Upgrade the power supply. Refinement of atx power supplies - computer and peripherals - circuits - file catalog - radio circuits magazines repair moding Powerful power supply from a computer

Refinement of power supplies CODEGEN and others, JNC-like... Sasha Cherny / 04/27/2004 00:56

This article (first draft) was written for my own project, which is currently in a dying state and will be repurposed. Since I believe that the article will be useful to many people (I judge by numerous letters, including from readers of your resource), I suggest you place the second edition of this creation.

good and stable work computer depends on many factors. Last but not least, it depends on a proper and reliable power supply. Regular user First of all, he is concerned about the choice of processor, motherboard, memory and other components for his computer. Little (if any) attention is paid to the power supply. As a result, the main criterion for choosing a PSU is its cost and the declared power indicated on the label. Indeed, when the label says 300 watts, this is certainly good, and at the same time, the price of a case with a power supply unit is $ 18 - $ 20 - generally wonderful ... But not everything is so simple.

And a year, two or three years ago, the price of cases with a PSU did not change and amounted to the same $20. But what changed? That's right - the declared power. First 200 watts then 235 - 250 - 300 watts. Next year there will be 350 - 400 watts ... Has there been a revolution in the power supply structure? Nothing like this. They sell you the same PSUs only with different labels. Moreover, often a 5-year-old PSU with a declared power of 200 watts produces more than a fresh 300 watt. What can you do - cheaper and more economical. If we get a case with a PSU for $20, then how much is its real cost, taking into account transportation from China and 2-3 intermediaries in the sale? Probably $5-10. Can you imagine what parts Uncle Liao put in there for $5? And you want to normally power a computer worth $500 or more with THIS? What to do? Buying an expensive power supply for $60 - $80 is, of course, a good way out when you have money. But not the best (not everyone has money and not enough). For those who do not have extra money, but have straight arms, a bright head and a soldering iron, I propose a simple revision of Chinese PSUs in order to bring them to life.

If you look at the circuitry of branded and Chinese (no name) PSUs, you can see that they are very similar. The same standard switching circuit is used based on the KA7500 PWM chip or analogues on the TL494. What is the difference between power supplies? The difference is in the parts used, their quality and quantity. Consider a typical branded power supply:

Picture 1

It can be seen that it is quite densely packed, there are no free places and all parts are unsoldered. All filters, chokes and capacitors are present.

Now consider a typical JNC PSU with a declared power of 300 watts.

Figure 2

An incomparable example of Chinese engineering! There are no filters (instead of them there are "specially trained jumpers"), no capacitors, no chokes. In principle, without them, too, everything works - but how! The output voltage contains transistor switching noise, sudden voltage surges and a significant drawdown under various computer operating modes. What a stable job...

Due to the cheap components used, the operation of such a unit is very unreliable. The actually output safe power of such a PSU is 100-120 watts. With more power, it will simply burn out and drag half of the computer with it. How to modify the Chinese PSU to a normal state and how much power do we really need?

I would like to note that the prevailing opinion about the high power consumption of modern computers is a little wrong. Packed system unit based on Pentium 4 consumes less than 200 watts, and based on AMD ATHLON XP less than 150 watts. Thus, if we at least provide a real 200-250 watt power supply, then there will be one less weak link in our computer.

The most critical details in a PSU are:

High voltage capacitors
High voltage transistors
High Voltage Rectifier Diodes
High frequency power transformer
Low Voltage Diode Rectifier Arrays

The Chinese brothers manage to save money here too ... Instead of high-voltage capacitors 470 microfarads x 200 volts, they put 200 microfarads x 200 volts. These details affect the ability of the unit to withstand a short-term loss of mains voltage and the power of the output voltage of the PSU. They put small power transformers that get very hot at critical powers. They also save on low-voltage rectifier assemblies by replacing them with two discrete diodes soldered together. The absence of filters and smoothing capacitors has already been mentioned above.

