What you need to know when buying memory?

Currently, the cost of memory has fallen so much that the profit of manufacturers and sellers is less than 20%. This forces them to use low-grade chips, parity generators, not really checking, chips that have already been used before, and to rebrand them when manufacturing memory boards. This article is an attempt to give memory buyers the information they need to make the right choice when purchasing system memory. It contains both technical information that may interest only "advanced users" and information from the "field of general knowledge and knowledge about nature."

The main part of this material is devoted to Dynamic RAM (DRAM), which is used today in most systems. Compared to the SRAM (Static RAM) used in the second-level cache, this is a cheaper solution, but DRAM is slightly slower due to the need to periodically update the memory content to avoid information loss. Currently, there are the following types of DRAM: Fast Page Mode (FPM) and Extended Data Out (EDO), which differ in the way they access data and interact with the central processor. More advanced and technological are Burst EDO (BEDO), Synchronous DRAM (SDRAM), Video RAM (VRAM), Window RAM (WRAM), Synchronous Graphics RAM (SGRAM) and RAMBUS RAM.

Static RAM (SRAM) and Read Only Memory (ROM) were not included in this list. SRAM does not require periodic content updates and is used in cache. ROM is mainly used to store the BIOS, where the information must be stored even when the power is turned off, which is what this type of memory allows. ROM also includes PROM, EPROM, EEPROM and FLASH ROM. EEPROM and FLASH ROM type memory are used in BIOS systems and can be updated using utilities provided by the manufacturer.

The second and third parts of this article are devoted to technical details and accepted terminology. The fourth paragraph discusses the construction of memory modules from chips and various technical solutions. The fifth chapter discusses the market of memory modules and its main players. If the technical details do not appeal to you, turn to the sixth chapter, which tells how to identify fakes with a possible higher probability. The last section is a brief overview of the entire article.

Memory Chips

DRAM memory modules are available as DIP (dual in-line package), SOJ (small outline J-lead) and TSOP (thin, small outline package). A DIP is a microcircuit with two rows of pins on both sides of the chip and these pins are soldered into small holes in the printed circuit board. Initially, DIP modules were installed directly into the motherboard. However, at present, they are used primarily in the second-level cache and are inserted into panels soldered to the motherboard. SOJs are "the same DIP, side view" because their leads are simply bent at the ends like the letter "J". Chips of the TSOP type are characterized by a small thickness and have contacts brought out on all sides. SOJ and TSOP are designed to be installed on printed circuit boards. However, some video card manufacturers mount contact pads for installing SOJ-type modules on their products.

Manufacturers of memory chips churn them out in huge quantities in large factories. When the first production lines were launched, not all the chips produced met the specifications and therefore required relabelling or even disposal (for example by rolling into asphalt :). With the improvement of technology, defects became fewer and fewer. However, due to the aging of the equipment, the amount of marriage increased again. Currently, most manufacturers are replacing technological lines.

In theory, every chip coming off the production line should be tested for reliability and speed according to the specification. However, the chip may have a lower access speed than what is written on it (work faster). For example, a 60ns chip can work as well as a 59ns – or even a 50ns chip. If the factory tests showed that the actual access speed of the chip is 61 or 69ns, then it will be labeled as a 70ns chip. Chips that have shown stable performance in all tests are class A (regardless of speed), chips with small defects will be classified as class Z, and chips with significant defects are usually destroyed.

Class A chips are the most reliable and are considered the highest quality chips. They are also the most expensive because they ensure stable operation in any conditions. Class C chips are used in devices that are not as demanding on system memory as modern computers, such as pagers, calculators and other household appliances. Some manufacturers additionally use another classification to identify chips.

Manufacturers apply marking on each microcircuit, which includes the name of the manufacturer, the configuration of the chip, the access speed and the date of manufacture. This marking is not applied to the surface, but embedded in the plastic case of the chip. The only way to remove this marking is to cut it off (with a peeler or a file :). Next, a protective coating is applied to the chip, giving it a presentable appearance. In addition, some manufacturers put a small embossed dot on the top of the chip to indicate the first pin of the chip and to identify relabeling done by hand.

