A happy customer gets a new computer, runs home that hair back. It connects… but it doesn’t work, or it works worse than expected. For a company that sells the best gaming computers in Russia, this is not even a problem, this is a real dramatic situation. And if he’s driven half the way across the biggest country in the world, honestly, it makes you break out in a cold sweat.
Have there been any similar cases specifically in our country? – for 8 years of experience and tens of thousands of computers sold, there were all sorts of things: factory defects,” extreme accuracy ” of postal services, the off position on the power supply, and just some inexplicable mysticism.
To ensure that such situations occur as rarely as possible, our technical control department, which is literally a real OTC with its own stamp and responsibility, has accumulated experience and risen to several levels.
In this article, we will share our experience. Let’s open up the backstage area of Digital Razor’s “kitchen” and tell you about the process of checking and configuring each of our computers.
And why is all this even necessary ?
We will immediately answer the key question: why check and thoroughly test the components at all, if they are only unpacked from the factory film? I collected it, turned it on, and saw the picture – well, that’s it, 99% of success is there, the remaining 1% will come out during the installation of Windows and various drivers.
To be honest, as part of the self-assembly of a computer, it is something like this. But here the law of large numbers and even Fermi’s theorem come into force: an almost impossible event becomes virtually inevitable as the scale increases.
The probability of getting into a marriage in the case of conditional 10 components, and about this number consists of a computer, is much, much more difficult than when 10,000 of them pass.
Speaking of marriage rates in modern reality, there are excellentstatisticson video cards from Mindfactory, the largest retailer in Germany. A little old, we are talking about video cards from the time of the first “superfamily” NVIDIA (~2020), but a sample of as many as 120 thousand copies.
NVIDIA video cards were returned on average in 2.11% of cases, and AMD in 3.29%. The most unreliable were the then flagships: GeForce RTX 2080 Ti (5.35%) and RX 5700 XT (3.6%). By the way, the most reliable budget solutions: Radeon RX 5500 XT (0.95%) and GeForce GTX 1660 Ti (1.2%).
What is most interesting, the statistics cover all cases that were recognized as factory defects. Not just the client returned the video card for their own reasons, but a factory defect. It is not always expressed in a completely dead map. It also takes into account partial working capacity.
Partial performance is the second and very big problem. It’s one thing when it doesn’t start at all. It’s quite different when it works, but worse than expected.
For example, the processor generally works, but for some reason one or two cores are “not ale”. Or the frequency doesn’t reach the set values. Maybe some of the PCI-Express controllers fell off, and the video card takes only 4/8 of the lines instead of the required 16. Or 2 RAM slots fell off. Or the memory controller for this particular sample turned out to be small, when loading RAM over the conditional 10 gigabytes, errors begin.
Broken memory is generally easy. It works, even starts at the specified frequencies (the XMP profile is activated), and the “blue screen” (also known as the blue screen of death, also known as BSOD) does not even come out, but there are hidden errors that affect performance. The Windows architecture allows you to forcibly keep the computer in working order, but the performance implicitly drops (for more information about this, see the request “WHEA”).
Broken memory on the video card is also generally easy. Again, everything seems to work, but in demanding games/programs, the entire memory is loaded, and errors appear on some last gigabyte. The system can eat/tolerate several errors, but sooner or later they are layered on top of each other, and then “The video driver stopped responding and was restored”.
The power supply may behave strangely. Modern video cards have increased not only the power limits, and 300+ watts is no longer surprising, but also the speed of increasing frequencies. In a fraction of a second, the frequency can increase from 400 MHz to 3 GHz, while the consumption can increase from 50 to 350 watts. Some PSUs perceive such a sharp jump as a short circuit and go into protection. Moreover, with the conditional RTX 4080 from the conditional ASUS, this problem can get out, but the conditional Palit does not get out (the power supply scheme of the core is implemented differently, different components, controllers, etc.).
And so with everything. Only case fans are highlighted. This is where there really were no problems at all.
Factory defects in the region of 2%, possible hidden defects, an extremely expensive reputation in the high-tech environment and corporate clients with an appropriate degree of responsibility simply oblige you to delve into the issue of thorough verification of everything and always.
And of course, no one canceled the human factor. This is not about the fact that with the motto “but strong” the memory bar was inserted in reverse, but about much more subtle things. The most banal thing is that a speck of dust got into the RAM slot.
PS Well, a couple of interesting cases from my personal practice of reviews: Core i5 with characteristics like Core i3; Ryzen 5, which is like Ryzen 7; in XMP-RAM profiles, not at all what is on the sticker; SSD with a volume of 1/8 of the declared (and this was not a Chinese pal); completely absent package contents of the processor cooler; all connectors did not work on the case; SSD, expensive video card, motherboard died right in the tests; the power supply unit could not withstand an overload of 10% (this is a standard test for reviewers); the power supply unit wires broke (Thermaltake BX, burn in hell); the PCI-E connector burned out from the load Power on the video card.
Customization
We skip the build stage simply because the article is not about that. Let’s assume that there are no problems in terms of physical assembly, and the computer at least turns on, shows a picture, and enters the BIOS. Then the configuration step follows.
