feipoa wrote on 2025-03-27, 04:08:

Here's an image of the 128 MB, 50 ns, EDO module I use on my M919. Shown is the front/back of the module. Only this module would allow for the exceptional performance and stability. Perhaps reach out to Memory Masters on eBay to see if they have any more.

The attachment Magic_M919_128MB_EDO_SIMM.JPG is no longer available

Thank you kindly for that photo, I was actually going to ask you for your "secret sauce recipe" 😀

Indeed research points to the fact the the drive current available for memory on this chipset may be comparatively limited, which probably makes a difference when pushing it's capabilities. It would only make sense then that the board would be sensitive to the exact characteristics of the memory ic's used, with "newer", higher density designs likely being favored/ less drive required

My next possible step was going to be to acquire some sticks with thin Hyundai branded chips to try as they pop up here and there, however now knowing that you use Samsung ones I'll try to hunt those down.

Preliminary searches show none anywhere...that's usually a clue to that it may be good stuff 😁

feipoa wrote on 2025-03-27, 04:08:

Here's an image of the 128 MB, 50 ns, EDO module I use on my M919. Shown is the front/back of the module. Only this module would allow for the exceptional performance and stability. Perhaps reach out to Memory Masters on eBay to see if they have any more.

The attachment Magic_M919_128MB_EDO_SIMM.JPG is no longer available

Spoke to seller, they had them in unlisted stock, so now I have 2 modules on the way. I'll report back once they show up.

I asked if they had a 64mb variant with those chips, which they did. As I'm perfectly happy with that amount on my machines I opted for those instead of the 128mb modules, as it's looking like the potential for things to work at high speeds is higher with the lower chip counts/chipset loading.

Interesting, so the 64 MB modules have the exact same 50 ns IC's? Did the seller send a photo?

feipoa wrote on 2025-03-28, 20:58:

Interesting, so the 64 MB modules have the exact same 50 ns IC's? Did the seller send a photo?

I sent them a photo of your modules and specifically asked for either identical ones, or preferably a 64mb with identical ic's citing that they have favorable characteristics for my application. They responded with that upon checking they do and will send them if I purchase the auction on which we were discussing this (a similar, but different IC equipped pair). I did not ask for a photo as I have dealt with this seller multiple times before, that and I'm too impatient to wait for more correspondence. Hopefully it works out, I kind of like the gamble I think.

Doing more tests yesterday I can confirm some oddball behavior of the 919 that I've seen someone else on here point out (possibly in the fastest 486 thread)- I decided to try a 128mb FPM stick I have (I had been ignoring FPM use on this altogether), and very interestingly- switching ram to "normal" in bios and using FPM actually consistently increased main memory speed by 1-2 mbps, along with adding 0.2 ish fps to quake. The downside was there was no chance in completing any tests while trying 2-2-2 with 1/0, only with 2/0, whereas I can pull off 1/0 SOMETIMES while using edo. So it seems like if all one can get stable is 3-2-3 (sorry sometimes I write 3-2-2) then in fact FPM is slightly faster.

Another interesting fact is that any expansion cards very much affect all of this- with the FPM 1/0 at 3-1-3 I could benchmark reliably all day with an s3 virge, but as soon as I swapped for a Matrox g450 (no windows drivers but works for benchmarks) and added a promise fasttrack 100 tx2 then I had to slow down to the trusty 3-2-3 to complete a win98 install (my ultimate test). I'm sure you've run into this.

EDIT: wanting to compare electrical characteristics of the km44v16100ak-6 chips on the 128FPM module to those of your samsung chips - I couldn't find a datasheet, but I DID find an old post of yours stating these are in fact 3.3 volt chips, and should likely not be used on these boards as there is no voltage regulator nor drop down diodes on the simm. THANK YOU for making a cautionary post about that in 2013 😁

Your comment concerning the existence of a 64 MB module, with the same IC's sparked me to message the seller as well. I have asked for a photo. I want to ensure that they are the same 50 ns 3.3 V Samsung TSOP chips.

It was probably me who pointed out that FPM increases performance speed on many 486 motherboards. However, when pushing these extreme limits, it seems we can sometimes get a more stable system with EDO. I think mkarcher explained why in some thread. On my MB-8433UUD system, which runs at 2x66, I have been slowly testing to see if using EDO results in a more stable system, albiet with slight worse performance.

I have also witnessed that adding expansion cards on a borderline stable and overclocked system can push it out of the stable boundary. It becomes a tiresome game of trying new cards. With more components, the bus gets loaded, the system has to do more, and there could be some small drop in Vcc. I am someone who likes to add oddities to each of my builds (e.g. DVD cards, MPEG cards, Gigamo drives, etc.), yet also clock them beyond their consumer limits. I think you'd get further to just use the bare essentials, but this is less interesting to me. Just insisting on an ethernet card in all my builds, for example, will limit my ISA or PCI bus speeds.