Manufacturers use different markings for chips of different classes. For example, Micron uses the MT marking for Class A chips, while chips in other grades are marked as USA or Laser depending on how bad they are. Other manufacturers use the country name to label lower-end chips. Thus, you can find chips with inscriptions such as "japan", "france", "korea", etc. Seeing a chip with this marking, a savvy buyer will notice that this chip is made of non-standard, cheap materials and does not fully meet the specification. In addition, manufacturers have chips of different price categories depending on the scope of their testing. For example, the same Micron published a document indicating the existence of four price categories for their chips. Chips of the upper price category are tested thoroughly and are guaranteed to be error-free in 99.9% of cases. Chips that have not been tested for speed and reliability have the lowest price, that is, the buyer buys chips "as is" and may be out of luck. In this case, testing is up to the buyer.

Chips of different capacities (measured in megabits - 1Megabyte = 8 * 1Megabit) are produced, for example 1 Megabit (in this context, the designation Mb is exactly Megabit), 4Mb, 16Mb, 64Mb and recently appeared 256Mb. Each chip contains cells that can store from 1 to 16 bits of data. For example, a 16Mb chip can be configured as 4Mbx4, 2Mbx8 or 1Mbx16, but in any case its total capacity is 16Mb. Thus, the first marking number of some manufacturers indicates the total number of cells in the chip, and the second - the number of bits in the cell. The number of bits per cell also affects how many bits are transmitted at once when accessing it.

The cells in the chip are arranged like a two-dimensional array, they are accessed by specifying the column and row numbers. Each column contains additional circuits for signal amplification, selection and recharging. During a read operation, each selected bit is sent to the corresponding amplifier, after which it enters the input / output line. During recording, everything happens exactly the opposite.

Since DRAM cells quickly lose the data stored in them, they must be updated regularly. This is called refresh, and the number of rows updated in one cycle is refresh rate (regeneration frequency). The refresh rates of 2K and 4K are most often used. Chips with a 2K refresh rate can update more cells at once than 4K and complete the refresh process faster. Therefore, chips with a regeneration frequency of 2K consume less power. When performing a reading operation, regeneration is performed automatically, the data received on the signal amplifier is immediately written back. This algorithm allows you to reduce the number of necessary regenerations and increase the speed of operation.

Multiple control lines are used to indicate when a row and column are accessed, which address is accessed, and when data is to be sent or received. These lines are called RAS and CAS (Row Address Select - row address pointer and Column Address Select - column address pointer), address buffer and DOUT / DIN (Data Out and Data In). The RAS and CAS lines indicate when a row or column is accessed. The address buffer contains the address of the required row / column to be accessed and the DOUT / DIN lines indicate the direction of data transfer.

The operating speed of an asynchronous memory chip is measured in nanoseconds (ns). This rate indicates how quickly data becomes available from the moment a signal is received from the RAS. Currently, the main speeds of microcircuits on the market are 70, 60, 50 and 45ns. Synchronous memory (SDRAM) uses the motherboard's external frequency for wait cycles, and so its speed is measured in MHz, not nanoseconds.

FPM was almost driven out of the market EDO RAM because access to the CPU EDO RAM is faster by about 60% than the FPM.

Access to data in the FPM is as follows (example for a read operation): when a logic zero (low voltage) is applied to the RAS line, the address buffer contains the number of the row from which the data should be transferred to the signal amplifier. Then follows the same operation, but with CAS and with the column number. Next, the DOUT line is turned on, indicating that data is available. To access another column of the same row, only CAS changes (this is called Fast Page Mode). Each time CAS is turned on, DOUT is disabled, denying access to data.

EDO type memory uses the same algorithms for RAS and CAS, but the DOUT line is not turned off when CAS is turned on. This allows you to start access from the next column without waiting for the end of transmission from the current column. This allows you to increase access speed within one page and increase system performance.

