This posting continues an earlier posting which explores my collection of SDRAM modules. In this post, we look at the larger 256 and 512 megabyte modules.
256 Megabyte Modules
Mitsubishi MH32S64APHB-6
Mitsubushi is a name you would normally associate with Japanese build, but in this case, the Mitsubishi chips are marked as Taiwan. The chips are the M2V28S30ATP-6 arranged as 4Mbits x 8-bits x 4 banks. Deceptively, the ones marked -6 are for 7.5ns 133Mhz operation, so those who were hoping for 6ns (166Mhz) will probably be disappointed. The PCB is rather tall, being full height, and while the chips are made in Taiwan, it seems the module may have been assembled in Japan. The date from the warranty label seems to suggest a November 2001 installation date.
Legend L3264P33-83AHS73C
Another one from Legend, and it looks very similar to their L1664PC3 certified module just with both sides populated. The chips used are Hynix HY57V28820HCT-H as was used in their own OEM module, with a configuration of 4Mbits x 8 bits x 4 banks per chip. This one comes complete with distributed capacitors and resistor packs, and the PCB has no provision for parity configuration and is thus a little shorter. The chip seems to be dated Week 11 of 2002.
Legend L3264PM3-723IS73C
An Infineon-based module built on a PCB with support for parity, containing distributed capacitors and resistor packs. This PCB is taller and has provisions for easier access to the top of the side latches for removal of the RAM. The chip used is an HYB39S128800CT-7.5 arranged as 4Mbits x 8 bits x 4 banks and dated Week 36 of 2001.
Nanya Technology NT256S64V8HC0G-75B
These Nanya Technology modules seemed popular with some OEMs, although Nanya is not as well known for making RAM anymore especially since Micron’s acquisition of some of their assets. This module is made with a PCB marked with PCSDRAM REV1.0, as was the earlier Micron Technology module. This seems to be no surprise that the layout is identical, with the Micron one omitting the silkscreening. Both have provision for parity which is not fitted, but this one has both sides fitted owing to the capacity. The chips are mixed, dated Week 48 and 52 of 2001, and are etched NT5SV16M8CT-75B (16Mbits x 8, so likely same as the others, with CL3 at 133Mhz and CL2 at 100Mhz operation). They appear to be made in Singapore as well.
PQI MS3856UPP
It seems this module is made by Power Quotient International, better known as PQI, as it uses PQI chips and the PCB has the PQI branding. The PCB has good design with distributed capacitors and resistor packs on data lines, and appears to be dated Week 30 of 2001. The chips themselves are dated week 35 of 2001, and are marked PQ3S168Q75 which appears to be their own remarked designation as I don’t think PQI have any fabrication capacity at all. The shape of the contacts deserves a mention, as they are rounded off at the board-end rather than being straight like other PCBs.
Pluss Technology Corporation (?)/Chipwin (?) Generic
This was a rather interesting module, as it has chips from Pluss with two ‘s’es. The company appears to have gone without a trace, but the PCB has PTC on the rear, which I’m guessing might have been Pluss Technology Corporation, and the front has the name Chipwin on it. Regardless, the chips appear to be Week 38 of 2001, and the module PCB seems dated Week 28 of 2000, and contains the necessary capacitors and resistor packs. The chips are marked ISGA8442PM-07 but no further details are available.
Generic Unknown
This is another unknown, although it’s really nothing particularly different. The PCB has provisions for parity, and is slightly taller but without any cut-outs for accessing the release handles. The chips used are the Hyundai HY57V28820HCT-H as has been seen before, and is dated Week 30 of 2001.
512 Megabyte Modules
On the whole, 512Mb modules were not that common, and were the domain of very specific requests. This was mainly because of their use of larger high density chips which were poorly supported by consumer chipsets. To my knowledge, none of the Intel consumer chipsets I had owned had even been able to recognize the 512Mb modules. Only lower-performing VIA boards managed to recognize them, and some of their chipsets were for early Athlons where DDR was also an option, and was preferred due to the better performance. As a result, my collection of 512Mb modules rarely saw any service, but were acquired in the hopes of being used.
