zuldan wrote on 2025-04-17, 12:03:

feipoa wrote on 2025-04-16, 12:49:

It's been a few months. How did the new chips work out for

The stick I made didn’t work at all unfortunately. I suspect I need to use different memory chips. I did manage to find an original coast module and the motherboard detects it but nothing uses the cache PCChips M919 Cache Stick not being used [FIXED]

Have you re-tested that one you made since you finally got the 256k one to work? Or, best not touch it anymore? 😀

bertrammatrix wrote on 2025-05-09, 20:48:

zuldan wrote on 2025-04-17, 12:03:

feipoa wrote on 2025-04-16, 12:49:

It's been a few months. How did the new chips work out for

The stick I made didn’t work at all unfortunately. I suspect I need to use different memory chips. I did manage to find an original coast module and the motherboard detects it but nothing uses the cache PCChips M919 Cache Stick not being used [FIXED]

Have you re-tested that one you made since you finally got the 256k one to work? Or, best not touch it anymore? 😀

I’m not going to touch the stick that’s in their now 😉 It wouldn’t be worth testing the stick I made since the motherboard didn’t detect it at all. Either I missed soldering 1 leg on memory chip or the memory chips I used aren’t compatible.

For anyone wanting to make their own stick I highly recommend using a soldering hot plate and solder paste. It’s a lot quicker and you probably have a much better chance of getting all the legs of each memory chip to make contact with the PCB. Soldering the stick together is a b***h.

Yesturday, I assembled another one of these M919 modules, using JLCPCB with 2U gold. I used 8 ns TSOP chips, which barely fits the solder pads. Plugged in the module, but it didn't work. A had my solder paste in the cooler, so I couldn't use it until it thawed, thus, I hand soldered the chips. Problem with TSOP and hand soldering for the M919 is that the iron tip cannot make direct contact with the PCB pads (TSOP leads are too long). To hand solder, I used flux and had to press down firmly on the TSOP lead so that it would also transfer heat to the pad at the same time. It took the solder just fine. I measured each pin to ensure they were making contact.

Why did the module not work?

Some analysis revealed that there's only 86 ohms between Vcc and GND, so either an SRAM IC is bad, a capacitor is bad, or some solder/lead is pushing through the solder mask somewhere. Looking for a small variation in the VCC to GND resistance should reveal the culprit as I probe around with the DMM. As one side of the cache stick gets much warmer than the other, I already have an idea where to look. My guess is that one of the SRAM chips is bad, either from the factory, or having to heat it too long with an iron tip.

Something else I noticed: this batch of gold PCBs has a thickness of only 1.55 mm to 1.56 mm, whereas my last batch was 1.61 mm to 1.62 mm. Measurement was taken off the gold contacts. That's a 70 um variation, which is more than I hoped it would be. The cache module fits looser in the slot and the slot pins don't make as much of a mark on the module's contacts. I don't know how much of a factor PCB thickness is in module performance, but of the 7 or so modules I've assembled, the best performers were of the thicker variety.

I looked on the JLCPCB website and was stated that they allow for +-10% tolerance on PCB thickness, or 140 um. I was wondering if we should be ordering these with the thicker 2 oz copper option for the outer copper weight? Their website mentions that the thicker copper adds 35 um to the height. Multiply that by two sides, and it is 70 um. Maybe this is how we can force a slightly thicker PCB? On the down side, this option increases the cost by $30, that is, for 10 units, the price goes from approx $62 to $92 USD.

feipoa wrote on 2025-05-10, 01:17:

Yesturday, I assembled another one of these M919 modules, using JLCPCB with 2U gold. I used 8 ns TSOP chips, which barely fits t […]
Show full quote

Yesturday, I assembled another one of these M919 modules, using JLCPCB with 2U gold. I used 8 ns TSOP chips, which barely fits the solder pads. Plugged in the module, but it didn't work. A had my solder paste in the cooler, so I couldn't use it until it thawed, thus, I hand soldered the chips. Problem with TSOP and hand soldering for the M919 is that the iron tip cannot make direct contact with the PCB pads (TSOP leads are too long). To hand solder, I used flux and had to press down firmly on the TSOP lead so that it would also transfer heat to the pad at the same time. It took the solder just fine. I measured each pin to ensure they were making contact.

Why did the module not work?