BEDO (Burst EDO) was developed by Micron Technology as an attempt to further increase memory speed. After being developed, this technology never entered widespread use, since SDRAM is "cooler". FPM, EDO and BEDO are not designed for a bus speed of more than 66MHz. At the moment this is not so critical for most motherboards, but in the near future the situation should change due to the use of high bus speeds. So currently, BEDO modules are used mainly for video memory caching in professional graphics systems.

SDRAM (Synchronous DRAM) is the most promising type of memory on the market. All operations in SDRAM are synchronized with the external frequency of the system. This eliminates the need for the RAS and CAS analog signals required for asynchronous DRAM, which increases performance. In the future, SDRAM technology will allow the use of a bus frequency of up to 125MHz. This is very important for overall system performance, as the I/O bus frequency is the bottleneck for most computers, limiting the functionality of modern systems. For more detailed information on SDRAM operation, refer to the SDRAM FAQ.

Printed Circuit Boards

Modern printed circuit boards consist of several layers. Signals, power and mass are separated in different layers for protection and separation. Standard PCBs have four layers, but some memory board manufacturers (such as NEC, Samsung, Century, Unigen, and Micron) use six-layer PCBs. While there is some debate as to whether this is actually better, the theory is that the two extra layers improve the separation of the data lines, reducing the possibility of noise and signal leakage between the lines.

You should pay attention to the wiring and the material from which the printed circuit board is made. For example, a common four-layer board is made with two signal layers on the outside, power and ground on the inside. This provides easy access to signal lines, for example, during repairs. Unfortunately, such architecture is poorly protected from noises arising from outside and inside. The best configuration is the location of the signal layers between the mass and power layers, which allows you to protect yourself from external noise and prevent internal noise from adjacent modules.

Unpleasant, but the only way I know of to determine the number of layers in the PCB and layout of signal lines - contact the manufacturer.

Memory Modules

Many people think that the memory modules they purchase are made by semiconductor manufacturers such as Texas Instruments, Micron, NEC, Samsung, Toshiba, Motorola, etc., whose markings are on the chips. Sometimes this is true, but there are many memory module manufacturers who do not manufacture the chips themselves. Instead, they purchase components for the production of memory modules either from manufacturers or resellers. It happens that such collectors stick stickers on ready-made modules for their identification. Although it is not uncommon to find modules without identifying marks at all, they are made by third-party manufacturers.

Major memory module manufacturers have contracts with chip manufacturers to source high-quality Class A chips. Usually the chip manufacturer's name remains, but some memory module manufacturers have special arrangements whereby the chip manufacturers apply their markings instead of their own. This is a factory relabeling, it does not affect the quality of the chip in any way.

Memory modules can be SIPP (Single In-line Pin Package), SIMM (Single In-line Memory Module), DIMM (Dual In-line Memory Module) or SO DIMM (Small Outline DIMM). The most consumed modules today are DIMM and SIMM. SO DIMM is more commonly used in laptops. The conclusions (contacts) of the memory modules can be gold-plated or tin-plated, depending on the material from which the memory slot is made. For better compatibility, you should try to use memory modules and slots covered with the same material.

There are two types of SIMM modules: 30-pin or 72-pin, depending on the number of module pins. 30-pin modules are 9 bits wide (8 bits and a parity bit), and 72-pin modules are 32 bits (without parity) or 36 bits (with parity). Since the 386 and 486 processors have a 32-bit bus, either 4 30-pin SIMM modules or one 72-pin module must be used. Systems based on Pentium, Pentium Pro, and Pentium II processors have a 64-bit bus, which requires the use of 72-pin SIMM modules in pairs or a single DIMM that is 64 bits wide and 168 pins wide. The required number of memory modules to fill the bus is called a memory "bank".

DIMM modules are divided by power supply voltage and operating algorithm. Unbuffered modules with a supply voltage of 3.3 volts are standard for PCs, so there are practically no others on the market. Unbuffered DIMMs can contain SDRAM, BEDO, EDO, and FPM memory, have a width of 64 or 72 data bits for parity, and 72 and 80 bits for ECC. These modules differ from other positions of keys (cuts) in the contact line. If you look at the module from the front side (from the side of the chips), then the left key (boom) should be in the extreme right position, and the middle one should be in the middle position. The left key determines whether the module is buffered, and the middle key determines the supply voltage.
DIMM configuration

Buffered modules also need to have more two-way radio on all the lines of data in personal computers class PC will not apply.