Let's try to fix this. First of all, you need to open the PSU and estimate the size of the transformer. If it has dimensions of 3x3x3 cm or more, then it makes sense to modify the block. First you need to replace the large high-voltage capacitors and put at least 470 microfarads x 200 volts. It is necessary to put all the chokes in the low-voltage part of the PSU. Chokes can be wound on a ferrite ring with a diameter of 1-1.5 cm with copper wire with varnish insulation with a cross section of 1-2 mm 10 turns. You can also take chokes from a faulty PSU (a dead PSU can be bought at any computer shop for $1-2). Next, you need to solder the smoothing capacitors to the empty places of the low-voltage part. It is enough to put 3 capacitors 2200uF x 16 volts (Low ESR) in the circuit + 3.3v, + 5v, + 12v.

A typical view of low-voltage rectifier diodes in cheap blocks is as follows:

Figure 3

or worse, like this

Figure 4

The first diode assembly provides 10 amps at 40 volts, the second - 5 amps max. At the same time, the following data is written on the cover of the PSU:

Figure 5

20-30 amperes are declared, but 10 or 5 amperes are actually issued !!! Moreover, on the PSU board there is a place for normal assemblies, which should be there:

Figure 6

The marking shows that this is 30 amperes at 40 volts - and this is a completely different matter! These assemblies should be on the + 12v and + 5v channel. The +3.3v channel can be made in two ways: either on the same assembly, or on a transistor. If there is an assembly, then we change it to normal, if the transistor, then we leave everything as it is.

So, we run to the store or to the market and buy 2 or 3 (depending on the PSU) MOSPEC S30D40 diode assemblies (per channel +12 volts S40D60 - the last digit D - voltage - the more, the calmer the soul or F12C20C - 200 volts ) or similar in characteristics, 3 capacitors 2200 microfarads x 16 volts, 2 capacitors 470 microfarads x 200 volts. All these parts cost about $5-6.

After we changed everything, the BP will look something like this:

Figure 7

Figure 8

Overvoltage on the 12 volt channel is very harmful to hard drives. Basically, HDD heating occurs due to increased voltage (more than 12.6 volts). In order to reduce the voltage of 13 volts, it is enough to solder a powerful diode, for example KD213, into the gap of the yellow wire that feeds the HDD. As a result, the voltage will decrease by 0.6 volts and will be 11.6 volts - 12.4 volts, which is quite safe for hard drive.

As a result, we got a normal PSU capable of delivering at least 250 watts (normal, not Chinese !!) to the load, which, moreover, will become much less heated.

Warning!!! Everything that you will do with your PSU - you do at your own peril and risk! If you do not have sufficient qualifications and cannot distinguish a soldering iron from a plug, then do not read what is written here, and even more so do not do it !!!

Comprehensive noise reduction for computers

How to deal with noise? To do this, we must have the right case with a horizontal power supply (PSU). Such a case has large dimensions, but it removes excess heat much better to the outside, since the PSU is located above the processor. It makes sense to put a cooler with an 80x80 fan on the processor, for example, the Titan series. As a rule, a large fan with the same performance as a small one, runs at lower speeds and makes less noise. The next step is to lower the processor temperature during idle or light load.

As you know, most of the time the computer's processor is idle waiting for the reaction of the user or programs. At this time, the processor simply runs empty cycles in vain and heats up. Program coolers or soft coolers are designed to combat this phenomenon. Recently, these programs have even begun to be built into the BIOS of the motherboard (for example, EPOX 8KRAI) and into the operating system. Windows system xp. One of the simplest and effective programs is VCOOL. This program at work AMD processor performs the Bus disconnect procedure - disconnecting the processor bus during idle time and reducing heat generation. Since the idle processor takes 90% of the time, the cooling will be very significant.

Here we come to the understanding that we do not need the rotation of the cooler fan at full speed to cool the processor. How to lower rpm? You can take a cooler with speed control with an external regulator. Or you can use the fan speed control program - SPEEDFAN. This program is remarkable in that it allows you to adjust the fan speed depending on the temperature of the processor by setting the temperature threshold. Thus, when the computer starts, the fan has full speed, and when working in Windows with documents and the Internet, the fan speed is automatically reduced to the minimum.

