I'm thinking I came across a 430VX board that had a 5v/3.3V jumper, so might just be automagic detection that's unique to TX.
Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.
amogus
BitWrangler wrote on 2022-11-08, 04:32:I'm thinking I came across a 430VX board that had a 5v/3.3V jumper, so might just be automagic detection that's unique to TX.
I don't see a way how auto-detection of 3.3V SIMMs could work. The presence detect encoding is highly ambiguos and missing on most SIMMs anyway, and there seems to be no other 3V indicator. Once could try to auto-detect voltage on DIMMs, assuming that you can read the SPD EEPROM at 3.3V supply voltage even for 5V EDO DIMMs (which might be possible), and one could auto-detect whether a module is installed in a SIMM or a DIMM socket and auto-switch to 3.3V supply voltage if a DIMM is inserted.
Also remember we are not talking about the voltage on the data pins. Already the HX chipset outputs something like 3.0-3.3V for high and 0-0.3V for low, which will be correctly interpreted by both 5V TTL-compatible SIMMs (they decode everything above 2.5V or even 2.0V as "high") and 3.3V CMOS-compatible SIMMs (they decode evertything above 2V as high as well). Also, the HX chipset is designed in a way that it perfectly works with SIMMs outputting 5V on the data lines. The only voltage we are talking about is the supply voltage, which is not controlled by the chipset. The question of supporting 3.3V SIMMs depends on board design, not on the chipset. The chipset is only supporting evidence. For example, SDRAM supports started to appear with the 430VX and started to have acceptable performance with the TX chipset. SDRAM DIMMs generally require 3.3V supply voltage. That's why SDRAM usually support implies that there is way to select 3.3V RAM voltage (usually using jumpers, not automagic, though). The HX chipset makes SDRAM support impossible and thus 3.3V memory support is far less likely on (consumer-level) HX-based boards.
When I got the first 430HX based PC in 1996, it was equipped with 16MB of RAM, which seemed enough at that time (that is, it had 2 8MB SIMMs). And there are plenty of 5V 8MB and 16MB SIMMs. It makes sense that 3.3V SIMMs using newer chips with smaller structure sizes were only common on 64MB and 128MB SIMMs, and those SIMMs were only affordable to server operators at that time, so it is understandable why a server board would support 3.3V memory, whereas no one cared about desktop boards supporting them.
Sphere478 wrote on 2022-11-08, 02:21:Hirsch wrote on 2022-11-07, 20:41:I don't like this because Imgur might change the URL or delete the photos in the future. Somebody reading this thread in the fut […]
RaiderOfLostVoodoo wrote on 2022-11-07, 06:52:You can upload on Imgur and post the pictures here via IMG tag.
I don't like this because Imgur might change the URL or delete the photos in the future. Somebody reading this thread in the future might miss the photos then. Uploading photos fortunatelly works again, so here they are:
20221106_183806.jpg
20221106_183817.jpgI think that the sticks survived, because the test with active cooling ran without errors and I didn't use them any more after that. I draw the conclusion from the discussion here that desktop/consumer boards always put +5V to the SIMM sockets (as mentioned by mkarcher), while server boards like Sphere478's Tyan S1564 might offer 3.3V / 5V selectable by jumper.
This means: The only way to run 3.3V only EDO SIMMs in a desktop board like the Asus P/I-P55T2P4 is a hardware mod on the mainboard. Do you agree or does anyone see another option (except from modding the EDO SIMMs)?
Perhaps a 3.3v simm interposer could be constructed?
I shal name it simmterposer
I like this idea. Might I suggest an extension:
P1 mainboards/chipsets like the discussed ones above usually require two identical (EDO) SIMMs each providing 32 bits which makes together 64 bits serving the memory controller. This might also work the other way around. A SDRAM-DIMM has 64 data bits which might be split into 2x 32 bits serving two SIMM sockets. The first SIMM socket holds a carrier PCB routing all signals from the 72-pin SIMM socket to a 168-pin DIMM socket (including the lower 32 bits). The second SIMM socket holds a smaller adapter PCB just providing the upper 32 bits by a ribbon cable to the carrier PCB. The following project gave me the idea:
Gravis Ultrasound GUS PNP/PNP PRO ram extender "for sale" questions...
I had a closer look at a Asus TX97-E mainboard which has SIMM and DIMM sockets. The manual says that you shall install either SIMM or DIMM. Having a look at the bottom of the PCB gives me the idea that two SIMM sockets are wired in parallel to one DIMM socket providing the circuit I described above:
Sphere will not rest until every interconnect in the PC is interposed upon
Hirsch wrote on 2022-11-08, 19:25:I like this idea. Might I suggest an extension:
P1 mainboards/chipsets like the discussed ones above usually require two identical (EDO) SIMMs each providing 32 bits which makes together 64 bits serving the memory controller. This might also work the other way around. A SDRAM-DIMM has 64 data bits which might be split into 2x 32 bits serving two SIMM sockets. The first SIMM socket holds a carrier PCB routing all signals from the 72-pin SIMM socket to a 168-pin DIMM socket (including the lower 32 bits).
This idea to split a 64-Bit DIMM into two 32-bit SIMMs should work. But it doesn't magically transform SDRAM into asynchronous RAM. Boards that support SDRAM usually have DIMM slots. Boards that do not support SDRAM could work with your suggested adapter, but only if you use an EDO DIMM in the interposer. That's a fringe use case in my oppion, as EDO DIMMs are way more rare than PS/2 SIMMs, so while this is technically possible, I fail to get the point of it.