The parity check consists of adding 8 significant bits of data when writing to memory and storing the result in the ninth parity check bit. During reading, the significant bits are added again and the result is compared with the parity check bit saved during writing. If the results match, it is considered that the data has not changed and their integrity is confirmed. This type of check can find, but not correct, a single-bit error. However, an error in two bits will remain unnoticed.

Taking into account modern memory production technologies, a parity error occurs approximately once every ten years for any module, but if it has already occurred, then the consequences can be any. Depending on the type of applications, parity control may not be a concern. For banking, military, and similar applications where data integrity is one of the necessary conditions, parity control is required, but most ordinary users do not need it.

A better level of data validation is achieved by applying ECC (error-correcting code), which uses 7 or 8 bits of parity control (depending on the width of the processor bus, 4 or 8 bytes, respectively). This allows not only to find an error in one bit, but also to correct it, as well as to find errors in 2, 3 and even 4 bits. Experience shows that 98% of errors occur in a single bit, so this level of parity control is acceptable for most data integrity-critical applications.

30-pin modules can be labeled as 1x9 or 4x9, corresponding to the number of bits that can be transferred simultaneously (including the parity bit), and 72-pin modules can be labeled as 1x32 and 1x36 (for 4 megabyte supervised or unsupervised modules parity). Almost all modern motherboards support 72-pin SIMMs, both with and without parity, as well as DIMMs.

The number of chips on a module is determined by both the size of the memory chips and the capacity of the entire module. For example, 32Mb is required for a module with a capacity of 4 Megabytes (8 bits is a byte, so the number of megabits must be divided by 8). Thus, a 4 megabyte module can contain either eight 4Mb chips or two 16Mb chips. Due to the fact that new chips with a larger capacity appear, memory modules with a larger capacity (more than 32 megabytes) are also becoming available, which allow you to increase the total amount of system RAM.

Installing a large number of chips on each module can lead to overheating and failure of the entire module.

Because 72-pin SIMMs are 32-bit, the banks for these SIMMs are also 32-bit. Sometimes, using standard chips, 64-bit memory modules are produced. These modules should be designed as two-bank. For example, to obtain an 8 megabyte SIMM module (requiring 64Mb), you can use: four 16Mb chips (8x2Mb configuration) - 32-bit Single-bank module, four 16Mb chips (16x1Mb configuration) or sixteen 4Mb chips (4x1Mb configuration) - 64-bit two-bank module. Note that four 16Mb chips (4x4Mb configuration) will not work, as this module will only use 16 bits of data, and if you use eight such chips, you will get a Single Bank 16 megabyte SIMM module. But sixteen 4Mb chips in a 1x4Mb configuration will also not work in a 4 megabyte module. Unfortunately, single-bank 8MB SIMMs require TTL circuitry to emulate two banks, which is not supported by some motherboards - so some 4-chip SIMMs sometimes do not work.

As a result, all modules 8MB SIMM (as well as a 32-megabyte) or is dvobankovyh or emulate this dvobankovyh to comply with the standard.

DIMMs are nothing more than a form factor, and in itself the question of whether they are better or worse than SIMMs is incorrect. The only known advantage of the 168-pin DIMM module is that they can be installed one at a time in a Pentium board, while SIMM modules are placed in pairs. It is obvious that this dignity is extremely insignificant. However, for, say, an EDO DIMM, it is actually the only one. Another thing is that almost all DIMMs currently produced are equipped with SDRAM type memory.

Market Memory

Most of the chips (about 80%) produced by major semiconductor manufacturers are not used in their own products, but are sold in bulk to other companies under contracts that stipulate fixed prices and volumes. Since the construction of factories is not cheap and takes a lot of time, such contracts guarantee the profitability of the enterprise and protect the manufacturer from market fluctuations. The rest of the chips are sold through the distribution network.