The combination of VCOOL and SPEEDFAN programs allows you to stop the cooler altogether while working in Word and the Internet, and at the same time the processor temperature does not rise above 55C! (Athlon XP 1600). But the SPEEDFAN program has one drawback - it does not work on all motherboards. In this case, you can lower the fan speed by switching it to work from 12 volts to 7 or even 5 volts. Typically, the cooler is attached to the motherboard using a three-pin connector. Black wire is ground, red +12, yellow - speed sensor. In order to switch the cooler to 7 volts, you need to pull the black wire out of the connector and insert it into a free connector (red wire +5 volts) coming from the PSU, and insert the red wire from the cooler into the PSU connector with a yellow wire (+12).

Figure 9

The yellow wire from the cooler can be left in the connector and inserted into the motherboard to monitor the fan speed. Thus, we get 7 volts on the cooler (the difference between +5 and +12 volts is 7 volts). To get 5 volts on the cooler, it is enough to connect only the red wire of the cooler to the red wire of the PSU, and leave the two remaining wires in the cooler connector.

Thus, we got a processor cooler with reduced speed and low noise. With a significant reduction in noise, heat dissipation from the processor does not decrease or decreases slightly.

The next step is to reduce the heat dissipation of the hard drive. Since the main heating of the disk occurs due to the increased voltage on the +12 volt bus (in reality, it is always 12.6 - 13.2 volts here), everything is done very simply here. In the break of the yellow wire that feeds the hard drive, we solder a powerful diode of the KD213 type. A voltage drop of approximately 0.5 volts occurs across the diode, which favorably affects temperature regime hard drive.

Or maybe go even further? Convert PSU fan to 5 volts? It’s just not possible to translate it like that - you need to refine the BP. And it consists in the following. As you know, the main heating inside the PSU is experienced by the radiator of the low-voltage part (diode assemblies) - about 70-80 C. Moreover, the + 5v and + 3.3v assembly experiences the greatest heating. The high-voltage transistors at the correct block (this part of the PSU is correct for almost 95% of the PSU, even for Chinese ones) heat up to 40-50 C and we will not touch them.

Obviously, one common radiator for three power rails is too small. And if the radiator is still normally cooled when the fan is running at high speeds, then when the speed drops, overheating occurs. What to do? It would be wise to increase the size of the heatsink, or even separate the power rails into different heatsinks. We'll take care of the last one.

To separate from the main radiator, a + 3.3v channel was chosen, assembled on a transistor. Why not +5v? At first, this was done, but voltage ripples were discovered (the influence of the wires that extended the terminals of the diode assembly + 5V affected). Since the channel is + 3.3v. powered by + 5v., then there are no more ripples.

For the radiator, an aluminum plate 10x10 cm in size was chosen, to which the + 3.3v channel transistor was screwed. The transistor leads were extended with a thick wire 15 cm long. The plate itself was screwed through insulating bushings to the top cover of the PSU. It is important that the heatsink plate does not come into contact with the PSU cover and heatsinks of power diodes and transistors.

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

After such refinement, you can safely set the PSU fan to +5 volts.

Video card. A more precise approach is needed here. If you have a video card of the GeForce2 MX400 class, then in most cases it does not need a cooler at all (which, by the way, many manufacturers do - they do not install a cooler at all). The same applies to GeForce 4 MX440 graphics cards, Ati Radeon 9600 - a passive radiator is enough here. In the case of other video cards, the approach may be similar to the above - switching the fan power to 7 volts.

Let's sum up. We've looked at noise and heat reduction measures for an AMD processor-based system. For example, I will give the following data. This article is currently being written on a very powerful computer. AMD Athlon XP 3200+, with 512 MB of RAM, video card GeForce 4 mx440, Hdd WD 120 gb 7200, CD-RW and has a processor temperature of 38C, temperature inside the case 36C, temperature inside the PSU measured with a digital thermometer on power diode heatsinks - 52C, hard drive just cold. The maximum temperature of the processor during the simultaneous test of 3DMark and the launch of cpuburn was 68C after 3 hours of operation. At the same time, the PSU fan is connected to 5 volts, the processor fan with the TITAN cooler is connected to 5 volts all the time, the video card does not have a fan. In this mode, the computer works without any failures for 6 months, at a room temperature of 24C. Thus, powerful computer has only two fans (running at low speeds), stands under the table and is almost inaudible.

