them can operate in the 80-90MHz range.
Selected few can hit 100MHz (2x50).
Slightly clock-to-clock faster than SXL2 processors. Feipoa saw on one of his motherboards the opposite.

Few links with more concrete data for your reference:
BL2 in Alaris Leopard
Bl3 in Alaris Cougar II @75MHz and unstable 100MHz
Bl3 in Alaris Cougar II @100MHz
Bl3 in DTK PEM-4036Y @100MHz - the more relevant info for the topic here.

Great reply! Thanks! 😀

So what I am ultimately leading to with my questioning is, I kinda wanna put one on my motherboard 🤣.

Where can I get one? Though?

I’m guessing maybe a pin mod and a voltage mod may be required though? So might not do it, but not impossible…🤷‍♂️

I have run a few more tests with the BL3 using the ultra low dropout regulator, TPS72501DCQR.

The max voltages I'm able to achieve are:
Sitting at the BIOS (no L1): 4.90 V
Sitting at idle MS-DOS (no L1): 4.85 V
MS-DOS w/L1 enabled running software: 4.7 V

That's a lot more flexibility compared to the AMS1117.

90 MHz is achievable at 3.3 V. To run BIOS at fastest timings, I need 3.6 V. Previous tests were with an ET4000AX, which were holding the system back with regards to BIOS timings. I'm testing now with the GD5434.

At 2x45 = 90 MHz, best I can get in DOOM is 3165 realtics, which translates to 23.6 fps. I ran the ISA at 15 MHz. Running it at 18 MHz results in a lower score. At 12 Mhz, the GD5434 scores 3500 realtics.

I ran some VRM temperature measurements at full load. 3.3 V = 41.2 C and 4.0 V = 44.4 C.

pshipkov, is 2x45 w/fastest BIOS timings faster than 3x33 on the BL3? If so, then I guess the system is already optimal. I can try 2x50, but I remember 50 MHz being problematic in general.

@Feipoa
You know how great i am with the hot gun ... cough ... Trio64V+ assembly. :}
Photos do not convey physical reality well enough which made me optimistic.

Btw, you don't need the exact same interposer. Any 386DX one will do.
Without L1 cache running the good CPUs deft can go over the 100MHz.
So far in all hardware configs i tried - L1 cache is the limiting factor.

Good info in your last post.
Will stare at it for some more, but that regulator is certainly the way to go. Will be going for it here - one of the interposers will be converted.

24 fps in Doom @2x45 is very good result.
If i remember correctly with 2 x 45MHz i get ~24 fps as well.

But 18MHz ISA bus results to be lower than 15MHz is unexpected. Are you sure ?
Maybe some internal auto-wait-state kicking in ?

The temperature difference between 3.3V and 4V is negligible.
This is good news. So leaving higher voltage does not impose much higher toll on the chip.
Need to check what i am getting at 5V ...

Cannot remember off the top of my head how 2x45 stacks against 3x33.
Will do a quick spin later tonight or tomorrow.

So far i didn't try voltage modulated.
Everything i run is at 5V.
So cannot comment on your question how different voltages affect the 4 CPUs i tested with so far.
I shared in the past my suspicion that it is not a significant factor.
At 5V i have 2 CPUs hitting 100MHz, 1 CPU hitting 90, 1 CPU not going much further than 80. Assuming complete stability.
One of the CPUs can do 110MHz under 12V Peltier. Inherently unstable. Motherboard is the main factor. That's why wanted to try on one of the OPTi boards here that can do 60MHz FSB, will do that later, even if it hits it - perf will be low - i know it. So, for now 100MHz is the practical limit i was able to achieve.
I think 5V with proper cooling is fine. Practical and functional, without significant risk for the CPU.

---

@Sphere478
The attached photo shows Forex chipset.
These don't work with BL3 processors, but you can try SXL2 - very good alternative to BL3. Another option will be DLC chip.

Btw, these Forex guys are not very fast.
Strapping a powerful engine to Trabi still gets you Trabi.

As for obtaining BL3 chips - these are scarce and expensive.
Usual sources are Cougar II boards, some IBM PCs and laptops, IO-DATA adapters.

Man, you are at risk of inserting yourself in one of the most weird retro computing niches.
Just a friendly warning. Watch your step. : )

I think L1 disabled and L1 flushing every I/O aren't the same situation. Being able to run 110 MHz with L1 flushing I/O is probably more encouraging that just 110 MHz and L1 disabled.

