Memory Buying Considerations
KNOW WHAT YOU ARE BUYING – The first consideration is to determine who is the chip manufacturer and who is the module manufacturer. There are a number of chip manufacturers, and it is difficult to know all of them. A list of known chip and module manufacturers listed at the end of this article (though there are certainly many others not listed). While all of the major chip manufacturers produce high-quality chips, they do sell their C-grade chips, and some may find their way to lesser-quality memory manufacturers, so it is important to buy your memory from a dealer who you can trust to get their product from known sources. There is no way to determine whether you are getting C-Grade chips without having a very good (and expensive) memory tester.
Many module manufacturers are offering lifetime warranties, even with the lower grade chips. The reasons can be varied, but essentially they are placing a pretty good bet, because most people don’t keep their systems for more than a few years – and many don’t push their systems enough to really give the memory a workout. Low quality chips (and even used ones) may work well for years, but their reliability is less than new, A-grade chips. It may take you many months (or longer) to figure out that those GPFs and application errors are due to bad memory chips, rather than the software. The dealers who are offering one, three or five year warranties are likely getting their product via the gray market, or are manufacturing modules using C-grade or used chips. These days it is best to avoid these dealers, if possible, unless your major consideration is price rather than reliability.
Unfortunately, just looking at the name on the chip is not enough, these days. With the market being so competitive, there are many questionable practices being used to cut costs for the reseller – usually at the expense of the consumer. Knowing what to look for can reduce the chances of your getting one of these low-grade SIMMs. This means understanding the markings on the chips and recognizing the signs.
PC100 SDRAM COMPLIANCE – Intel has helped to confuse the marketplace with their SDRAM specifications, which have gone through at least 6 revisions. The most recent (rev 1.63) should be used for any memory modules being built today. One must understand that this specification was written to indicate how modules should be built, not the chips. One of the most often misunderstood and misquoted pieces of information from this specification is the tCLK value. Many have used this to claim that for PC100 (100MHz) operation, the chip only needs to be 10ns. The fact is that tCLK indicates the system clock speed, not the chip speed.
The chart provided in the specification shows two columns of information – one with a tCLK of 66MHz (15ns) and the other with a tCLK of 100MHz (10ns). This indicates what the module specifications must be to reliably operate at that system clock speed, it is not a recommendation of the chip’s ‘clock cycle time’ (tCK). While it is possible for a module using 10ns chips to be PC100 compliant, chances are that problems will manifest themselves. Almost all highly regarded memory and module manufacturers recommend 8ns chips to ensure proper operation at 100MHz.
CAS2 vs. CAS3 SDRAM – This really refers to CAS Latency (CL), which is expressed in clock cycles. The CL value is derived from the column access time of the SDRAM, which is the amount of time that must transpire between the time that CAS\ falls and the data actually appears on the output line. At 100MHz, this would mean a CL2 part would need to have a column access time (tCAC) of 20ns or faster. On the other hand, a CL3 part would have a tCAC time between 21ns and 30ns.
Many vendors push their CL2 modules (usually marketed as CAS2) as being faster and/or more overclockable. The fact is that most CL2 parts at 100MHz are CL3 at 125MHz, while CL3 parts at 100MHz are also CL3 at 125MHz. In addition, the difference in actual performance between a CL2 and CL3 module at 100MHz is only a few percent at best. Unless you have a specific application which requires CL2 modules, you should save your money. In fact, if the choice is between a generic CL2 module and a name-brand CL3 module for the same price, you would probably be better off choosing the name-brand CL3 module.
REMANUFACTURED MODULES – As stated above, each manufacturer puts a date stamp on the chip (except for TI, who puts it on the PCB). This date stamp will be in the form of ‘YYWW’, where ‘YY’ is the year and ‘WW’ is the week of manufacture. For example, a stamp of 9622 means the chip was manufactured in the 22nd week of 1996. This practice is being used less than when EDO was the dominant DRAM memory used, but it is still something to watch out for, especially as the SDRAM market matures.
When you look at a memory module, the dates of all chips of like configuration should be the same date (or very close), and the manufacturer should be the same for all chips, regardless of configuration (sometimes this ‘rule’ is violated, but not often). This is how every reputable memory module manufacturer makes their SIMMs. You may see that the data chips and the parity chips have different dates from each other (i.e., they are different configurations), but they should not be too far apart.