P.S. Perhaps in the summer (it will be +28 in the room) you will need to install an additional case fan (with +5V power supply, so to speak - for peace of mind ...), or maybe not, wait and see ...

Warning! If you do not have sufficient qualifications, and your soldering iron is similar in size to an ax, then do not read this article and, moreover, do not follow the advice of its author.

Bookmark the article


	Similar content

Once upon a time, there were computers. They were able to count quickly and a lot, and even display two-dimensional graphics on the monitor screen. And everything on the computer screen was flat and dull. People also wanted three-dimensionality, a sense of space, cinematic graphics. They modestly dreamed of a miracle. And a miracle appeared to the world in the face of 3Dfx Interactive.

Part 1 - Theoretical. As well as an excursion into history

Founded in 1994 by four enthusiasts, the company 3Dfx Interactive introduces the Voodoo Graphics chip to the world for the first time. Rather, not even a chip, but a chipset - PixelFX And TexelFX Engine with support for up to 4 MB of local memory, which was a miracle at the time. And a miracle happened - 3D graphics became a mass phenomenon for a personal computer.

In January 1998, 3Dfx introduced a new miracle in the form of the second generation of graphics chips - Voodoo2, along with the advent of SLI technology, which allowed several chips to Voodoo2 work in parallel. SLI (S can L ine I interactive) [not to be confused with NVIDIA SLI = S calable L ink I nterface], allowed several Voodoo2 cards to work in parallel, thereby increasing fps in games.

Games! In fairness, it should be said that among the revolutionary developments 3Dfx also had at its disposal a unique API - Glide. The vast majority of games of that time were developed specifically for this API. Until now, many people remember THAT games with great warmth. And many still play these classic games.

But that's not all. No less significant were the subsequent developments of 3Dfx.

For example, support for multi-chip solutions using SLI technology, but this time within one (!) board for an AGP slot.

It's a graphics chip. VSA-100, which contained interesting features - multi-chip image processing, full-screen anti-aliasing is very High Quality and good texture compression.

For the first time on one “household” video card, it combined two (Voodoo5 5500) and even 4 (in the legendary Voodoo5 6000) 3Dfx graphics chips. The latter, unfortunately, did not manage to get into the series. 3DFX ceased to exist independently since December 2000, tk. bought by NVIDIA.

video card 3Dfx Voodoo5 6000 also known for being a harbinger of the advent of technology Quad SLI.

Four video chips on one printed circuit board. Since it was equipped with an AGP interface, and there were no motherboards with two AGP ports, we can assume that the Voodoo5 6000 was the first graphic solution, combining four video chips in one system. Similar product nVidia showed only! SIX! years later, by releasing Quad SLI-enabled drivers to combine a pair of dual-chip GeForce 7950 GX2 graphics cards.

If we talk about multi-chip solutions, then we cannot fail to mention the company Quantum3D. And its technologies heavy metal on 3Dfx chips.

Before starting a description of Heavy Metal technology, it must be said that this technology belongs to the HI-END class (we should not forget that we are talking about 1998-2000). So, Heavy Metal is not just a graphic station, it's something more.

Heavy Metal is a high performance graphics station for all the needs that the most advanced software(at that time) for users who do not care about the price of the product, they use the most perfect.

These users were: military training bases, NASA, some major graphic studios. Such things were also used to train specialists in helicopter control and missile guidance, when it was necessary to recreate scenes of military operations in real time with maximum realism. The system was also used by civilians at the Ford Research Laboratories in Dearborn, Michigan.

Lockheed Martin chooses an open architecture imaging system AAlchemy by Quantum3D to increase the realism of the C-130 aircraft simulator.

It was for such tasks that Heavy Metal stations were designed. In particular, the most powerful VSA-100 3Dfx solution in history is the AAlchemy modules.

AAlchemy graphics subsystems have a separate metal case, a cooling system consisting of two 150 CFM fans and other components. The AAlchemy deck fits into a Heavy Metal body. Moreover, the number of such decks can reach four.

AAlchemy contains from 4 to 32 VSA-100 chips, to obtain bandwidth memory from 12.8 to 102 gigabytes per second. Alchemy uses this architecture to get 4x4 or 8x8 sub-sample, single-pass, full-scene, sub-pixel anti-aliasing at FillRate of 200 Mpixels/sec. up to 1 Gpixels/sec. AAlchemy4 was only sold as part of the Heavy Metal GX+.