Unbranded Rambus-lookalike (512UCUNZWMEA)
It’s interesting to think that the whole heatspreader movement started by Rambus RAM was indeed mirrored by this generic module. Whereas Rambus modules really did get hot and really did need a heatspreader, SDRAM generally didn’t. Despite this, they decided to give the modules a golden heatspreader and an RDRAM look including the placement of the rivet holes. Apparently this module has a lifetime warranty, but I have no idea who to claim that from. The chips inside and their organization are not clearly known, but is likely to be 8Mbits x 8-bits x 4 banks or similar. The part number appears to be 512UCUNZWMEA, and it seems some people are claiming that it is made by Micron but the logo doesn’t look like it. Others claim it has Micron chips, but I didn’t try to verify that.
Panram 168SD512M338
I wasn’t too sure of the branding, but judging from what I found in this manual, it looks to be a Panram-branded module. This module is made with Mosel-Vitelic Corporation V54C3256804VBT7PC chips, with the 7PC designation indicating it is rated for 143Mhz operation. The label also alludes to an impressive CAS 2 latency rating, where most modules would have CAS 3. The chips are 8Mbits x 8 bits x 4 banks in organization. This module looks well designed with a few capacitors scattered and resistor packs on the lines. From what I could find, Mosel Vitelic was an arm of ProMOS Taiwan, and closed in 2009.
Unbranded Samsung-based
This seems to be a surprise module which I have no information on whatsoever. There are no identification labels, nor resistor packs, and very few capacitors. The PCB is also size-minimised for cost reduction, and it seems quite likely that this is a generic module after all. The module is based around Samsung K4S560432D-TC75 chips, of differing date codes. The closest information I could find seems to show the RAM as being 16Mbits x 4 bits x 4 banks. This seems to be an odd arrangement compared to the other PCBs which typically use 8-bit wide RAM, although it probably could work. The designation appears to indicate the chips can operate at 133Mhz CAS 3.
Unbranded 512PAURTWNAA
With a suspicious font, not-unlike the earlier Rambus-lookalike modules, this module seems to have remarked RAM with a mixture of markings. The date codes are Week 25 of 2006. The chips themselves have the “bite” mark on the sides which does allude to them being remarked Micron memory, as does the batch code location. Each chip claims to be 64Mbits x 8 bits with no information as to their banking. On the whole, it seems that 512Mb modules are mostly asking for trouble.
Conclusion
SDRAM had a profound influence of improving memory bandwidth by moving to a more complex, synchronized memory interface. Its support was widespread, and modules were available from a large number of manufacturers, utilizing various similar PCB designs and interchanging ICs at their whim. This commodity market led to strong competition, keeping prices affordable for the consumer. With standardization by JEDEC and specifications by Intel ensuring better compatibility, and incorporation of SPD configurations, memory upgrades were mostly plug and play.
Sadly, the large number of DRAM manufacturers have dwindled over time, as many failed to continue securing a large slice of the market and eventually, became bankrupt. Compatibility issues with high density memory modules appeared towards the end of SDRAM’s lifetime, and as higher speed modules appeared, machines expecting slower modules occasionally had compatibility issues due to SPD or the organization of the memory itself. On the whole, SDRAM was very reliable and less finicky than previous standards, as it didn’t require matching multiple modules for the most part.
The legacy of SDRAM continues on in embedded applications, where SDRAM is still used in some cases due to its low prices and simplicity of implementation compared to other standards, especially where high bandwidth is not a requirement. JEDECs efforts have been successful in ensuring the same climate continued through to newer DDR-based standards which kept pace with memory bandwidth requirements of modern day computers.