If these dates do not match, then there is an extremely good chance that the memory has been remanufactured. As long as the module is being represented as remanufactured, there is no legal or moral issue. There may, however, be a reliability issue, since the chips are typically heated to remove them from their original PCB, and must be re-soldered to a new board (remember, heat can affect the reliability and speed of the chip). In addition, these chips are typically used chips, which you may not be willing to buy regardless of price.
Another issue in regards to remanufactured modules is that the manufacturer may be using 4Mb chips to build a 16MB module, or 16Mb chips for a 64MB module. This can cause excessive heat to build up in the module during use, and can damage the chips, thereby affecting it’s reliability, or rendering it unusable. See the section on Memory Modules for an explanation of composite vs. non-composite modules.
REMARKED CHIPS – Another practice to be aware of is the remarking of chips. Sometimes, in order to make the manufacturer’s names and/or dates match on the module, the original markings will be sanded or etched off, and a new on stenciled on. This is extremely questionable, if not outright fraudulent. There is virtually no way to know anything about that chip once its markings have been removed, without testing it.
You can identify remarked chips sometimes by looking at them. As indicated in the section on manufacturers, a new chip should have a somewhat reflective surface due to the protective coating placed on it and many manufacturers have small recessed ‘dots’ that will have a polished surface (look close, they are very small). Since a remarked chip must be sanded or etched, the surface of the chip will be very dull, or matte, in appearance. Also, the recessed ‘dots’ will not have sharp edges, and will no longer be reflective when held under a light. Finally, if you can scratch the manufacturer’s markings off with a fingernail or knife, the chip has most certainly been remarked.
As stated previously, some chips are remarked at the factory. Just because the chip does not carry the name of a major IC manufacturer, does not mean the chip has been fraudulently remarked. By observing the above signs, you should be able to determine if the chip was remarked at the factory, or by some unscrupulous manufacturer or vendor trying to cut costs.
USED MEMORY – A practice that is used mostly at computer shows and swap meets is to sell used memory. Many times, people have traded in their old memory so they can get a price reduction on ‘new’ SIMMs. These dealers then turn around and sell the memory back to the next guy, and make a pretty good profit, to boot! You can avoid this by checking the dates on the chips, as specified previously. Any SIMM with a date more than a year old has likely been used before – and anyone who says that it may have sat on a shelf for that long is either rationalizing, lying or is simply being argumentative. Memory is just too volatile for any dealer who wants to stay in business to have sitting around for any length of time. Another way of identifying used memory is to look at the leads and see if it is ‘dimpled’ or scratched up, though if the dates are pretty recent, it may have just been put into a machine to be tested.If you are buying memory at a computer show, stick to buying from those vendors who can test each memory module as they sell it in an actual memory tester (putting it into a machine and booting is not good enough – all memory cells are NOT tested). If you have to return a module to a vendor that is many miles away, the cost and time of returning it will offset any savings you may have thought you had received.
FAKE PARITY – If you have a need (or desire) for parity memory (as opposed to ECC), be aware that there used to be quite a bit of ‘logic’ or ‘bit’ parity being sold. This is where the parity bit is not stored at write time, but is instead generated at read time. This guarantees that an ‘OK’ signal is always sent to the memory controller, so that, in effect, there is no parity checking at all. This may have made sense for 30-pin memory when memory was still expensive (and parity was about 12% more cost, since it required 9 chips instead of 8). If you didn’t want parity, you couldn’t use non-parity memory in the 30-pin slots, so the fake parity was a reasonable solution. Unfortunately, the practice has spilled over into the 72-pin market. A lot of what is sold as parity memory is, in reality, ‘logic’ parity.
My question is, if you can buy and use non-parity modules, what is the purpose of fake parity? My guess is to defraud people out of their hard-earned money. Fortunately, there is a fairly easy way to identify fake parity. On the parity chip, all manufacturers that I have checked out place a ‘BP’ stamp along with their regular marks. This stands for ‘bit parity’. As long as the chip has not been remarked, you can identify it. True parity should cost from 8-12% more than the equivalent non-parity, due to the additional chips that are necessary, so if you see someone selling parity modules for only a couple of dollars more than the non-parity – chances are they are selling fake parity.
Be the first to discuss this article!