On my SiS Rabbit board, I ran a few clock-for-clock comparisons between the BL2 and SXL2 at 80 MHz internal. Results are with GD5434. BIOS settings identical between CPUs.

SXL2 at 2x40
ISA 13.3 MHz - 3775 - 19.79 fps
ISA 16 MHz - 3703 - 20.17 fps

BL2 at 2x40
ISA 13.3 MHz - 3562 - 20.97 fps
ISA 16 MHz - 3570 - 20.92 fps

Notice how increasing the ISA rate on the BL2 worsens the result, while it improves the result on the SXL2.

On this motherboard, the BL2 is only 5% faster, per clock, than the SXL2. However, the BL2/3 will be limited to 16 MB cacheable range, while the SXL2 should be able to cache however much RAM is on the motherboard.

OK, if you have two CPUs which are stable at 100 MHz and 5 V, then for a quickt est, I can solder the BL chip onto another interposer without VRM and see how it goes. From my tests at 90 MHz, there wasn't any benefit in going above 3.6 V, so it is likely you can run how chips cooler and at less voltage with the VRM.

I ran a few more tests. I can confirm that trying to use an ET4000AX was limiting my CPU speed to 83 MHz. I don't know how this makes sense, but with the GD5434, I can go up to 112.5 MHz at 4 V.

I was able to go up to 105 MHz at 3.6 V and 108-112.5 MHz at 4 V. I could run chkcpu at 120 MHz and 4 V, but it would crash when trying to run DOOM. I didn't increase the voltage further, in fear I might damage the CPU.

100 MHz at 3x33 and ISA at 13.3 MHz, then DOOM = 3511
105 MHz at 3x35 and ISA at 11.7 MHz, then DOOM = 3334
105 MHz at 3x35 and ISA at 14.0 MHz, then DOOM = 3345
108 MHz at 3x37.25 and ISA at 12.4 MHz, then DOOM = 3238
112.5 MHz at 3x37.5 and ISA at 12.5 MHz, then DOOM = 3108

So it took 112.5 MHz to beat the 90 MHz result of 3165 realtics.

pshipkov wrote on 2023-01-21, 06:58:

@Feipoa You know how great i am with the hot gun ... cough ... Trio64V+ assembly. :} Photos do not convey physical reality well […]
Show full quote

@Feipoa
You know how great i am with the hot gun ... cough ... Trio64V+ assembly. :}
Photos do not convey physical reality well enough which made me optimistic.

Btw, you don't need the exact same interposer. Any 386DX one will do.
Without L1 cache running the good CPUs deft can go over the 100MHz.
So far in all hardware configs i tried - L1 cache is the limiting factor.

Good info in your last post.
Will stare at it for some more, but that regulator is certainly the way to go. Will be going for it here - one of the interposers will be converted.

24 fps in Doom @2x45 is very good result.
If i remember correctly with 2 x 45MHz i get ~24 fps as well.

But 18MHz ISA bus results to be lower than 15MHz is unexpected. Are you sure ?
Maybe some internal auto-wait-state kicking in ?

The temperature difference between 3.3V and 4V is negligible.
This is good news. So leaving higher voltage does not impose much higher toll on the chip.
Need to check what i am getting at 5V ...

Cannot remember off the top of my head how 2x45 stacks against 3x33.
Will do a quick spin later tonight or tomorrow.

So far i didn't try voltage modulated.
Everything i run is at 5V.
So cannot comment on your question how different voltages affect the 4 CPUs i tested with so far.
I shared in the past my suspicion that it is not a significant factor.
At 5V i have 2 CPUs hitting 100MHz, 1 CPU hitting 90, 1 CPU not going much further than 80. Assuming complete stability.
One of the CPUs can do 110MHz under 12V Peltier. Inherently unstable. Motherboard is the main factor. That's why wanted to try on one of the OPTi boards here that can do 60MHz FSB, but BL3 does not work on that chipset unfortunately. So, for now 100MHz is the practical limit i was able to achieve.
I think 5V with proper cooling is fine. Practical and functional, without significant risk for the CPU.

---

@Sphere478
The attached photo shows Forex chipset.
These don't work with BL3 processors, but you can try SXL2 - very good alternative to BL3. Another option will be DLC chip.

Btw, these Forex guys are not very fast.
Strapping a powerful engine to Trabi still gets you Trabi.

