The head of Apple said that the Chinese spy chips in Supermicro's servers are fiction. What is a parity chip and how is it different from the rest? What are the main reasons for chip failure? Can a chip break "from old age"?

Page 7 of 9

How do major producers differ from the rest?

The main difference is the presence of its own silicon production. This is impossible without significant investments, including in the development, respectively, there are no fly-by-night firms in this market.

Consequently, major who have a certain status are least interested in losing it and practice the most severe final controls for their products. As a result, the risk of acquiring a marriage by the end consumer is reduced. By the way, in the "normal" consumer market, which does not deal with anyone other than "majors" (we are not talking about a conscious choice of the consumer, the choice is made by the manufacturer of the module or the trader serving the consumer), the norm is a lifetime warranty for memory.

Brands that do not have major status do not have their own silicon production facilities, and are either specially created for the implementation of low-grade silicon (comments are superfluous here), or in the very best case the end consumer cannot be sure about the origin and quality of silicon packed in these chips (but in reality one can be sure - silicon was "which was cheaper"). That is, to say that the quality of major and non-major are as different as heaven and earth might be an exaggeration ... but there is certainly a difference.

Why give up modules assembled from "left" chips if they are cheaper and suit me in terms of quality?

First, do not give up anything that suits you in terms of quality! For this will be pure snobbery. Anyway high and low quality say, a certain brand is always a statistical concept (roughly speaking, the percentage of defective goods that reaches the buyer), so you can be lucky to buy the perfect product cheaply. Again, since memory, like most, is pure electronic devices, does not deteriorate from old age, the very fact of long flawless work is very good. Nevertheless, here are some considerations for which I myself will never buy an "nameless" memory:

1. Is the quality bar at the right height? How many times have we all heard of "Mazda" Windows 95, which glitch, freeze and crash? In my opinion, yes, it happens to them, but not at all so often as to make a problem out of it (3.11 were no better). There is a serious suspicion that all that "mazda" is not responsible for overclocked processors and suspicious motherboards, is on the conscience of memory (fortunately, she is now without parity, she herself will not complain). Yes, Windows 95 is pretty ugly, but if it is also inoperable - do not rush to say that everything is in order with your hardware.
2. If everything is in order - is the memory working in the most intense mode (for example, if it is marked at 60 ns - 66 MHz in your bus?)? Will it work?
3. The current drop in prices for memory, on the one hand, makes manufacturers keenly desire to reduce costs, not throwing the discarded silicon into the trash, but selling it to "basement" (exaggeration, of course) packers. On the other hand, for a consumer in these conditions, the difference between the cheapest and the most expensive (but similar) memory is an insignificant fraction of the price of the entire computer. Do you have to risk the machine's performance (or haul a replacement memory, for that matter) for a few dollars?
4. Liquidity issue. Many computers are subject to constant upgrade, and the "extra" parts have to be implemented. What do you think, what will they buy from you with great pleasure (and for a lot of money, as a result) - Hitachi or "50 years of the People's Liberation Army of China"?

Purely in the form of a conclusion - most of the above reasoning has and reverse side, and in fact they are not proof of the correctness of only my point of view. The motto is "acceptable functionality for minimum price"has a right to exist. So the choice is ultimately yours. See the beginning of the answer.

A table of markings used by various memory manufacturers.

							Standard
		F in "brackets" above and below
Goldstar (now LG)
	Japan, Korea, Malaysia, USA
	Japan, Singapore, China
	Korea, Portugal
	Germany, Korea, France, Taiwan, UK
Texas Instruments	USA, Japan
	Japan, Malaysia

Memory Chip Marking

The issue of reading the chip markings seems to be extremely important. Of course, if you need to figure out which memory chip you have in your hands, the surest thing that you can do is refer to the manufacturer's documentation. But usually the user does not have the relevant documentation at hand and, despite this, there are ways to try to determine, with varying degrees of accuracy, what kind of memory chip you are holding (or soldered to your existing module).