Some analysis revealed that there's only 86 ohms between Vcc and GND, so either an SRAM IC is bad, a capacitor is bad, or some solder/lead is pushing through the solder mask somewhere. Looking for a small variation in the VCC to GND resistance should reveal the culprit as I probe around with the DMM. As one side of the cache stick gets much warmer than the other, I already have an idea where to look. My guess is that one of the SRAM chips is bad, either from the factory, or having to heat it too long with an iron tip.

Something else I noticed: this batch of gold PCBs has a thickness of only 1.55 mm to 1.56 mm, whereas my last batch was 1.61 mm to 1.62 mm. Measurement was taken off the gold contacts. That's a 70 um variation, which is more than I hoped it would be. The cache module fits looser in the slot and the slot pins don't make as much of a mark on the module's contacts. I don't know how much of a factor PCB thickness is in module performance, but of the 7 or so modules I've assembled, the best performers were of the thicker variety.

I looked on the JLCPCB website and was stated that they allow for +-10% tolerance on PCB thickness, or 140 um. I was wondering if we should be ordering these with the thicker 2 oz copper option for the outer copper weight? Their website mentions that the thicker copper adds 35 um to the height. Multiply that by two sides, and it is 70 um. Maybe this is how we can force a slightly thicker PCB? On the down side, this option increases the cost by $30, that is, for 10 units, the price goes from approx $62 to $92 USD.

I finally got that memory module, so tested and sent a message.

Long story short, it's starting to look more like the 12ns chips on my 1mb module just cant cope with faster timings at 60mhz 😒

Some can, some can't.
I have one 12 ns module that works fine at 60 MHz and 2-1-2, and another module which does not like this speed.
I have one 8 ns module that works fine at 60 Mhz and 2-1-2, and another module which does not like this speed.
I have one 10 ns module that works fine at 60 Mhz and 2-1-2, and two other modules which does not like this speed.

I guess socketing the IC's on module would make it too big?

For anyone interested, I have one of the 1024mb Cache sticks made by Pancake Puppy, if someone wants to buy it. I've had mine for a few years now and never used it. I've decided to move on a lot of vintage gear as it just takes up space here.

I am in Australia though so shipping to the US gets expensive. UK even worse.

I can't respond to messages yet 🙁 (Joined 11 years ago but haven't posted a lot, I guess). Happy to move the module for $30AUD plus $10AUD post Australia wide. International, what ever the postage costs over $10AUD would be added.

Tandy1000TL wrote on 2025-07-23, 10:15:

I can't respond to messages yet 🙁 (Joined 11 years ago but haven't posted a lot, I guess). Happy to move the module for $30AUD plus $10AUD post Australia wide. International, what ever the postage costs over $10AUD would be added.

I’ll take it. Will PM my email.

Tandy1000TL wrote on 2025-07-23, 10:15:

I can't respond to messages yet 🙁 (Joined 11 years ago but haven't posted a lot, I guess). Happy to move the module for $30AUD plus $10AUD post Australia wide. International, what ever the postage costs over $10AUD would be added.

This has been sold now.

anyone else selling a module ? i have M919 3.48-f with no cache 😉
im living i denmark, if you know someone who sells you can also direct me to where, and i will do the rest thanks 😉

lovely community efforts 😉

regards
jan

Without knowing the history,I bought one of these infamous M919's. Mine has a IBM (Cyrix) DX4-100. Had I known the history, I'd have probably bought it anyway just for the unique (to me) CPU. The wierd looking onboard cache modules caught my attention. Long story short, all kinds of googling and chatgpt landed me here. I too, would be interested in purchasing L2 cache module. I would need the 3.3v version.

Tandy1000TL wrote on 2025-07-23, 10:32:

Tandy1000TL wrote on 2025-07-23, 10:15:

I can't respond to messages yet 🙁 (Joined 11 years ago but haven't posted a lot, I guess). Happy to move the module for $30AUD plus $10AUD post Australia wide. International, what ever the postage costs over $10AUD would be added.

This has been sold now.

I was like a hawk on this thread...the one time I don't check BAM and gone 🤣🤦‍♂️ Dammit. It is what it is.

I do wish someone would do another batch of these, obviously there is a market

A little notice

Sasmung make also 1MBit SRAM with this pinout (non jedec pinout, "revolution"-Pinout)
K6R1008

But Sasmsung type doesn't work fine (why..? )on M919 Megacache. (only very slow timing is possible)
But on SOJ-DIP perform SAMSUNG SRAM fine on other 486 Motherboard.