Specification:

Support 4 or 8 VSA-100 chips on one board.

Support for 1, 2, 4 channels in Heavy Metal GX+

Support for precise synchronization of SwapLock and SyncLock.

Support for 16 bit Integer and 24 bit Z-buffer with 8 bit Stencil

Support for 32 bit and 22 bit rendering

Single, Double, Triple Buffering

Support for perspective correct bilinear, trilinear and selective anisotropic texture filtering with per-pixel LOD MIP mapping with Gouraud modulated, detailed and projected texture mapping

transparency and chroma-key support

Per-pixel and per-vertex atmospheric effects with simultaneous OpenGL compatible alpha blending

Support for 16, 24, 32-bit RGB/RGBA and 8-bit YIQ and color-indexed compressed textures

Support for FXT1 and S3TC texture compression

Support for textures up to 2048x2048

32 or 64 Mb Framebuffer

Support for 3dfx Glide API, Microsoft Direct3D, OpenGL and Quantum SimGL

Memory bandwidth 12.8 - 102.4 Gb/sec.

66 MHz PCI 2.1 interface with multi-chip transfer capability

Built-in geometry pipeline with a capacity of 2,100,000 textured polygons per second.

135 MHz RAMDAC with Stereo support

Support for T-Buffer technology

Given all of the above, it becomes clear why 3Dfx has acquired a huge army of fans of its products. Over time, turned into fans-collectors. And just gamers who love and appreciate old, classic games.

Again, if in the 2000s many did not dare to dream of graphics system Heavy Metal AAlchemy GX+, because even with one AAlchemy module it cost $15,000, now all this equipment can be bought for more reasonable money. It is possible in parts.

How do you like it - to fulfill the dream of your childhood, youth, youth ... who likes it? Decorate your collection with such a beauty? The author of the article is one of the fan-collectors of 3Dfx and Quantum3D products.

When I got a chance to purchase a single graphics module from the Heavy Metal AAlchemy GX+ system, I naturally didn't miss it.

But collecting computer hardware differs from collecting, for example, stamps, in that the hardware also works. Having admired enough of the man-made miracle, it occurred to me that it would be very cool to run Quake on a video card with EIGHT graphics chips on board, removed from a military or aerospace simulator! I got down to business.

The video card has a PCI interface, which makes it compatible with any modern computer.

Remind me of the next decision Voodoo5 6000:

has an AGP 2x interface, requires a motherboard with a chipset no older than 333, not compatible with many motherboards (even if they support AGP 2x)

and is such a rarity that it only appears on e-bay no more than once a year at a price of 1000 euros. And it has a performance two times lower compared to AAlchemy. Of course, these are incomparable things, but still.

It would seem that it is easier. Board for PCI slot. This is in almost all computers ... But, as always, there is a “BUT”. A specialized power supply is required to power this graphics monster. With these parameters:

Impressive? 2.9 V and 75 A!!! Almost a welding machine! The only comfort is that 75 A is required for two AAlchemy video cards combined in SLI. Half is enough for one, and this is 30-35 A.

3.3 V and 30 A is still real. There are many power supplies from 400 watts. But where to get 2.9 V?

Buy branded (native) power supply? You can certainly try, but this thing is extremely rare. And it costs a lot of money. Even on such a worldwide flea market as E-Bay, it is rarely found.

Many Western enthusiasts get out in different ways. There is an option using converters 12 V to 3.3 V DC / DC-Converter Artesyn SMT30E 12W3V3J

At first glance, it is simple and accessible. But the price of such a device is about 50 euros, and you need three of them. And getting them in Russia is not easy. And buying abroad ... long, troublesome and expensive.

There is an option using a powerful laboratory power supply and powerful current relays

I tried to figure out how much such a power supply could cost. I found 20 A 5 B. The price is twenty-odd thousand rubles. How much will a seventy-ampere one cost!?