Typically, the marking of the chip carries information such as the manufacturer of the chip, the country of origin and the date of its manufacture. In addition, most often there is some "service" information (for example, it can be the code of the technological line that produced the given chip). However, the most important, of course, is information about what kind of chip we see in front of us (that is, the type of memory, organization and, as a result, capacity, access time, packaging and some other architectural and technological details). Such information is necessarily present on the chip in the form of a string, as a rule, this is the longest string of all available. This is, so to speak, the article of the product; knowing him, you can always get detailed information about it from the corresponding databook. Information about the chips currently being produced is usually available on the manufacturers' websites. Nevertheless, the question remains - what to do if there is no databook at hand (which is usually the case), and the site is unavailable / does not answer / does not contain the necessary information?

Fortunately, the vast majority of manufacturers adhere (at least to convey the organization of the chip) to more or less standard markings. Having some idea of ​​this marking, it is almost always possible to determine with a high degree of certainty what kind of chip is in front of you, without resorting to reference books. However, some background information is essential. You will find a table of markings in the above.

Example 1

An example of marking a microcircuit from Micron Tecnology.

This example gives a fairly detailed information about the decoding of the markings of memory chips, but this example is 100% true only for the specified company.

Below are some more examples of markings for memory chips. Also, these examples show how it is possible, with a minimum of reference information, to obtain at least a little information about memory microcircuits by logical conclusions.

Example 2

Before us is a chip with the inscription: HM514400CS7

Looking at the table, we find that HM is the markings used by Hitachi. After making sure that the picture (logo) also belongs to Hitachi, from the same table we can see that 51 is the code used by Hitachi for asynchronous DRAM. Those. before us is an FPM or EDO chip, but not SDRAM, since SDRAM is a synchronous memory (remember how SDRAM stands for abbreviation). Let's move on to the end of the label. The last group letters (here CS) almost always carries information about the type of packaging (the letter S, as in this case, or quite often J means, as a rule, SOJ). The very first letter in this group most often refers to the initial letters of the alphabet, because it is designed to denote a revision (that is serial number subject to change) specifications for this product. V in this case this is revision C. There is no general principle for reading information in this group, but it is not very important either (revision is informative only for very advanced specialists, and you can already see the type of packaging).

The last digit is 7. Other manufacturers could have one of the following groups of characters in its place: -7, 70, -70. It is already clear that it comes about the access time, just some of the manufacturers write it in full, while others discard one zero. As a rule, this does not cause problems with determining the access time, since the characteristic times for asynchronous DRAM are 50-150 ns. It would seem that there is a risk of confusing the old 100 ns chip, which had one zero dropped, with the modern 10 ns SDRAM, but there is still great amount features (product code, packaging, operating voltage, production time, etc.) to distinguish them from each other.

Finally, there is a group of 4 digits in the middle - 4400. It is translated as follows:

a) The last zero with overwhelming probability means that this chip belongs to the fast page type. For EDO, almost all manufacturers put a different number in its place (usually 5, see the table). If the figure that you see in this place is not zero and does not coincide with the figure declared by the manufacturer for EDO, the question requires additional study. This can be either a reserve digit for the same fast page or EDO, or an indication of a special chip architecture.

b) All zeros in front of the last digit can be ignored - they only fill the free space that could be claimed if the chips had a different organization.

c) The remaining digits at the beginning of the group under consideration are 44. They first encode the capacity of the chip, and then the number of input-output lines. In this case, it is not difficult to divide these two numbers - the capacity is 4 megabits, the bus width is 4. By simple division, we find out that we have a 1x4 chip. So, studying the article showed that we have a Hitachi fast page DRAM 1x4 SOJ 70 ns chip.

Two summaries on this matter:

General - in order to highlight the group of numbers responsible for the organization and type of the chip, you need to discard the front alphanumeric code of the manufacturer and product class, and the back - the letters responsible for revision and packaging, as well as information about the access time.

Private - for 4-Mbit chips, this group of digits is 4 in length.

Example 3

Chip with a logo in the form of the letter F marking: MB8117405B-60

MB (like the logo) is given to us by Fujitsu. B-60 - revision and 60 ns. Such access times (as well as the fact that we have SOJ in front of us) make you doubt that we have SDRAM in front of us. Therefore, the product code is 81. We are left with the numbers 17405.

