Nope, definitely cannot fit SOJ sockets. The PCB would need to be a good deal larger. Unfortunately, the swap around of modules will take quite a bit of effort without the sockets.

I wish I knew if these were clones. Below, you can see that on one of the SRAM chips, I whipped a little isopropanyl on it, and the wording faded quite readily. This is a little suspicious. Reference the bottom right corner, and up one IC. On the 3rd module, I'll be taping the IC tops before cleaning the flux off. It is not all that uncommon for lettering to fade from surfaces, so I wouldn't put too much weight on this, but the fading felt a bit quick for just one gentle wipe.

Below are the photos of the black module. I think it looks better on the motherboard compared to the green unit.

I have nine more black PCB's that I've notched out already. I found that this jumper notch helps with alignment when I go to place the module onto the DIP sockets.

In these photos, I used solder paste. But when the system couldn't do 1 ws at 66 MHz, I figured there was a marginal connection, and decided to hit each pin with the iron and some solder wire. It didn't help the situation though.

I assembled a third unit:

The results from this unit were even worse. Cache shows up in cachechk and works in DOS, but I couldn't even get 2 ws working at 2x66. 2-1-1-1 doesn't work at 3x40 with the 3rd assembly, but works with the 1st & 2nd assembly. Maybe there's a bad IC on the 3rd assembly.

Does anyone know of an authentic source for CY7C1009D-10VXI? Alternately, a different part number which is compatible?

The only good sample I have is the green unit, with IC's sourced on digikey from 5+ years ago.

bertrammatrix wrote on 2025-12-17, 17:37:

That's interesting. I'd assume there would be value differences, but not that big. No wonder there is a difference in achievable […]
Show full quote

feipoa wrote on 2025-12-17, 14:24:

I measured the resistance between Vcc and GND on, both, the Digikey and AliExpress CY7C1009D chips. […]
Show full quote

I measured the resistance between Vcc and GND on, both, the Digikey and AliExpress CY7C1009D chips.

On the Digikey part, Vcc-GND resistance = 120 K-ohm

On the AliExpress part, Vcc-GND resistance > 50 M-ohm

Why would there be such a stark difference on the same part number and does it suggest that the AliExpress part is fake?

I assembled a module with the AliExpress chips, and while it works, I had to increase the DRAM wait-states from 1 ws to 2 ws for it to work properly at 2x66. I could run the Digikey parts at 1 ws. I could also run my DIP stacked assembly at 1 ws. I'll assemble one more module before concluding that the IC's are no good.

That's interesting. I'd assume there would be value differences, but not that big. No wonder there is a difference in achievable performance.

It makes me wonder if these are actually relabeled or a clone of another makers IC's that would normally have those values, or, if this hints at some different manufacturing process where the different resistance (likely just one amongst other internal properties that may or may not be different) is a side effect that "normally shouldn't matter", but clearly does matter when used in our (finicky) application. Reminds me of when you buy a cheap electronics product only to find half of the filtration caps inside unpopulated because "cheaper, and it still works" - sure it does, but how well.

Since you've mentioned this I think I should grab a meter and poke around some of the Chinese dip sram I have. I have long since abandoned any plans of using it, but it still may be interesting to find out why exactly it performed as horribly as it did- no matter how much I played musical chairs with it I could never get a "good" batch put together, even though others here on vogons have. I started to doubt that any I got was just not "binned", and am thinking that for the most part it's just greatly inferior product altogether

I decided to measure an extra Cypress 32kx8 IC I have, also from digikey. CY7C199D. Resistence between Vcc and GND is 118 K-ohm, which is consistent with CY7C1009D chips from Digikey. The Aliexpress sourced CY7C1009D chips are rubbish.

feipoa wrote on 2025-12-23, 06:00:

I decided to measure an extra Cypress 32kx8 IC I have, also from digikey. CY7C199D. Resistence between Vcc and GND is 118 K-ohm, which is consistent with CY7C1009D chips from Digikey. Aliexpress sourced CY7C1009D chips are rubbish.

Any idea what is difference between 1009 and 109 ? edit: slightly wider footprint it seems.

https://eu.mouser.com/ProductDetail/Infineon- … %252BovpA%3D%3D

https://eu.mouser.com/ProductDetail/Infineon- … jMnnsNGow%3D%3D

since 109 is still being sold on mouser as well as digi-key

Yup, the 109's are the 400 mil SOJs. They would require a larger PCB. But 109's are still end-of-life, so we'd be in the same situation soon enough. Digikey lists Flip Electronics as the market place distributor for the 1009's, but they require a order quantity of 690 at $1.70 each.