I agree that EDO DIMMs are less common than SDRAM. When I search for "edo dimm" at Ebay I currently have ~400 hits. Having a brief look at the search results tells me that large modules are more common and much cheaper than large EDO SIMMs.
Hirsch wrote on 2022-11-08, 22:18:I agree that EDO DIMMs are less common than SDRAM. When I search for "edo dimm" at Ebay I currently have ~400 hits. Having a brief look at the search results tells me that large modules are more common and much cheaper than large EDO SIMMs.
Yes, sorry, I really should have checked instead of assuming from what I experienced on PCs built from custom components. It seems EDO DIMMs were in fact widely used in COMPAQ Servers (as servers often need lots of RAM, getting 64 bit per module is a more pressing issue there), in HP printers and in some series of Apple PowerMac computers. Also, EDO SODIMMS were common in laptops, which absolutely makes sense, because there appears to be no universally accepted SOSIMM standard and placing two standard SIMMs into a pentium-class laptop just uses too much space. You might think about that 2*SIMM->DIMM adapter using SODIMMs to save space - but standard-sized EDO DIMMs of large capacity are more common and thus cheaper. Your choice, should you ever actually start constructing an adapter like that. At least, actually looking at ebay offers convinced me that there is a point in making that adapter.
Hirsch wrote on 2022-11-08, 19:25:P1 mainboards/chipsets like the discussed ones above usually require two identical (EDO) SIMMs each providing 32 bits which makes together 64 bits serving the memory controller. This might also work the other way around. A SDRAM-DIMM has 64 data bits which might be split into 2x 32 bits serving two SIMM sockets. The first SIMM socket holds a carrier PCB routing all signals from the 72-pin SIMM socket to a 168-pin DIMM socket (including the lower 32 bits). The second SIMM socket holds a smaller adapter PCB just providing the upper 32 bits by a ribbon cable to the carrier PCB.
DRAM is not SDRAM. SDRAM has state machine build in, incompatible.
Hirsch wrote on 2022-11-08, 22:18:I agree that EDO DIMMs are less common than SDRAM. When I search for "edo dimm" at Ebay I currently have ~400 hits. Having a brief look at the search results tells me that large modules are more common and much cheaper than large EDO SIMMs.
you either get 16MB edo dimms, or 128/256MB Buffered ECC modules.
So you want an interposer (~20-40 end cost with shipping) to reuse $10 256MB Buffered ECC module, a module you would most likely have to mod to remove buffering and ECC.
Meanwhile 64MB 72 Pin EDO Memory SIMMs are:
64MB 72 Pin EDO Memory SIMM 5V 3V 60ns
Condition: New
Quantity: More than 10 available / 165 sold
Price: US $9.65
Hirsch wrote on 2022-11-08, 19:25:I like this idea. Might I suggest an extension: […]
Sphere478 wrote on 2022-11-08, 02:21:Hirsch wrote on 2022-11-07, 20:41:I don't like this because Imgur might change the URL or delete the photos in the future. Somebody reading this thread in the fut […]
I don't like this because Imgur might change the URL or delete the photos in the future. Somebody reading this thread in the future might miss the photos then. Uploading photos fortunatelly works again, so here they are:
20221106_183806.jpg
20221106_183817.jpgI think that the sticks survived, because the test with active cooling ran without errors and I didn't use them any more after that. I draw the conclusion from the discussion here that desktop/consumer boards always put +5V to the SIMM sockets (as mentioned by mkarcher), while server boards like Sphere478's Tyan S1564 might offer 3.3V / 5V selectable by jumper.
This means: The only way to run 3.3V only EDO SIMMs in a desktop board like the Asus P/I-P55T2P4 is a hardware mod on the mainboard. Do you agree or does anyone see another option (except from modding the EDO SIMMs)?
Perhaps a 3.3v simm interposer could be constructed?
I shal name it simmterposer
I like this idea. Might I suggest an extension:
P1 mainboards/chipsets like the discussed ones above usually require two identical (EDO) SIMMs each providing 32 bits which makes together 64 bits serving the memory controller. This might also work the other way around. A SDRAM-DIMM has 64 data bits which might be split into 2x 32 bits serving two SIMM sockets. The first SIMM socket holds a carrier PCB routing all signals from the 72-pin SIMM socket to a 168-pin DIMM socket (including the lower 32 bits). The second SIMM socket holds a smaller adapter PCB just providing the upper 32 bits by a ribbon cable to the carrier PCB. The following project gave me the idea:
Gravis Ultrasound GUS PNP/PNP PRO ram extender "for sale" questions...
I had a closer look at a Asus TX97-E mainboard which has SIMM and DIMM sockets. The manual says that you shall install either SIMM or DIMM. Having a look at the bottom of the PCB gives me the idea that two SIMM sockets are wired in parallel to one DIMM socket providing the circuit I described above:
20221108_194504.jpg
I also have had this thought obviously you can only use edo Dimms with it though not sd.
There doesn’t seem to be a actual benefit to this other than being cool.
maxtherabbit wrote on 2022-11-08, 20:50:Sphere will not rest until every interconnect in the PC is interposed upon
There is truth to this..
I agree that an EDO-DIMM adapter doesn't make sense when it is used to replace rather common sized EDO-SIMMs. When it comes to 50ns modules it is even hard to find 32MB EDO-SIMMs. Finding 128MB 50ns 5V SIMM modules seems almost impossible for me. Having such ones would be nice in order to install the max memory supported by an Intel 430HX chipset. 50ns would be nice in order to run a K6-III+ with 83MHz FSB on this board. 5V support is a must as I learned in this thread. Therefore I'm interested in this idea.
Regarding the problems mentioned:
