Memory Buyer’s Guide

Pages: 1 2 3

Additional Considerations

A couple of issues to take note of, which are not related to fraudulent practices, have to do with compatibility, power and performance. The more you know about your motherboard (and system in general) the better your protection against fraud and incompatibilities. You should not rely upon some ‘professional’ to have your best interests in mind, unless you have personal experience with him/her, or have been given very good references.

MEMORY SPEED – Every motherboard will have a document that specifies what speed memory to use. This is usually specified similar to ‘Use 70ns or faster’. This means that a 60ns module will work, but an 80ns will likely cause system crashes and hangs. While it is possible to mix speeds on a motherboard, all modules in a given bank should have the same speed rating. Note that a speed rating is the *minimum* speed of the chip – it may actually perform faster. PC100 memory will work fine on a 66MHz bus, so if you can purchase PC100 SDRAM for the same (or close) price as PC66, the PC100 modules would be the better buy for future upgrades (if that is a consideration). The bottom line, however, is to not pay more than is necessary for features or speed you may not need.

EDO vs. FPM – Many Pentium motherboards (and a few 486 boards) will accept EDO modules. While these were originally advertised as being significantly faster than FPM modules (see Memory Chip section), in actual practice the greatest impact on performance is the quantity and type of L2 cache is on the motherboard. Without L2 cache, memory accesses are 20% or more faster with EDO over FPM, but when at least 256k of synchronous cache is present, the performance increase is only about 1%-2%. Therefore, you should not pay much extra for EDO memory, especially since it does not cost any more to produce it than FPM. Another thing to consider is, if you want parity checking you are stuck with FPM, since EDO is difficult to find with parity (usually, it is offered with ECC).

PARITY and ECC – Parity is a simple form of error checking, while ECC (error checking & correction) is a much more advanced method. If your motherboard supports it, ECC is preferable to parity. FPM modules were made in only Parity and non-Parity flavors, while EDO could be found as Parity, ECC or non-Parity (though ECC EDO modules are very difficult to find). SDRAM modules do not use Parity at all, only ECC and non-ECC. Many motherboards that were made with Intel’s 430HX chipset did not implement ECC properly, and therefore cannot use ECC modules. Many of the motherboards made today (Slot 1 and Super Socket 7) do support ECC properly. The real question is whether you believe you need the ‘extra protection’ offered by these more expensive modules.

If you believe that your data is extremely important, and you cannot suffer any data errors whatsoever then ECC is for you. ‘Soft’ errors in the memory modules can potentially cause data to be corrupted and either stored on your hard drive, or used in subsequent calculations, unless there is some form of error checking in place. ECC will allow single bit errors to be corrected, and will stop the OS in the event of 2, 3 or 4 bit errors. If you believe that occasional crashes, possible system instability or lost data is not a problem (meaning, reboots and regeneration of data is acceptable to you), then ECC is probably not worth the extra cost.

SDRAM vs. EDO – SDRAM has now become the commodity DRAM that EDO and FPM once were. As a result, SDRAM is now as cheap, or cheaper, than the equivalent sized EDO module. For this reason, it is advisable to purchase SDRAM over EDO if your system can accept it. With SDRAM you are more likely to be able to utilize the memory for future upgrades, as opposed to EDO. Most motherboard manufacturers are no longer supporting EDO, even if the chipset does because of the faster access and larger module sizes (allowing for greater maximum memory configurations). Unfortunately, even SDRAM is likely to be replaced with a newer technology within a few years.

TIN vs. GOLD LEADS – There have been some reports of corrosion on the leads and connectors when the metals do not match. Generally speaking, you should be sure to buy modules with leads that match your connectors. COMPAQ computers are notorious for having gold connectors, so you should buy modules with gold leads. If you have tin connectors, buy tin leads. Apparently there is some type of reaction that occurs in humid environments that corrodes the metals, and can break the contact. This can cause problems from intermittent memory related errors to the machine unable to boot.

DOUBLE-BANK MODULES – Some motherboards will use two banks with the 8 and 32 MB modules so that one slot remains empty, while other motherboards may not recognize these modules at all. As a result, it is very important to check your motherboard manual before buying 8 or 32 MB modules. One very common situation arises on 486 boards with both 30-pin and 72-pin slots. Usually, with these boards if you fill up the 30-pin slots you cannot use double bank modules in the 72-pin slots. Again, check your motherboard manual to prevent buying a module that you cannot use, or that requires you remove the memory in your 30-pin slots (unless that is what you intended to do in the first place).

2 and 4 CHIP MODULES – Although these types of modules are generally no problem, some motherboards will not accept 4 chip memory (8 MB modules made with 16Mb chips), due to the use of a non-standard module design. As described in the section on memory modules, it is possible to make a single-bank 8 MB module with the 16Mb chips, but they must ’emulate’ a double bank module via a special TTL chip. There are motherboards that do not like this configuration, causing the modules to be unrecognized, or rendering the system incapable of booting. I have not heard of any incompatibilities with 2-chip memory modules.

REFRESH RATE – With EDO modules, you should try to find memory modules that have a 2K, rather than 4K refresh rate (see the section on Memory Chips for an explanation of refresh). These modules will require less power, and will be better and more reliable performers in the long haul, due to less heat buildup. They will also be able to more reliably maintain the data due to the faster refresh rate.

MIXING MEMORY – There have been numerous postings and questions on mixing memory on a motherboard. The general rule is that as long as you have the same type of memory within each bank, and the motherboard supports the type of memory you are installing, you probably won’t run into any problems. There may be exceptions to this, and as always, the final determination is based upon what your motherboard documentation states.

If your motherboard support EDO, for example, it will very likely not complain if you install the EDO in one bank and FPM in another bank. Some motherboards however will treat all of the modules as FPM in this case, while others will be able to utilize the modules properly. Some motherboards require the EDO to be installed in the first bank in order for the module to be recognized as EDO. Also, if your motherboard does not explicitly state that it supports EDO, the board may or may not be able to utilize these modules at all (meaning it may not even boot!).

Another area of controversy is mixing modules with different speeds on the motherboard. Again, generally speaking as long as you install the same speed modules within the same bank, you should not have a problem. One issue to recognize is that if the memory is slower than the motherboard requires (this is based upon the bus speed), you may need to increase your wait states (BIOS setting) to ensure the motherboard will run properly. This is because the bus ‘expects’ the data to be available on the next transfer cycle. If the memory is too slow, the board may experience lock ups, addressability errors, etc., so making the bus ‘wait’ another cycle will ensure the data is waiting when the bus makes the transfer.

Many people have posted information saying that the system will run at the speed of the ‘slowest’ module, so mixing of speeds should be avoided. While I have no hard data to prove or disprove this, it seems unlikely that this really occurs, except as outlined here. Since the speed of a chip is an ‘internal’ rating, that is, it is defined as the maximum time that will elapse from the time RAS falls until the data is available on the output lines, the bus knows nothing about how fast the chip actually is. As described in the section on memory chips, a chip that is rated at 60ns may actually perform much faster. All that matters, as far as the bus is concerned, is that the memory be available for the next data transfer. This means that all modules – regardless of their speed – will perform the same in terms of the overall performance, because it is the bus that determines how fast the data is delivered to the cache or CPU. If you bump up the wait states to accommodate slower memory, then all of the memory accesses will slow down because the bus waits the extra cycle(s). This does not mean, however, that the memory chips actually ‘slow down’ internally.


Pages: « Prev   1 2 3   Next »

Be the first to discuss this article!