The very first thing an employee does is update the motherboard’s BIOS. Not that this is necessary, which is more detailed inthis article, but it certainly won’t be superfluous, and in general the rule of good taste.
The video card firmware is usually not updated. Only in exceptional cases, when AMD or NVIDIA releases significant fixes for a particular model. It happens, but it’s rare.
Next, the initial BIOS setup. It includes :
- Activating the TPM security module (otherwise Windows swears, but now TPM is almost always enabled by default);
- Activation of RAM profiles (XMP for Intel and EXMO for AMD);
- Activating Windows installation over the Network;
- Setting processor power and heat package limits when needed (not for all processors);
- Setting up the fan and pump temperature/RPM relationship;
- Configuring RAID arrays, if any.
During subsequent testing, the OTC adjusts the BIOS if necessary. For example, if you have a very good temperature margin, you can reduce the fan speed and make the computer quieter. Of course, there are also reverse situations.
Next, Windows is installed. All basic programs are already in its shell :
- Video card drivers, motherboard chipset, Wi-Fi, sound, and more;
- All the necessary utilities for controlling the backlight, depending on the components;
- Proprietary SSD drive software, like Samsung Magician and Kingston Manager. If required, the disk firmware is updated immediately;
- Various additional software based on the TOR.
Testing
If the computer has passed all the procedures described above, it is likely that everything is fine with it. But to say for sure, an automatic 8-hour stress test is used.
During this testing, various popular and not-so-popular benchmarks, stress tests, and checks are randomly launched. They simulate both everyday use and heavy load, sometimes even going beyond the specifications. For example, during testing in the OCCT stress test, the video card may not just eat up the entire power limit, which is impossible in games and work applications, but also slightly exceed it.
List of test programs:
- OCCT Power-nominally a power supply test, since it loads all the components of the computer as much as possible. In fact, it can be called a general test of the computer for stability, overheating and fault tolerance;
- LinX6. 5-Checking the processor and RAM;
- AIDA64 – test of RAM performance, latency, and CPU cache;
- AIDA64-General system Report;
- Furmark-Video card stress test;
- 3DMark PCI-Express-PCI-Express bus bandwidth test (check the speed of communication with the video card, roughly speaking);
- Cinebench 15-CPU and Graphics Card Bundle test;
- Cinebench 20-checking the processor in rendering;
- Cinebench 23-CPU check in rendering (long version);
- Octan1 – checking the video card in rendering;
- Crystal Disk Mark – checking the drive speed under various loads;
- Crystal Disk Info – checking SMART and health indicators of storage devices;
- Iperf-test of network connections (Wi-Fi and Ethernet).
Auto test is “friendly” with the internal API of the listed tests/benchmarks and issues a warning if problems or inconsistencies are detected. The situations are different: fan speeds or temperatures are much higher than expected (not only the processor/video card, but also memory, power nodes – all available thermal sensors); problems with SMART indicators; memory and WHEA errors; the results are checked for compliance with the reference values.
The noise level is evaluated periodically in all modes. After passing the auto test, the logs are checked for errors and deviations from the norm.
It’s not hard to notice that there are almost duplicate tests. The technically unskilled audience probably has a question: why run the same processor several times? It would seem that 100% of the download is there – and it’s fine. The fact is that the processor core is not just a field of a million transistors. Inside, it is divided into blocks with different functionality: instruction blocks, encoding/decoding, security, computational ALU and FPU blocks, memory and power controllers, cache memory, registers, and in addition a bus for communicating with other cores and the chipset.
Tests load different blocks in different ways. The point is not to create the maximum possible load (although this is also the case), but to check as extensively as possible. It is the diverse workload that allows you to identify problems.
After that, all connectors, buttons, and the operation of the backlight are checked (including flickering and lack of synchronization). Finally, a final look at the cable management, the overall build quality, the presence of visual defects a la a scratch came out from somewhere, once again checking with the assembly sheet (whether everything corresponds to the TOR), and transferring to the packaging.
Top issues
At the top of the list is a banal marriage: the component part either does not work at all, or there are problems in load mode.
A very common problem is that the RAM does not want to work with the settings specified in the XMP profile. Rarely can the motherboard be to blame, but almost always the root of the problem is in memory.
Flickering of the backlight, lack of synchronization. Also not uncommon, treated by reinstalling the appropriate software.
If the computer restarts during testing – it is most likely a problem with the RAM or video card.
CPU throttling – stuffy case, weak cooling, or errors when installing the cooler.
Results
Such meticulous checking and testing is not to say that it is strictly mandatory, so most computer manufacturers do not do this. Long, labor-intensive, and you can do much less: a couple of popular-basic tests within half an hour. Basically, that’s what everyone does.
At Digital Razor, we have chosen a more complex, but better way to identify problems as much as possible while still “on the beach”. Of course, this does not give a 100% guarantee that problems will not appear in the future. But it guarantees that your computer is completely healthy when you leave the Digital Razor store.
In the end, with the position ” Third class is not a marriage. If it arrives under warranty – we’ll answer “you’re a fool!” you can win on the short, but at a distance it is always a loss.