The main manufacturers of memory modules Kingston, Century, Unigen, Simple, Advantage, etc. are purchased directly from chip manufacturers. The best manufacturers use Class A chips to guarantee the reliability of their products. Some small "left" manufacturers buy chips from either the gray market or lower-end chips from manufacturers, and may use chips removed from old or defective modules. This allows you to keep prices low, sacrificing quality and reliability.

A finished memory module requires little to produce: a printed circuit board, a few chips (and of course chip-mounting equipment), and some assembly information. The quality of the finished module is determined by the quality of the chips and the manufacturing process. Although the main part of the module is the chips, the quality, compatibility and reliability of the products are also affected by: the quality of the printed circuit board, production management and circuit wiring. The manufacturer must be careful to preserve the functionality of the chips, as the high temperature used during soldering can damage the product or reduce its reliability and performance, even leading to inconsistency in markings.

As mentioned earlier, the speed of the chip is printed on its outside, along with other data. It is usually indicated after the "-" sign or the last one or two digits in the marking. For example, a 60ns access rate might be labeled as "-6", "-06" or "-60", or something like "GM71C17400AJ6" where we are interested in the last number. The access speed shows how much time the chip needs to respond to the central processor, so it is better if it is less.

In general, the faster the bus speed, the faster the memory should be used in the system. For example, an 8MHz bus requires only 150ns modules, 33MHz only 70ns modules, and 66MHz only 60ns modules. The use of faster modules does not cause problems, but the use of slower modules can lead to errors in the operation of applications and freezes. It is desirable that the speed of all memory modules installed in the system, especially in one bank, coincide. Even if all the modules have a speed faster than the CPU needs, it does not cause any side effects. As mentioned earlier, the speed that is written on the chip indicates only the lowest speed of its operation. In theory (and in reality), a memory module could, for example, have one chip at 52ns, another at 56ns, and the third at 60ns.

Buyer

KNOW WHAT YOU'RE BUYING! First, identify the chip manufacturer and the module manufacturer. There are many chip manufacturers and it is very difficult to know them all. See a list of chip and module manufacturers. Although all the leading manufacturers make high-end chips, they also sell their C-class chips, and you can simply attack low-quality products. Therefore, find a good, well-known seller who can clearly explain where he got this memory, because we cannot offer another way to distinguish C-class memory.

Many module manufacturers offer a lifetime warranty, even on modules with lower-end chips. The reasons for this can be different, but nevertheless they benefit, because most use their systems for more than a few years and few load the computer so much that the memory is exploited in hard mode. Low-end chips (even used ones) can work satisfactorily for several years, but still have lower reliability than new Class A chips. It can take a long time to establish the fact that application errors are due to bad memory, and not with program errors.

Unfortunately, currently, it is not enough to look at the manufacturer's name on the chip. Due to the increasing competition in the memory market, there are many ways to reduce the price, mostly all of them are at the expense of quality. Knowing what to pay attention to will help reduce the likelihood of buying low-quality modules, that is, it is necessary to understand the dwarfs and icons applied to the chips.

REPEAT the product of the module. One of the popular ways to save and something to pay attention to is the so-called "remanufactured modules". As mentioned above, every manufacturer puts the date on the chip (except TI, which puts it on the PCB). This date has the form "YYYY", where YY is the year and NN is the number of the week in which the product was manufactured. For example, the inscription "9622" means that the chip was made in the 22nd week of 1996.

Keep in mind that all chips of the same configuration on the same module must have the same (or close) date and the manufacturer of all chips must be the same (sometimes this rule is broken, but not often). Data chips and parity chips can sometimes have different manufacturing dates, as they can be configured differently. However, the dates should not differ too much.

If these dates differ, then there is a high chance of running into a "remanufactured module". There is nothing wrong with this, except that it can affect reliability due to reheating of chips when resoldering (remember, temperature can damage chip reliability and speed). Also, these chips are most likely already used and you wouldn't want to buy them if it weren't for the price.