I didn't like these options. In general, I saw such a solution: three power supplies - ordinary, computer ones. Combine Pc-ON wires. Combine common (black) wires. And somehow modify one of the power supplies to get the desired 2.9 V from it. The first two positions were decided without problems. I have two power supplies:

1. Linkworld LPQ6-400W. It's a pretty thin block. But to power my retrocomp, it will do.

2. FCP ATX-400PNF A more modern block has a current of 28A on the 3.3 V line. Practically what you need.

But from what to get 2.9V? Basically, I have a single Quantum 3D Alchemy 8164. Half of 75 will be enough for her. The power supply is designed for SLI of two Quantum 3D AAlchemy 8164. I have only one available. According to the experience of foreign users, 30 amperes is enough.

And then I remembered Powerman HPC-420-102DF. I have circuit diagram very close to this block. And I decided to take him for the base.

click on the picture to enlarge

In power supplies made according to this scheme, 5 and 3.3 V are taken from one winding of the transformer. This means that such a block has a power reserve along the 3.3 volt line. But there are two small problems. Protection against exceeding the maximum load current and protection against overvoltage and undervoltage. There is also such a thing, which is called - "voltage skew due to uneven load along the lines." How to deal with these troubles, I did not consider. Decided to "tackle problems as they come". If during operation the unit starts to turn off, then I will bother.

I opened the block and refreshed my memory by downloading and reading the datasheet on SG6105. It is on this chip that my power supply is made. The large, twenty-pin connector has three orange wires. These are 3.3 V lines. One of them is connected to the brown (usually) Vsens wire. Sometimes it is the same color, but thinner than the others. This wire controls the change in voltage at the output of the unit along the 3.3 V line.

The wire goes to the power supply board.

And through the resistor R29 it goes to leg 12 of the SG6105 chip. The leg is called VREF2. The value of this resistor determines the output voltage of the power supply along the 3.3 V line.

According to the scheme 18kOhm. I found this resistor on the block board:

Soldered one leg of this resistor, thus turning it off. You can see it in the photo. I measured the actual resistance with a multimeter. It turned out to be 4.75 kOhm. Wow! Schemes and life often differ from each other!

Now I take a variable resistor with a worm gear with a resistance of 10 kOhm. Such resistors are very popular with overclockers, because. allow you to smoothly change your resistance. Turning the resistor engine with a screwdriver, I set it to the required 4.75 kOhm. I control the value with a multimeter and solder instead of R29 from the side of the printed tracks.

I do this for adjustment. Then I make a hole in the block housing to access this resistor.

Now we need to make the connecting wires of the block with the video card. AAlchemy has a special board with connectors. You can connect to it with the help of petals. But the design of my homemade case is such that the video card is upside down. Therefore, I will screw the wires directly to the card itself. Right here:

I find orange wires in the harness. I cut it, clean it, carefully tin it and solder two wires with a cross section of at least 2.5 mm square to them. I do the same with the black wires.

(common, ground, minus power supply). I also take three wires so that the cross section of the outgoing wires is equal to the cross section of the incoming wires.

I assemble the block, isolate the soldering points of the wires with electrical tape. And the verification process begins.

For the load, I used a furniture spot with a power of 20 watts. All assumptions turned out to be correct and everything worked correctly. 2.9 V was set without problems. If you repeat this moment, then notice that I turned on the power supply without fan blowing. It is possible for a short time. But it's better to run with airflow.

For a long time I have had a homemade water-cooled case, the hero of the article.

Now it contains a retroconfiguration:

CPU Athlon 1700
MB EP-8KTA3L+
Mem 3 at 256mB
Video cards GeForce GTS
QUANTUM3D AALCHEMY

I install all three power supplies on it.

Blocks are connected according to the following scheme.

I connect the green wires of the connector of all power supplies. Now all blocks will turn on at the same time. I connect any black wire of each power supply to each other.

This building is very spacious. Such a giant as Quantum 3D Alchemy. If the first block is loaded - motherboard, processor, hard drive, GeForce GTS video card, then the rest of the load is only on the 3.3 volt line. Voltage distortion will not occur in this case, because. 3.3 V is stabilized separately from 5 V and 12 V. But the 5 V and 12 V lines cannot be left completely unloaded. Therefore, I hang neon and fans on them. Such beauty is obtained:

My Quantum 3D AAlchemy turned out to be an old revision and required a power supply not 2.9 V 2.7 V. I adjusted it without any problems variable resistor the correct voltage.