The last five, both according to the table and simply as a rule, means EDO, we discard 0 in front of it. The capacity and width of the tire are in the numbers 174. The assumption that the capacity is 1 gives us a very strange tire. Dividing these numbers elsewhere, we get 17 megabits and 4 I / O lines. Better lines, but why 17 ???

The answer is that 16Mbit chips have one more parameter that distinguishes one chip from another and is called "refresh depth". Rather, any chip has this parameter, but only for sixteen-megabit chips, chips of the same organization began to be produced with different values ​​of this parameter. Without going into details, we just point out that for 16-megabit chips, the number 16 in the marking was used to transfer this parameter, so that the former 16 began to equal:

16 for 4k refresh

17 for 2k refresh

18 for 1k refresh

(similarly, for 64-Mbit chips 64 can be equal to 65 ...)

So, 174 is 16 megabits on 4 I / O lines, i.e. 4x4 chip (with 2k refresh). Fujitsu chip, 60 ns EDO.

Note also that 16-megabit chips already have 5 digits for transmitting the information that 4-megabit chips fit into 4 digits.

Example 4

Chip labeled by Toshiba TC5118165BJ-60 .

TC - Toshiba, 51 - asynchronous DRAM, BJ - SOJ revision B (or something like that - this is the least important for us), 60 ns. The remainder is 18165. We see that 5 is EDO, and 1816 is 16 megabits on the 16, 1k refresh bus, that is, a 1x16 chip.

Example 5

Chip SEC KM416C1204AJ-7.

We look at the table, we see Samsung, minus KM4, minus AJ-7, and there are 16C1204 left. Something is wrong?

If you look closely at the table, you can see that Samsung uses non-standard markings. Fortunately, it is (relatively) easy to read. The 4 at the end is EDO, zero can still be dropped. What does the two mean - I do not know, it seems that I will have to discard it too (what to do - non-standard Samsung ...). The remaining 16C1 is 1x16, where instead of x we ​​put C and swapped the multipliers. This is how most Samsung markings are read.

The other renegade - Micron - has much less logical markings (it seems that inside one class of chips everything is roughly clear, but different classes are marked according to a different principle, even for EDO there is no single digit), so the access time is easily determined, and as for the rest - you need to take the markings from the site and learn by heart

Some more examples: OKI M5116405B-60

16405 gives us 4x4 4k refresh EDO (note by the way that OKI, like some other manufacturers sometimes, omits the first two letters of the marking in this case)

LGS GM71C4403CJ60

Goldstar 60 ns. The first C must be discarded, because it means 5-volt (a 3-volt chip would have the letter V in this place, other manufacturers, as a rule, do not label 5-volt chips in any way. 4403 is a 1x4 EDO.

TI TMS417809DZ-50

17809 - 2x8 2k refresh EDO

"Stylized H" HY51V65404 TC-60

Hyundai, low voltage (V) TSOP (TC) with 65404 combination, which corresponds to 64 megabits on 4 I / O lines (i.e. 16x4) EDO. Here 65 means 4k refresh, 64 would mean 8k.

Until then, we only considered asynchronous memory with you. Let's understand a little about SDRAM labeling.

Let it be NEC D4516821G5-A10-7JF

NEC's TSOPs have hard-to-read "end" markings. Without looking at this very end, we only note that the cycle time of this chip is 10 ns (A10). The combination 16821 we are interested in consists of 16 (megabits), 8 (bus), the rest of the numbers we will have to ignore, since their meaning is not entirely clear. So this is a 2x8 SDRAM chip.