And a notice:you can drive 3.3V CY7C1019D33 with 5V, (out of spec, but no problem 😀 )
I guess: use same chip from 5V Type CY7C1019D

I have bag with used Samsung & Cypress SRAM with this pinout.

I have here one Cachestick left (with Cypress CY7C1019D-10VXI )
I am from germany.

I tried using CY7C1019DV33-VXI (3.3V) on the original Megacache module but could not get it working. Since you say that CY7C1019D-10VXI (5V) is working, maybe I should try these 3.3V modules again. All it takes is one bad chip or an incomplete solder joint for the stick not to work. I am currently working on assembling several modules with IDT 71V124SA10Y and ISSI IS63LV1024-8T chips. I managed to find the sources of error on my bad modules, which was usually a bad IC or joint.

If I'm not mistaken, the Megacache module takes voltage from the motherboard's voltage regulator circuit, regardless of what CPU you use or how you have the CPU voltage jumpered. This means that only 3.5 V or 4.0 V gets to the Megacache module. So if you are using 5 V chips like CY7C1019D, these chips will only get 3.5 - 4.0 V, which is under their specification.

On the other hand, are you saying that you modified the Megacache PCB so that it always takes voltage from the 5 V rail? If you modified the PCB, did you host the gerbers somewhere?

I have checked SRAM with 5V CPU (in this case DX2-66 ) at frist.

CY7C1019D-33 eaten without issuses 5V (I know long time, cypress CY7C1019D-33 eaten 5V without problem )
CY7C1019D eaten 3.3V at most case, only few case is tick slower timing requiered.
Samsung K6R1008 (5V type) doesn't like both voltage (only very slow timing is possible.) Proably high energy consumption of old SRAM.

now is it clear ?

Madao wrote on 2025-10-31, 13:23:

I have checked SRAM with 5V CPU (in this case DX2-66 ) at frist. […]
Show full quote

I have checked SRAM with 5V CPU (in this case DX2-66 ) at frist.

CY7C1019D-33 eaten without issuses 5V (I know long time, cypress CY7C1019D-33 eaten 5V without problem )
CY7C1019D eaten 3.3V at most case, only few case is tick slower timing requiered.
Samsung K6R1008 (5V type) doesn't like both voltage (only very slow timing is possible.) Proably high energy consumption of old SRAM.

now is it clear ?

What Feipoa is saying is it doesn't matter what the voltage of the CPU is on the m919 motherboard - this would only affect the i/o voltage the sram chips see, and as far as I know everything should be 5 volt tolerant there. The VCC of the sram on modules on the other hand on the megacache design is actually fed power from the cpu regulator- which stays active and regulating EVEN IF you use a 5 volt cpu and jumper set the board for 5 volt vcore. So, on the 919 with a megacache module there is ALWAYS only between 3.3 - 4volts (depending on how the one regulator jumper is set) at the sram vcc

bertrammatrix wrote on 2025-10-31, 16:32:

What Feipoa is saying is it doesn't matter what the voltage of the CPU is on the m919 motherboard - this would only affect the i/o voltage the sram chips see, and as far as I know everything should be 5 volt tolerant there. The VCC of the sram on modules on the other hand on the megacache design is actually fed power from the cpu regulator- which stays active and regulating EVEN IF you use a 5 volt cpu and jumper set the board for 5 volt vcore. So, on the 919 with a megacache module there is ALWAYS only between 3.3 - 4volts (depending on how the one regulator jumper is set) at the sram vcc

Unless Madao redesigned the PCB so that it pulls 5 V from the motherboard's 5 V rail. It could be convenient to have two voltage variants of the MEGACACHE module.

Has anyone been able to run a Cyrix 5x86 at 120 MHz using the 1024K megacache module? I am finding that the motherboard will not POST w/1024K and Cx5x86-120. I tried altering the settings to 3-2-2, 2/2 ws, but same issue (no POST). On the other hand, if running at 100 MHz, there is no problem. I tried multiple cache modules with differing SRAM brands, but same result. If I install a 256K module, I can run Cx5x86-120 just fine. In general, the Cyrix 5x86 chips tend to need slower timings, especially with more SRAM and higher frequencies, but I've not run into a case whereby it wouldn't work at all at 120 MHz. It is likely related to the lack of onboard memory buffers.