I took a gamble on the 50 pieces from Aliexpress and it was a loss. I just ordered 9 pieces from another supplier on eBay to test out. If they test good, I'll attempt to return the 50 pieces to Aliexpress, although it has probably been too long.

feipoa wrote on 2025-12-17, 14:24:

I measured the resistance between Vcc and GND on, both, the Digikey and AliExpress CY7C1009D chips. […]
Show full quote

I measured the resistance between Vcc and GND on, both, the Digikey and AliExpress CY7C1009D chips.

On the Digikey part, Vcc-GND resistance = 120 K-ohm

On the AliExpress part, Vcc-GND resistance > 50 M-ohm

Why would there be such a stark difference on the same part number and does it suggest that the AliExpress part is fake?

This clearly indicates the AliExpress parts use different silicon design than the Digikey parts. If both chips claim to be the same type, one of them likely has a fake marking. What polarity did you use to measure the resistance? I assume you used a test current that is positive at Vcc and negative at GND. With a standard digital multimeter and leads in the sockets the way they are supposed to be inserted, this is the red lead at Vcc and the black lead at GND. This polarity would reverse bias ESD protection diodes (if present). Typical CMOS inputs have a diode from GND (anode) to the input (cathode) to short negative input voltages to GND) and another diode from the input (andode) to Vcc (cathode) to short overvoltage at the input to Vcc. These paths are essential for CMOS inputs, because the actual logic input is two isolated gates on two (complementary) FETs. These fets are irreversibly damaged if the gate voltage is more than 30V above Vcc or below GND. As these FETs have a very low input impedance, only a quite small amount of charge is required to reach these destructive voltages. This means that any CMOS chip with a standard input structure will show around 1.4V "diode voltage drop" if measured in reverse (i.e. red lead at GND, black lead at Vcc).

It's possible that the AliExpress chips are "true CMOS" chips (and maybe require CMOS levels at the input, i.e. a switching threshold of around 2.5V), while the Digikey chips have TTL-like inputs with a switching threshold of around 1.7V. As long as everything on that bus uses CMOS output levels, both kind of inputs are fine.

feipoa wrote on 2025-12-17, 14:24:

I assembled a module with the AliExpress chips, and while it works, I had to increase the DRAM wait-states from 1 ws to 2 ws for it to work properly at 2x66. I could run the Digikey parts at 1 ws. I could also run my DIP stacked assembly at 1 ws. I'll assemble one more module before concluding that the IC's are no good.

You already did so, and it seems to confirm the chips you got from AliExpress are not meeting the designated 10ns speed. They are likely re-marked 15ns chips by a different vendor that uses a topology that does not "leak" current from Vcc to GND. The "high resistance" does not necessarily mean the chip is bad, but it obviously means the chip is different from what the label says -- and that is bad in itself, because you no longer know what kind of chip you are working with.

Common knowledge says that "10ns ISSI 128k x 8 DIP-32 chips" bought off ebay have an expected failure rate of 10% and do not meet the 10ns claim. I expect you AliExpress lot to be similar to that experience. Having two modules that basically work is not that bad, and if the 2nd module works at 3-2-2-2 @ 40MHz, I would not call it a "broken" chip, just an exceptionally slow one, which still might be in spec for what it was originally sold as.

Yes, red DMM lead to Vcc, and black DMM lead to GND. I also tried the reverse. I measured this on an IC that was sitting in the manufacturer package, not a unit that was already soldered.

I assembled two 1024K modules from AliExpress. The second module didn't work at any timing. It failed at 33 & 40 MHz, 3-2-2-2, 1/1 ws. I decided to order from another supplier, hoping for better results.

feipoa wrote on 2025-12-25, 22:33:

Yes, red DMM lead to Vcc, and black DMM lead to GND. I also tried the reverse. I measured this on an IC that was sitting in the manufacturer package, not a unit that was already soldered.

I assembled two 1024K modules from AliExpress. The second module didn't work at any timing. It failed at 33 & 40 MHz, 3-2-2-2, 1/1 ws. I decided to order from another supplier, hoping for better results.

OK, so clearly a bad chip or a bad solder joint. Given the high quality of the solder work you've shown on the thread and the fact that the module "mostly" works (in DOS), I don't think soldering can be the issue at this point. All address and data lines are already used in DOS, but if you limit yourself to the first MB + HMA, only the low 640K of cache memory will be used. Bad bits in the top 384K will not show up until you use protected mode software (like Windows) or EMS provided by EMM386.