It should be noted right away that all manufacturers have the greatest inconsistency in the SDRAM marking, so it is highly recommended to consult the databook. Finally, in spite of the relative harmony of the described system, it is absolutely not noise-resistant, and each manufacturer strives to introduce more of its own interference. Above, some deviation of Goldstar has already been mentioned, we also note that for TI for 16-megabit chips 4k refresh 16 for some reason equals 26. And, say, video SOJ 256x16 is marked by almost everyone as 426x (x - fast page or EDO), i.e. .e. again 16 equals 26. Especially a lot of variety is demonstrated when marking SDRAM

My home computer already retired, has existed for many years, periodically changing the case, then the platform, then the monitor. About 4 years ago the configuration was settled - Motherboard on AM2, quad core Phenom processor 2,920, 4 gigabytes of RAM (4x1gb). Everything was fine, but some time ago I had to give 2 planks to my parents' computer, because of which the amount of memory was drastically reduced. I hardly play games, with one exception - every few months I remember about the uncompleted Skyrim. It was he, or rather, the terrible brakes when playing it, that prompted me to search budget solution Problems.

We hammer “DDR2 4gb” into the search bar on Ebay.com, sort it in ascending order of price, and we come across interesting lots: low price and the mention in the title of the AMD platform. We are surprised, in the product description it is indicated in big red letters:

1> .This Memory RAM High Density ONLY work On with AMD CPU Chipset Motherboard,
2> .Not support All Intel CPU chipset motherboard.
3> .Not compatible with Apple.IBM, DELL, HP, COMPAQ, Acer.Gateway, Emachines, Packardbell, Lenovo Computers

What is it like? The first time I hear about some special memory for AMD platform. Searching the internet yielded scanty results.
Here similar memory, but already DDR3 (the meaning is the same). Small clipping from there:

The integrated memory controller for AMD processors supports 11-bit column addressing and 16K bit page size. Standard memory controllers found on other platforms use 10-bit columns and 8Kb page size. The new OCZ memory modules help you get the most out of the integrated memory controller in AMD processors. With this organization of memory access, each 16 Kb page can contain 2048 entry points. This allows the Socket AM2 processor memory controller to stay on one page twice as long as compared to the "standard" memory controller.

We conclude: for some reason, it is more profitable for the Chinese to make just such modules, while universal ones are much more expensive. I didn't go further into the jungle of theory, I learned the most interesting thing for myself.

Okay, I have AMD platform, you can order. I had a little doubt about the volume and decided to “walk like that”. The motherboard supports 16GB as much as possible, and the memory in bulk turns out to be decently cheaper, I found the most inexpensive option(I decided not to chase 400 MHz and stopped at 333).
Sent postal service ePacket, it took 2 weeks to reach Siberia, which made me very happy. Each memory stick was in an individual plastic container, all 4 were packed in a regular white paper envelope. During the delivery process, nothing was damaged and I brought home brand new green memory sticks.


They look good, the chips on the strips are arranged in 2 rows on both sides.




The chips turned out to be Samsung, full marking K4T1G044QC-ZCE6


Look, everything is correct, DDR2-667 chip, 1 gigabit capacity. If the chips are not outright rejection, then they should work.
We plug in all 4 modules, start the computer and voila, everything started, the amount of memory at startup showed 16gb. Windows booted without blue screens, open the task manager and see a magic picture:


Already good, open AIDA64 and look at the contents of the SPD:


The memory pretends to be Kingston :) Well, it's not a pity, everything else seems to be normal.

Skyrim, beloved by me, began to work just fine, it opens and closes quickly, does not think for a long time when changing locations, there are no brakes during the game. My wife's Photoshop also took a deep breath, and in general, now you can keep at least a hundred tabs open in chrome, before I would not have dared to do this. I feel like a white person, and this is damn nice :)

Let's summarize:
The price of memory in stores just skyrocketed, DDR2, which even before the collapse of the ruble was expensive, began to cost just indecent. And here, for 3000 rubles, just a wagon of memory was purchased, for my tasks it will be enough with a large margin.
So, if you have AMD with DDR2 or DDR3 memory, you want a lot of memory for relatively little money, you can take a chance. I consider myself lucky.

A little kote

UPD
I ran the memory tests at the request in Aida, I post it here.

RAM, or random access memory, is an important part of modern computing systems, whether it be personal computers, tablets or laptops, in general, any equipment that can process information.

V desktop computers DDR3 RAM is mainly used today. The choice of manufacturers is simply enormous nowadays.