As for obtaining BL3 chips - these are scarce and expensive.
Usual sources are Cougar II boards, some IBM PCs and laptops, IO-DATA adapters.

Man, you are at risk of inserting yourself in one of the most weird retro computing niches.
Just a friendly warning. Watch your step. : )

Yep, I’m gonna be using a sxl2 soon on it using the adapter feiopa and I are almost done with.

What’s the hotrod motherboard of the 386 world that I should keep an eye out for?

@Feipoa
Your scores, converted in frames per second:
100 MHz at 3x33 and ISA at 13.3 MHz, then DOOM = 21.3 fps
105 MHz at 3x35 and ISA at 11.7 MHz, then DOOM = 22.4 fps
105 MHz at 3x35 and ISA at 14.0 MHz, then DOOM = 22.3 fps
108 MHz at 3x37.25 and ISA at 12.4 MHz, then DOOM = 23 fps
112.5 MHz at 3x37.5 and ISA at 12.5 MHz, then DOOM = 24 fps

I don't have SXL2 processor that can handle 2x40.
Some numbers with DTK PEM-4036Y, BL3, GD5434.

Doom high-res:
80MHz (2x40) - 22.4 fps
100MHz (3x33) - 24.16 fps
100MHz (2x50) - 27.2 fps
120MHz (3x40) - 28.55 fps <- Can this be the highest Doom number around for 386 hardware ?

At 3x40 the system works pretty well but is not fully stable. It will eventually hang. It is the CPU.

Quake 1 chugging along

@Sphere478

I got some of the PCBs as well. Thank you guys for making the effort.

Hotrod 386 mobos ?
Look for Alaris Cougar II (performs well, otherwise flaky sensitive ballerina), or stuff based on SiS Rabbit or Symphony Haydn. Implementation and assembly vary a lot, so simply the presence of such a chipset is not an insurance for perf/quality.

pshipkov, the Symphony Haydn II is just faster than the SiS Rabbit. Do you have any results on a SiS Rabbit? Do you also have a clock-for-clock comparison with SXL2 and BL2 on the Symphony? I thought I recalled the SXL2 being clock-for-clock faster than the BL3 on the Symphony.

Given your results up to 120 MHz, it likely that I am already using an ideal BL3 CPU. I wonder if your 120 MHz system could be more stable by dialing down the voltage?

Your results at 120 Mhz are encouraging. Maybe I'll up the voltage further to see the min. voltage I can run DOOM at is.

Unfortunately, my Symphony board doesn't like the bare BL3 hack. Since you now have one board which has working BL3 and one which doesn't, I'd start probing around to find how the CPU sockets are connected differently.

The two SiS Rabbit boards i had are long gone.
Cleared most of the 386 motherboards inventory.
Only handful still around - couple of OPTi 495 and UMC UM481/491. To my surprise the UMC ones don't like BL3 processors.
I think i have some ALI 1429 one(s), but they are slow.

I tried freezing the CPU with 12V Peltier - still not fully stable.
From my 486 related experience with this kind of cooling - a 12V Peltier can turn around CPUs that otherwise do not tolerate 5V very well at 180/200 MHz.
Still, there is a chance to potentially improve stability at 120MHz by reducing input voltage. Will get to that at some point soon.

About DTK PEM-4036YB not liking BL3 CPU.
It actually works with them just fine.
But i tested it with a dead BL3 chip.
So from the 4 BL3 guys here one fell silent already.
That misled me for a moment to think that the mobo is incompatible with this processor.

I do have a comparison of SXL2 and BL3 on Symphony chipets at 1x50 MHz.
BL3 is slightly more efficient on a clock-to-clock basis.

That's too bad you don't have any Rabbits left. I think it is a good practise to keep around one board from some of these iconic chipsets which perform well. I know I've preached this before.

OK, this is the first I've heard that your industrial DTK PEM-4036YB can work with BL3 chips after all. Did you test it in Windows? I will have to try my DTK board again, now that I've determined there's a strong graphics element limiting even mild overclocking.

The two SiS Rabbit boards were from Asus.
Overclocked to 50mhz but 64kb l2 cache only and up to 8mb of ram only.
Kind of meh.

4036yb handling windows with bl3 - of course.
Your Symphomy board should be able to handle bl3. It is the same hardware like the 4036's hjust placed in different locations.