1Gio modules were also made, for some i845 systems, but mostly for servers. Intel 82845 Brookdale is their fastest PC SDR SDRAM controller. If I could hoard everything, then I would have kept all which I came across, but I have no space to keep even one. I cannot find where I wrote it down, IIRC, i845 attains SDR SDRAM throughput approximately 450 Mo/s. Second place Intel PC controller is i830 Almador, IIRC 350 Mo/s.
1 Gio SO-DIMMs exist, though not ever sold commercially.
Could ‘Unbranded Rambus-lookalike (512UCUNZWMEA)’ be Micro Center house-brand whitelabelled? That M looks familiar, though I have not seen it on a slant before. Micro Center is a big box electronics store in USA. They have the usual big box store branded merchandise: CD-R/DVD-R, USB Sticks, and once even IBM PC clone brand.
I don’t think 512Mbit SDR SDRAM was ever common, and most registered DIMMs used 256Mbit ones.
Ah this takes me back! I have quite a bit of SDRAM in a box somewhere as I went through a few PCs during its hey-day. I also collected a little as the machines gradually came in for replacement, it was harder to get SDRAM through the ordinary channels so it was worth keeping some around from old machines in case you desperately needed to get one working. I’ve seen a few sticks in other electronics as well, usually printers or other devices with small computers in them.
Sparcie
Indeed, SDRAM is still quite often used in embedded systems, although soldered in rather than modular. Often when they were modular, they were in SODIMM form factor, especially in higher end printers where they took over from FP/EDO RAM. I still keep it around to make sure I can get the most out of any “donated” or visiting P2/P3 era machine, as they make good platforms for demoing older hardware – not too old for ISA, not too new to entirely fail at MS-DOS and Windows 95/98/98SE driver support, and not too new to lose things like hardware serial and parallel ports, but maybe lucky enough to have onboard USB/Ethernet or be amenable to taking a PCI-USB/Ethernet card for shuttling data back and forth.
If I had known that one day I would be running a site and cataloging my collections, I wouldn’t have disposed of the older, slower sticks as they are historically relevant as reference … but I was too quick and too practically minded to notice. It’s always a regret of mine.
– Gough
Yeah most of the ones I’ve seen are in the SO-DIMM form factor as well, but I’ve seen the larger ones as well. Usually on the print server add-in card.
I completely understand about disposing of stuff, I’ve regretted a few disposals as well. It’s always difficult to decide what to keep and what to get rid of, it’s unfortunately necessary when in a limited space.
Sparcie
In fact, the main difference between Intel’s 440BX and 440GX chipsets (released back in 1998) is support for the 256Mbit SDRAM chips. It took until Intel’s 815 chipset in 2000 before they finally added 256Mbit SDRAM support to that line too (unfortunately still with a 512MB RAM limit).
Thanks for your insights! Very helpful.
– Gough
Another note: This article has references to “parity”. DIMMs used ECC not parity.
And yes, Intel never officially supported x4 chips on unbuffered DIMMs due to loading issues.
In fact, the 256Mbit SDRAM story is actually quite interesting. From what I can figure out, it looks like the original plan (circa 1996-1997) was to do 256Mbit DRAMs on 0.25um as a successor to 64Mbit on 0.35um. The problem is that the die was too big to fit into a standard 400 mil TSOP package, as well as being far too big to reach crossover even on 0.21um. In response, DRAM makers did 128Mbit SDRAM, which I think reached crossover once 0.18um processes became common. Of course, as you said, by the time 256Mbit SDRAM reached crossover in 2002, SDR SDRAM was almost obsolete. Trivia: I think Micron tended to be ahead on process technology during this period.
Trivia: Both the EU and US imposed high duties on Hynix in 2003 on DRAM. The DRAM fab in Eugene, Oregon was exempt. At the time they were imposed, that fab was only making 256Mbit SDRAM and DDR SDRAM chips. The modules made from Korea-made DRAM chips would be subject to the same duties as the DRAM chips themselves if they were imported standalone, and most of the market for 128Mbit SDRAM chips was in the upgrade market.