It can be AMD, Adata, Kingston, Kingmax, NCP, Patriot, Corsair, PQI, and Silicon power, and many others. Number of existing models random access memory and even more: each manufacturer produces several models. Manufacturers such as Crucial and Kingston may have dozens of them.

If in search engine you enter the query "how to choose RAM", then you will see a huge number of materials, each of which can be interpreted in different ways. This review based on the analysis of various forums and sites. These recommendations can be roughly divided into two categories: the choice of RAM for systems without overclocking and the choice of RAM for systems with overclocking.

1. The number of RAM channels must be a multiple of the number of memory modules

Let's look at an example. Dual-channel platforms include FM1, AM3, LGA 1156, LGA 1156, LGA 775, the number of modules in this case should be two or four. The three-channel platform is LGA 1366, the number of modules should be 3 or 6. The four-channel platform is LGA 2011 - the number of modules in this case should be 8 or 4. If the number of modules in the RAM is not a multiple of the number of channels ( in a two-channel system, a three-module system is used), the third module will operate in Flex mode. Flex mode was first implemented by Intel... Later it was used on AMD platforms as well. But for some reason, the opportunity to work in this mode can be limited, so the requirement regarding the multiplicity of the number of modules by the number of channels remains relevant.

2. It is necessary to use identical models of RAM modules

Using given advice, you will reduce the likelihood of errors when assembling the system and its further exploitation... Some users have been able to successfully interfere with 18MB and 256MB modules from different vendors and still successfully use EDO SIMM on a Tomato with a 486 processor from AMD. But here there is a high probability of compatibility problems, for example, the impossibility of overclocking modules on the Intel platform with the nominal frequency if they are installed on the AMD platform. This error occurs because the timings used by the manufacturer are outside the timing ranges that are available on AMD platforms. Also, errors can be associated with incorrect determination of the volume of modules, the need to put a slot with RAM in other slots after BIOS updates on the motherboard, and, if necessary, install the module contrary to the manufacturer's recommendations.

3. Which is better: whales or single modules?

When it comes to RAM with a frequency of 1333 MHz, in most cases it makes no difference whether the modules are sold individually or compiled into whales. Also modules released the largest manufacturers, such as the SEC, Hynix are sold exclusively by the piece. These manufacturers just don't make memory kits. Companies such as Silicon Power and Kingston produce modules both in kits and individually. Overlocking modules, on the other hand, are usually sold by whales. The number of modules in the kit should be a multiple of the number of RAM channels, carrier platforms for which the kit manufacturer recommends them.

4. What to give preference to - RAM with or without a heatsink?

Long gone are the days when overclocking RAM, operating at a frequency of 1600 MHz and higher, required a voltage of 1.8 V. Then the need for additional cooling of this element personal computer really existed. These modules were usually obtained by overclocking slower boards. The voltage was increased from the standard 1.5 V to 1.8 V. By introducing new technological processes, managed to improve frequency characteristics RAM.

Today, most overlocker modules use a voltage not exceeding 1.65 V. Therefore, today radiators primarily perform a purely aesthetic function, and only after that the function of cooling. If the memory module contains heatsinks with a height higher than the standard, then this can be considered as significant disadvantage, which prevents the use of these modules in systems with air supercoolers, which include Thermalright HR-02, Zalman CNPS 12X, Coolink Corator DS, Scythe Mugen 3, Noctua NH D14.

This is just a recommendation. V this issue everything will depend solely on your budget. But don't forget that the price of a 4GB DDR3 RAM module has dropped significantly. Thus, by purchasing two 4 GB modules, you can save a lot.

Today, the Windows 7 operating system has become widespread among users. The 64-bit version provides support for 16 GB of RAM. Most often, applications are implemented in the 64-bit version. This allows large amounts of RAM to be used.

Preliminary results: if the budget allocated for the purchase of RAM is very limited, it is better to focus on the first and second recommendations. If the budget allows you to purchase a large amount of RAM, then it is better to take a closer look at several modules or a set of 4 GB modules.
Let's move on to considering recommendations for different platforms... The platforms are sorted alphabetically.

Socket AM3: Which RAM to Choose?