Another negative point would be to use 16-megabyte modules 4Mb chips.

Peremarkirovanyy chips.

Although you can sometimes identify rebranded chips if you keep the following in mind. As written in the manufacturing section, the new chip should have a smooth, shiny surface, and many manufacturers also put an embossed dot on this surface (this dot is small, so look better). Since the surface of the chip is sawed off during re-marking, it will look matte and not smooth. The edges of the dot will no longer be as sharp, and the reflective properties of the chip surface will also be lost. And finally, if the marking is easily erased with a nail or a knife, then this chip must be re-marked.

As mentioned earlier, some chips peremarkovuvaty the factory.

MEMORY used. Another method - selling has already been used before memory. Often people sell their old memory. The seller who bought this memory from them sells it again and makes a really good profit. This can be avoided by checking the date on the chips as mentioned above. Any year-old module has most likely already been used. Memory units sell well, so new memory can't sit on the shelf that long. Another method of determining the used memory is to look at the contacts of the module. They should not be scratched. Although, if the manufacturing date of the chips is not very old, then this module could simply be inserted into the motherboard for testing.

FAKE parity control. If you want parity-checked memory, you need to keep in mind that there is bitwise (true) and logical (none) parity. If in the first case, the parity control is really performed according to the previously described algorithm, then in the second case, the following happens: when writing, the parity bit is not written anywhere, and when reading, this bit is simply generated based on the output data. This ensures that the parity check signal is always applied to the memory controller. Thus, no control actually takes place. Creating such modules made sense when 30-pin SIMMs were used and memory chips were quite expensive (the cost of an additional chip is about 12% of the cost of the module, in the case of using 8 data chips). If parity control is not required, and the system does not support modules without parity control, then the use of logic control is a perfectly acceptable solution. Unfortunately, this practice was continued in modules with 72 contacts. Therefore, a lot of memory with fake controls has already been sold.

The meaning of such frauds is to extract additional profit from the seller. How to distinguish a real bit parity check from a logical forgery? There is a very simple way. All modules that implement real parity control have a control chip marked, among other things, with the letters "BP". This stands for "bit parity". So if this chip is not relabeled, it can always be found on the module. In addition, bearing in mind that a module with this parity control should be 8-12% more expensive due to an additional chip, if the seller offers memory with parity, the most expensive by a couple of dollars, then it is safe to say that parity control on such logical modules.

USE cheap technology.
***

SPEED MEMORY. In the manual for any motherboard, it is said which chips with which access time are recommended to be used. It is generally recommended to use chips with an access speed of 70ns or faster. This means that modules with chips with an access speed of 60ns will work fine, while using modules with 80ns chips may lead to errors and freezes. The access speed of modules in one bank must be the same, while the use of modules with different speed characteristics in different banks is allowed. But at the same time, the system will work at the speed of the slowest module.

EDO and FPM. Almost all modern motherboards, including some boards for 486 processors, allow the use of EDO RAM. EDO type memory has been advertised as significantly faster compared to FPM, but in reality this advantage is not so obvious due to the use of a fast level 2 cache on the motherboard. Without cache, the performance of systems with EDO type memory is 20% faster than systems with FPM memory, but if the size of the L2 cache is at least 256 kilobytes, this advantage does not exceed 1-2%.

SDRAM. SDRAM definitely gives a performance gain over Edo with 60ns access times, but not quite six times as much as you might think from looking at the labeling. In particular, with a system bus frequency of 66 MHz on a 430TX chipset (VX does not optimally use SDRAM), EDO 60ns memory allows you to organize serial access according to the 5-2-2-2 scheme, and SDRAM with 10ns marking - according to the 5-1-1 scheme – 1, which gives a profit of about 30%. (In fact, the gain is significantly less, since access to memory is far from always consistent, and the cache is much more important) However, when the frequency of the system bus increases (the same Intel has not yet officially announced processors operating at a higher external frequency, but obviously which is not far off) up to 100MHz SDRAM 10ns will still be able to support the 5-1-1-1 scheme, and EDO 60ns will either not work at all or will work on a much worse scheme.