Concerning the 3.3 V memory chips, the modules above indicate that there's a 3.3 V voltage regulator, so although the voltage fed to the memory module is 5 V, the IC's are receiving 3.3 V. But there are some SIMMs which do not contain a voltage regulator (or a two series diodes used to reduce voltage). Those are the modules you should watch out for. However, even with the IC's running at 3.3 V, wouldn't we need to ensure that the IC's are 5 V I/O tolerant? If not, should the modules have a voltage level translator chip on them? Would this introduce too much delay? I feel like this was discussed once when people were developing 3.3 V 30-pin SIMMs. Unfortunately, I forgot the outcome. Or maybe it was discussed in a youtube video I watched.

feipoa wrote on 2025-03-31, 03:50:

Your comment concerning the existence of a 64 MB module, with the same IC's sparked me to message the seller as well. I have ask […]
Show full quote

Your comment concerning the existence of a 64 MB module, with the same IC's sparked me to message the seller as well. I have asked for a photo. I want to ensure that they are the same 50 ns 3.3 V Samsung TSOP chips.

It was probably me who pointed out that FPM increases performance speed on many 486 motherboards. However, when pushing these extreme limits, it seems we can sometimes get a more stable system with EDO. I think mkarcher explained why in some thread. On my MB-8433UUD system, which runs at 2x66, I have been slowly testing to see if using EDO results in a more stable system, albiet with slight worse performance.

I have also witnessed that adding expansion cards on a borderline stable and overclocked system can push it out of the stable boundary. It becomes a tiresome game of trying new cards. With more components, the bus gets loaded, the system has to do more, and there could be some small drop in Vcc. I am someone who likes to add oddities to each of my builds (e.g. DVD cards, MPEG cards, Gigamo drives, etc.), yet also clock them beyond their consumer limits. I think you'd get further to just use the bare essentials, but this is less interesting to me. Just insisting on an ethernet card in all my builds, for example, will limit my ISA or PCI bus speeds.

Concerning the 3.3 V memory chips, the modules above indicate that there's a 3.3 V voltage regulator, so although the voltage fed to the memory module is 5 V, the IC's are receiving 3.3 V. But there are some SIMMs which do not contain a voltage regulator (or a two series diodes used to reduce voltage). Those are the modules you should watch out for. However, even with the IC's running at 3.3 V, wouldn't we need to ensure that the IC's are 5 V I/O tolerant? If not, should the modules have a voltage level translator chip on them? Would this introduce too much delay? I feel like this was discussed once when people were developing 3.3 V 30-pin SIMMs. Unfortunately, I forgot the outcome. Or maybe it was discussed in a youtube video I watched.

I'm aware of the fact that it would be fine as long as long as the module has provisions to drop the voltage, however the 128mb fpm I was using does not have this. I just received a 64mb EDO one from the same seller I purchased that 128 FPM last year ((memgate)) today (I bought it just because it was super cheap). It only had a stock photo, and when I asked the seller what chips it has he said he couldn't guarantee- well guess what, again it is a module with no voltage dropdown, and again IT HAS 3.3V chips. Auction claims 3v - 5v, but the datasheet lists the top operating voltage as 3.6v. I could have known. I asked the seller who claims he's never heard of any problems with running them at 5 volts, and that I can return it if I want. It does not seem to run abnormally warm, however I don't think I'll be making it my main stick, as performance is average anyway.

I sure hope the Samsung chipped units from Memory Masters will indeed be the secret ingredient when they get here. I was going through some datasheet, and it looks like their electrical characteristics are indeed superior. For instance, all of the operating/driving currents listed are roughly 20-30 mA lower then my current best performing stick chips, which sounds promising.

I hope I didn't ruin pricing on the Samsung chipped modules for others...After my purchase I suddenly see an auction from Memory Masters with no photo titled "...EDO with Samsung chips..." ...with a price definitely higher then what I just paid...

There is also an auction for 2 64mb ones from a different seller right now which have the 5 volt version of those Samsung chips you have and no regulator (obviously). I wonder again how these would perform in comparison, maybe I will stare at datasheets a little longer. Perhaps the overall power draw of the module would be a little lower if it's not dropping voltage through the linear regulator

After some back and forth, it appears that this sell does not have the 64 MB, single-sided, 3.3v VRM, EDO modules with K4E640412D-TC50. For Samsung, they only have: KM44V16104BS-6. I will be interested to hear what modules you actually receive.

bertrammatrix, you should have your 64 MD modules now. Would you be willing to upload a photo of the module and identify which IC's are on the PCB? Thanks.

feipoa wrote on 2025-04-09, 01:12:

bertrammatrix, you should have your 64 MD modules now. Would you be willing to upload a photo of the module and identify which IC's are on the PCB? Thanks.

I should but I don't 😁 Courtesy of efficient us-Canada shipping it will only take another week.

Will update once it shows. I can already confirm it won't be what I wanted, seller claims a miscommunication with the shipping department, all he can assure is Samsung chipped modules but not that specific part (they are out)

I don't give up so easily though, I ordered some memory with 50ns tsop Siemens chips I seen elsewhere, as electrically the parameters were pretty close to the samsung chips of your ram.
I'm thinking the key here is likely the tsop package, it looks like once you get into those the characteristics are a tad better compared to the old packaging regardless of the manufacturer.

I was also noticing that TSOP's tend to perform the best when overclocking old 486 systems.