AM3 is AMD's flagship platform for last years... The range of manufactured processors is quite wide: from processors with one AMD Sempron core, to processors with six cores - AMD Phenom II X6. AM3 is a two-channel platform, therefore it is recommended to install modules on it in pairs. To ensure the full functioning of the system without additional overclocking, it is enough to use modules with a frequency of 1333 MHz.

This frequency is officially recommended for memory modules in this platform. A little later, the AMD Dragon platform was released, in which the AMD Black Edition Memory Profiles initiative was launched, which in some way was an analogue of Intel XMP. This made it possible for the AM3 platform to realize the ability to support 1600 MHz memory. However, despite this, the initiative did not receive widespread acceptance.

If you are going to overclock the processor in the future, then you may need RAM that can work for more high frequencies... This will improve the performance of the device.

In most cases, a feature of the RAM controller does not allow the use of memory with a frequency exceeding the value of 1866 MHz. The most successful memory controller from this point of view is used in AMD Phenom II X6 processors.

Which brands and modules should be preferred? The best of all are modules with a frequency of 1600 MHz from AMD Memory, Silicon Power, Kingmax, Hynix, Value RAM, SEC. Modules from AMD Memory, Value RAM, SEC, overclocking modules are well overclocked.

Socket AM3 +: how to choose the right RAM?

This socket was first introduced to a wide audience in 2011 during the emergence of architecture AMD Bulldozer... This platform was the top one in the AMD portfolio and is represented by the flagship eight-core AMD FX-8150 processor. Unlikely this platform will be used in the nominal mode, but just in case, we present the recommended RAM modules for the nominal: 1600 MHz Hynix, SEC, ValueRam, Kingston, Kingmax. Thanks to the use of an updated dual-channel RAM controller, AMD processors FX ways to support frequencies up to 1866 MHz. The RAM is in reality capable of supporting the frequency of 2133 MHz. For overclocking, it is better to purchase the following modules: AMD Memory, SEC, G.SKILL, Kingston overclocking modules.

Socket FM1: choose RAM

One more new platform from AMD is socket FM1. This is where the AMD Fusion initiative is moving forward, which is represented by the Accelerated Processing Unit (APU) architectures. The meaning of the APU is to combine the graphic and central processing unit... Most often, APUs can be found on desktop and mobile devices... Socket FM1 was designed for desktop APUs with TDP up to 1000W. The socket supports up to two channels of RAM. The frequency of the nominal device is 1600 MHz. For overclocking, models from SEC or AMD Memory at 1600 MHz are well suited.

LGA 1155 socket: choosing RAM

To date Intel platform LGA 1155 is quite popular. The platform debuted in January 2011. Quick start prevented by a bug found in the P67 / H67 / H61 chipsets. This malfunction could lead to an error in the functioning of the SATA ports. However, today the problem has already been fixed. Today LGA 1155 is represented by a wide line of processors, ranging from Intel Celeron and ending Intel Core i7.

For overclocking, it is better to choose overclocking memory. This has been proven by studies of the effect of timings and frequencies on the performance of systems using Sandy processor Bridge.

The best choice would be memory modules that support operation at a frequency of 2133 MHz. Many manufacturers have claimed that their motherboards are capable of operating at 2400 MHz, but in reality, full system operation at that frequency is quite rare.

LGA 1156 socket: choosing RAM

If we are talking about this socket, then we can say that it is already completing its life cycle... However, boards and processors for it can still be found on sale. Therefore, you may be faced with the question of choosing RAM for this socket... Due to the peculiarities of the architecture, it becomes necessary to use only high-frequency RAM modules. The LGA 1156 platform, like all those discussed above, is dual-channel. For this platform, it is recommended to select 1600 MHz RAM modules from silicon Power, Hynix, Samsung, Kingston and Kingmax. Modules from AMD Memory, SEC and overclocking modules are well overclocked.

LGA 1366: choosing RAM

This platform is long-lived. It was introduced back in 2008. The secret of LGA 1366 lies in the use of powerful Intel processors based on the core Gulftown. Due to the peculiarities of overclocking processors, it is necessary to use high-frequency RAM modules. Intel LGA 1366 is the only three-channel platform on the market today. For this platform, it is recommended to use RAM modules from Samsung, Kingston, Silicon Power, AMD Memory with a frequency of 1600 MHz.