TIN AGAINST GOLD. Some sources report that the material from which the contacts of the memory module and the corresponding connector are made must match. That is, when buying new memory modules, make sure that their contact pads are made of the same material as the contacts in the connector of your motherboard. This comparison can be made purely visually, as gold contacts are yellow and tin contacts are white. Obviously, this recommendation is based on the fact that in some (for example, wet) environments, oxidation reactions are possible in the collision zone of different metals. This can lead to unstable operation of the memory system and even failures.

Two-bank MEMORY. Some motherboards can use dual-bank memory modules of 8 and 32 megabytes. Therefore, it is important to make sure that your motherboard supports these modules before purchasing such modules. For example, many 486-based boards using all banks of 30-pin SIMMs cannot work with dual-bank 72-pin SIMMs.

TWO AND FOUR CHIP MODULES. Although the use of such modules does not cause problems, some motherboards may not support four-chip modules with a capacity of 8 Megabytes, consisting of 16Mb chips due to non-standard configuration. As mentioned earlier, it is possible to create single-bank 8-megabyte modules with 16Mb chips by emulating two banks using a TTL circuit. But not all motherboards support this configuration, not recognizing these modules or refusing to load. On the other hand, I have not heard of any problems with dual-chip modules.

Refresh.

MIXING DIFFERENT TYPES OF MEMORY. There are many recommendations for this problem. However, the general rule is to install memory of the same type in one bank, and not to install memory that is not supported by the motherboard. Although some exceptions are possible, following this rule will avoid any problems.

If the motherboard supports EDO, for example, and if you install FPM RAM in one bank and EDO in the other, everything will most likely work correctly, although it is possible that EDO will work as FPM in this case. Some boards require EDO type memory to be installed in the first bank in this case. If the motherboard cannot correctly determine the type of memory installed, the system will work incorrectly or not work at all.

Another aspect of this issue is related to mixing memory modules with different access times. If you use modules of the same speed inside one bank, then there will be no problems. When using slower memory than the motherboard requires (and these requirements are based on the frequency of the system bus), it becomes necessary to add additional wait cycles when the processor works with memory. This operation is performed by changing the Setup BIOS parameters. In this case, the central processor will wait a little longer for the data to be ready. If the memory is so slow that even adding more wait cycles doesn't help, then multiple application errors and hangs are possible.

Many sources claim that all memory modules in the system run at the speed of the slowest module. However, I do not see any arguments that can confirm or refute this position. It seems unlikely to me, since the speed of the chip is an internal characteristic determined by the time from the moment the RAS signal is applied to the appearance of data at the output. Sheena has no idea how quickly they appear there. As already mentioned, the chip is marked, for example, how 60ns can work even faster. The main thing is that the memory is available before the next data transfer. This means that all modules, regardless of their speed, will work with the same overall performance, which is determined by how fast the processor or cache data will be needed. If additional wait cycles are set to apply to a slower memory system, then all memory accesses begin to perform more slowly as the bus idles for additional time. However, this does not mean that inside the chips start to work more slowly.

And another important point is the use of DIMM modules together with SIMM modules. It is not recommended to install them together due to the fact that DIMM modules are powered from 3.3 volts, and SIMMs - from 5. At the same time, most motherboards have a common power supply for SIMM and DIMM slots. In this regard, when installing modules in both types of connectors, an increased voltage of 5 volts will be applied to the DIMM. This circumstance can lead to the failure of the chips on the module. And although there are facts that the modules work together normally, using them in freelance mode, if it does not immediately cause them to fail, then it leads to a shortening of the service life.

Conclusion

It is important to understand that with memory, like everywhere else, you get what you pay for. If some memory is offered at a cheaper price, there is a good chance that it is also of lower quality. Even if a warranty is given for cheap memory, it is often not useful because problems are discovered after the expiration of its term.

Even the third memory products manufacturer with a well-proven chips can lead to a significant deterioration in their quality.

So it is better not to save - and take good memories of good firms, who also can produce it tested.

Buy memory in Uzhgorod. Order