I think I should polish my UM8881 cache diagnostic tool to include a "cache whole test RAM" mode that works at reasonable performance which could tell you which chip is bad.

mkarcher wrote on 2025-12-26, 18:28:

I think I should polish my UM8881 cache diagnostic tool to include a "cache whole test RAM" mode that works at reasonable performance which could tell you which chip is bad.

How is that coming along?

I received my order of 9 pieces of CY7C1009D from eBay. They have a datecode of 0837. Aliexpress units had a datecode of 1927, and Digikey units had 2101. I only ordered 9 pieces due to the cost. At $3/each, they are way expensive if they are counterfeit. The Aliexpress pieces were $1/each. Some encouraging information is that the resistance from Vcc to GND was 118 K-ohm, same as that of the digikey IC's. The eBay package:

Here it is evident that the sheen of the eBay chips is quite different than the Aliexpress chips:

The chip bottoms are stamped with Philippines, like the Aliexpress chips were. Curiously, each IC has a different two-character marking, e.g. AM, AN, BQ, BV, etc. What could this be for? The bottoms:

I proceeded to remove the IC's from my 2nd black Aliexpress PCB and soldered on the eBay chips. I decided to hand solder them. I used more solder than necessary to reduce uncertainty with open bridges. Do the eBay IC's look any more authentic than the Aliexpress chips? Neither chips look identical to the Digikey chips. Assembled:

Unfortunately, as with the previous IC's on this PCB, I could not get the unit to work reliably. I started to wonder if one of the pins on the PCB wasn't soldered well, or if one of the IC's had an open bridge. It took about 2 hours, but I checked for continuity of every pin on the SRAM package to ensure it went to one of the PCB's pins. They all checked out good.

I managed to do further testing with the green DigikeyPCB and the 1st black Aliexpress PCB. I have two Cyrix 5x86 QFP S1R3 CPUs which run well at 136.5 MHz (45.5x3), namely, CPU-A and CPU-B. After extensive testing, it was determined that the 1st black Aliexpress PCB and the green Digikey PCB were, both, able to handle 2-1-1-1, 0/0 ws with CPU-A (in Win95), but not CPU-B. CPU-B was only long-term stable with the green Digikey PCB. For added perspective, CPU-A is stable at 3.60 V, while CPU-B needs 3.73 V. This is with BTB, FP_FAST, LSSER, LINBRST, MEM_BYP, and DTE_EN set to their optimal settings. BWRT needs to be disabled on S1R3 CPUs.

It is this new found stability with the Aliexpress IC's which are encouraging. Previously, it was stated the the Aliexpress IC's needed increased wait-states compared to the Digikey IC's. While this is true, that finding appears to be CPU-dependent. I have now run CPU-A with the 1st Aliexpress IC's for several hours in Windows 95, running all sorts of software at 2-1-1-1, 0/0 ws.

I will be assembling a 3rd black Aliexpress PCB.

I suspect more stable results would be achieved by injecting 5V externally or generating it on-board, rather than relying on the 5V routed through long traces on the motherboard.
My remarked DIP IS61C1024-15 (three of four are AS7C1024s) became extremely unstable even at 33MHz FSB when the 5V dropped below 4.9V. I think this is a significant factor even for a good IC at speeds above 60MHz. Supplying around 5.2V might be even better.

PS: I just bought ten CY7C1009-15VCs on AliExpress (I ordered 10VC, but received 15VC. likely used, with various markings). The VCC-GND resistance was in the 30MΩ range. I'll post a photo when I get home. (I soldered two of them to a SOJ-DIP adapter I made myself, but both failed the SRAM test with Xgceu T48, so I lost motivation and haven't been able to test anything further.)

That's a good note. I'll measure Vcc-to-GND on each module when powered. For 5V integrity, I've found that it is important for the AT connector to be in good shape. I've seen 0.2 V drops from the AT connector when they are worn out.

I can confirm, it can be hard to maintain motivation when the quality of parts is so variable. Part of the difficulty with the work on these modules is that the MB-8433UUD motherboard is SRAM fussy, even with DIPs.

EDIT: 5.07 V from the PSU and 5.00 V at the SRAM chips.

EDIT2: managed to get all 3 black PCB's working at 3x 45.5 w/cx5x86 SoR5. 2-1-1-1 & 0/0 ws.