LGA 2011 socket: choosing RAM

This platform from Intel replaced the outdated LGA 1366. At the beginning of 2012, its cost could not be called low. The server roots of the platform are revealed by the presence of four channels. But this did not stop the manufacturer from providing overclocking kits for her. For this socket, it is recommended to use 1600 MHz modules, which work well both in nominal and when overclocked. Modules from AMD Memory and Samsung are well suited. If you need to use your computer for heavy computing tasks, take a look at the high-speed modules from Kingston.

"Spy chip", which was described in the material Boomberg, looks, according to media journalists, just like this

Last week, Bloomberg Businessweek published an in-depth article about a Chinese spy microchip that was secretly installed on the motherboards of Supermicro servers. With the help of this chip, the interested party (we read - the Chinese) was able to remotely control any server connected to the Internet. Also, Bloomberg journalists said that 30 large American companies including Apple.

Despite the server maker's denial, Supermicro is up 50%. Last week, Thursday was Supermicro's worst day on the stock exchange. In addition, questions began to be asked to companies using Supermicro servers, for example, the same Apple. The corporation has tens of millions of users who have entrusted them with their data. If Apple data centers are open to the Chinese, then the question arises - what is the company doing in order to solve the problem and what guarantees of confidentiality does it provide.

Apple sent a letter stating its position on this issue to the US Congress. Cook did not give any guarantees, he simply said that this whole story is a fiction, the company's servers are under reliable protection... He also said that employees of the corporation conducted an extensive audit of their data centers several months before the ill-fated article appeared on Bloomberg. During the check, nothing similar to spy microchips was found. In addition, the company was unable to detect software vulnerabilities.

In principle, this is not the first time a corporation has had to report on its work in the field of protecting user data. So, back in 2013 year Apple had to strain, because then Edward Snowden talked about the PRISM platform, which allows US intelligence to access any information in almost any company in the United States and some other countries.

Following the publication of Snowden's material, Apple and other tech startups began posting rebuttals written by word masters to the media. The words were carefully chosen, and the facts given in the ill-fated documents were tried to refute. But it did not work, because, although some mistakes were made in the publication of the materials, the general message was quite correct - intelligence would really get access to a large number confidential data. In other words, the companies have failed to whitewash themselves because the facts in the Snowden documents have been proven, verified, and cross-checked by numerous experts.

Now, after the "stuffing" from Bloomberg, experts and ordinary users wonder if this could all be a repetition of earlier history. That is, if Bloomberg is mistaken in particulars, is there a possibility that the basic information provided in the article is pure truth, and corporations are again trying to disown the obvious.

Apple continues to argue that a thorough investigation by its experts led to the opposite of the conclusions of the Bloomberg article. No spy chips were found, and there weren't any. "We were never checked or warned about such a danger by the representatives of the FBI." In addition, the corporation said that the FBI agents did not contact after the entire article was made public.

An animated illustration of a Bloomberg article that clearly shows the same chip. So far, no one has found him

Not only Apple had to make excuses, Amazon also built its line of behavior in about the same way. Her press office made a streamlined statement: “Neither in the past, nor now, nor in the future have we had and will not have any problems associated with the use of modified by third parties computer systems or software.

Some time after the US government commission heard representatives from technology companies, the Ministry internal security stated that there is no reason to doubt the words of company representatives. However, all suspicions have not been cleared.

True, during the time that has passed since the publication of the Bloomberg publication, none of the journalists has confirmed the facts given in the material. Usually, when something out of the ordinary happens, and some media publishes it, there are journalists from other publications who confirm the words of their colleagues. But not this time - so far no one has confirmed the validity of Bloomberg's accusations, and photos of boards with a "bug" have not appeared on the network either.

If, as Bloomberg journalists argued, the problem would have affected about 30 companies, then sooner or later the truth would come out, the covers would be torn off. So far, this has not happened, so it remains only to wait.