Has anyone else run into this problem with 1024K and Cx5x86-120? Thanks!

feipoa wrote on 2025-11-03, 11:41:

Unless Madao redesigned the PCB so that it pulls 5 V from the motherboard's 5 V rail. It could be convenient to have two voltage […]
Show full quote

bertrammatrix wrote on 2025-10-31, 16:32:

What Feipoa is saying is it doesn't matter what the voltage of the CPU is on the m919 motherboard - this would only affect the i/o voltage the sram chips see, and as far as I know everything should be 5 volt tolerant there. The VCC of the sram on modules on the other hand on the megacache design is actually fed power from the cpu regulator- which stays active and regulating EVEN IF you use a 5 volt cpu and jumper set the board for 5 volt vcore. So, on the 919 with a megacache module there is ALWAYS only between 3.3 - 4volts (depending on how the one regulator jumper is set) at the sram vcc

Unless Madao redesigned the PCB so that it pulls 5 V from the motherboard's 5 V rail. It could be convenient to have two voltage variants of the MEGACACHE module.

Has anyone been able to run a Cyrix 5x86 at 120 MHz using the 1024K megacache module? I am finding that the motherboard will not POST w/1024K and Cx5x86-120. I tried altering the settings to 3-2-2, 2/2 ws, but same issue (no POST). On the other hand, if running at 100 MHz, there is no problem. I tried multiple cache modules with differing SRAM brands, but same result. If I install a 256K module, I can run Cx5x86-120 just fine. In general, the Cyrix 5x86 chips tend to need slower timings, especially with more SRAM and higher frequencies, but I've not run into a case whereby it wouldn't work at all at 120 MHz. It is likely related to the lack of onboard memory buffers.

Has anyone else run into this problem with 1024K and Cx5x86-120? Thanks!

Both of my 919s / cyrixes worked fine in that configuration at 120mhz, with 3-2-2-2 with 1/0 on the memory with the 1024 cache module (in conjunction with one of the top notch ram modules). I could even complete most DOS benchmarks at 3-1-1, however windows was never stable.

HOWEVER, I do remember the combination being very picky when it came to PCI cards and their position/order - ie a card with which the system posted at 3-1-1 may not allow the system to post at all if at the same settings in the "wrong " slot. This problem seems not to be present in the same setup while using the AMD 5x86 at 180 at 2-1-1-1 - I've moved things around and even added a USB card, and haven't had any cache/memory instability pop up

My third 919 board - version 1.5 with dip sram - is a different story. I gave up on 1024kb since the availability of decent dip sram in those capacities is nil and opted to try 512 (double banked) with the Cyrix since I have multiple sets of "legit" chips in that capacity. No matter how long I played "musical chairs" with the sram I could not get post with anything but 3-2-2-2, regardless if using the 1.5's stock bios (has the above 56mb cache bug) or the better stock 3.4 b/f bios (what I normally run on this board)

This information is quite telling. Thanks. Sounds like I will need to play 'shuffle the cards' to get Cx5x86-120 working with 1024K. I'll give the shuffle a whirl before moving onto the more interesting Cx5x86-150 S1R3 configuration.

The issue with Cyrix 5x86-120 chips and the 1024K module have been resolved. It seems that IBM 5x86c-100 chips soldered onto Thinkpad interposers are not well liked by the M919. Switching to some other interposers resolved the issues at 120 MHz with 1024K cache. This is odd because every other socket 3 motherboard I tested worked well with the Thinkpad interposers.

I was able to achieve a stable M919 with 1024K and Cx5x86-120 MHz with 2-1-2 and 0/0 ws. I used 8 ns SRAM (TSOP) and 50 ns EDO (TSOP).

Note that if you need to use 1/0 ws with EDO, you can get slightly better performance if you have some FPM that runs with 1/0 ws. For example, on the M919, and (I think) MB-8433UUD when running an Am5x86-180, Cachechk v7 as follows:

1EDO 0/0 ws --- memory read speed = 66.8 mb/s
2FPM 1/0 ws --- memory read speed = 55.4 mb/s
3EDO 1/0 ws --- memory read speed = 52.6 mb/s

And when running a Cx5x86-120, Cachechk v7 as follows:

1EDO 0/0 ws --- memory read speed = 55.2 mb/s
2FPM 0/0 ws --- memory read speed = 55.2 mb/s
3FPM 1/0 ws --- memory read speed = 43.9 mb/s
4EDO 1/0 ws --- memory read speed = 41.3 mb/s