I intend to try my Symphony board again when I'm finished with the SiS Rabbit. For now, I was able to get 2x50 working. I had to take the voltage to 4.10 V at 2x50, even though 4.0 V was sufficient for 3x37.5 (112.5 MHz). I ran the ISA at 16.7 MHz and received 2856 realtics = 26.15 fps. I didn't try running the ISA at 20 MHz for this case. I am using an AHA-1522B card and wasn't sure if the SCSI would hickup at 20 MHz.

Edit1:
tried 20 MHz ISA, but the resultant was slower, 3024 realtics. For each configuration, it looks like there is a maximum ideal ISA frequency, for which going higher reduces performance.

Edit2:
Took the BL3 up to 4.63 V at 120 MHz, but enabling the L1 would cause a crash. That was as much voltage as the VRM would output for the current demand at 120 MHz. In summary, on the SiS Rabbit board used, here are some tallies with GD5434:

80 MHz at 2x40 and ISA at 13.3 MHz, 3.3 V, then DOOM = 3562 = 20.97 fps
90 MHz at 2x45 and ISA at 15 MHz, 3.6 V, then DOOM = 3165 = 23.60 fps
100 MHz at 2x50 and ISA at 12.5 MHz, 4.1 V, then DOOM = 3162 = 23.62 fps
100 MHz at 2x50 and ISA at 16.7 MHz, 4.1 V, then DOOM = 2856 = 26.15 fps
100 MHz at 2x50 and ISA at 20.0 MHz, 4.1 V, then DOOM = 3024 = 24.70 fps
100 MHz at 3x33 and ISA at 13.3 MHz, 3.6 V, then DOOM = 3511 = 21.27 fps
105 MHz at 3x35 and ISA at 11.7 MHz, 3.6 V, then DOOM = 3334 = 22.40 fps
105 MHz at 3x35 and ISA at 14.0 MHz, 3.6 V, then DOOM = 3345 = 22.33 fps
108 MHz at 3x37.25 and ISA at 12.4 MHz, 4.0 V, then DOOM = 3238 = 23.07 fps
112.5 MHz at 3x37.5 and ISA at 12.5 MHz, 4.0 V, then DOOM = 3108 = 24.03 fps

For 108 & 112.5 MHz, I did not bother to narrow down the minimum voltage needed. It would be somewhere in the 3.7 - 4.0 V range. 90 MHz would probably make for a more ideal and longer lasting configuration anyway.

Edit3:
Tested the system with 16 MB and 32 MB to ensure DOOM was still functional at 2x45. It was. 16 MB is max for BL2/3 chips.

If we are able to get the SXL2 chips running at 90 MHz rather than 80 MHz, we can expect a DOOM score of about 22.5 fps, rather than 20.2 fps at 80 MHz.

Very nice.
Did you test past the Doom test ?

That 2x50 score in Doom is more or less the same to what i get here.
You had a picture of the motherboard you are running this on.
Just need a visual refresher ?

I should know soon enough if these SXL2 adaptors can put SXL2 chips on 90MHz.

There's some photos of the SiS Rabbit here:
Re: Creating a voltage regulated 386 BL3 module from existing parts
and
Re: Creating a voltage regulated 386 BL3 module from existing parts

I didn't go past DOOM because this motherboard doesn't work in Windows with the BL3 hack. Probably just 1 in 40 motherboards work with BL3 in Windows without the Evergreen unit. This is why the SXL2 interposer is the way to go in general.

I will try again the SiS Rabbit in Windows 3.11 using the GD5434. Who knows, maybe the issue with ET4000 stretches into Windows as well. It is unlikely, but I want to rule this out. I can use the Evergreen BL3 adaptor in Windows though. I don't really want to modify that interposer though.

I'm pretty sure the 90 MHz SXL2 limit will be a binning process; at some point, it doesn't matter how clean you get the signals. I hope you ordered at least a dozen SXL2 CPUs. CPUShack always has some SXL2 chips if the regular sources for cheap SXL2's run dry.

Thanks.

Btw, i just found a BIOS for UMC 82C481/2 and 82C491 chipsets that work with BL3 chips just fine.
Also, retested the 2 OPTi 495### motherboards (one of them VLB guy) with working BL3 chips and they turn-on as well.
I was so mislead recently by the dead BL3 CPU.
This feels good.

So far if a board lights up and boots in DOS with BL3 chip, Windows works as well.
Surprised you are experiencing an issue with that.

Good times.