After checking everything again, I started the system. The monitor has only been connected to the GeForce GTS so far. After loading the operating system, I checked the supply voltages on AAlchemy. The 3.3V line turned out to be normal. But 2.7 V dropped to 2.65 V. I adjusted it again to 2.7 V.

The operating system immediately saw a new device and requested a driver. I got the driver from here.

Here it is, the legend, it works. I connect the second monitor to the output of AAlchemy. And I run the test.

AAlchemy works as a video accelerator in a regular computer. The image in 2D is displayed by a regular video card, and AAlchemy displays Glide applications.

Part 2 - F.A.Q.

After a successful experiment in upgrading a conventional power supply and launching AAlchemy (hereinafter abbreviated "AA5") on a regular motherboard, I tried to assemble the native package of the graphic station Heavy Metal Alchemy GX+:

2 Pentium processor III - 1000 MHz/100/256
2 x processor motherboard Intel L440GX+
Embedded Video CL-GD5480
1.5 Gb SDRAM ECC Sync. PC100R

The board has two types of PCI connectors 66 MHz and 33 MHz.

I drove AA5 on it. In the process, some subtleties of operation became clear. At first I wanted to write a continuation of the article. But I realized that it would be more useful to state all the developments in the form F.A.Q. and place it at the end of the first article. Pros - all the information in one place and clearly presented.

Actually this F.A.Q is presented to your attention:

1. Where can I get a manual for AA5?

2.What operating system use?

The graphics station was designed for use with Microsoft Windows NT4 and Windows 2000. But it also works fine with Windows XP.

3.Where can I get the driver for AA5?

There is a huge selection of drivers for 3DFX here

4. Where can I ask questions and discuss AA5?

Part 3 - Extreme. Practical tests

The third part is the most extreme. In the first two parts, it turned out that a single AA5 video card is not so difficult to run on a regular home computer. The price of the issue is an easy upgrade of a separate power supply. But .. Again “but”. Now you can purchase a module consisting of two QUANTUM 3D AALCHEMY 8164 and nVSensor post-processor. 16 GPUs! But then 75 amps will be required to power two video cards! With non-standard 2.7-2.9 V.

For such currents, the above modification is not applicable. Firstly, part of the power goes to other lines 5V, 12V, -5V, -12V. The 5V line had to be loaded with a light bulb, otherwise there was still a voltage imbalance and the unit stopped working correctly. And this is additional power loss.

The overload protection also worked. In short, it was required to get honest 75 A from the power supply at an adjustable and stabilized voltage of 2.7-2.9 V. Twice as much as the unit can give. But if the power supply is capable of delivering 400-480W on all lines, then why can't it be forced to give out all this power in one line? Can.

The original plan was this. I turn off all protections and monitoring of all voltages. I solder all the extra parts. And I make the block work only on one line. And honestly give out everything that he is capable of in ONE this line with adjustable voltage 2.7-2.9 V. This variation is due to the fact that there are two versions of AA5. There are with 2.7 V power supply, and there are also with 2.9 V.

I study in more detail the datasheet on the SQ6105. And I'm developing ways to disable all protections. The principle is simple. It is necessary to deceive SQ6105. There is a so-called "duty room" in the block. This is an independent 5 V source. From it, power is supplied to the SQ6105, before the entire power supply is turned on.

For example, how to disable 5V monitoring? Apply a voltage of 5 V to the SQ6105 output responsible for this monitoring. And I will take it from this very “duty room”. Monitor +3.3V? I’ll take 5 V from the “duty room” and use a resistor divider to supply the required 3.3 V to the SQ6105! The only problem is with 12 volts. But I solved it too. Anyway, to power a computer with AA5 installed, I use three power supplies. I'll take +12 V from any of them.

What I did, I state strictly point by point. I redid the power supply codegen 480 watts. I haven't upgraded it yet. Simple, no frills. And reliable. The only thing weakness- diode assemblies. But I changed them a long time ago. After the previous alterations, it looked like this.

It has a diagram very close to this one:

Scheme No. 1

Let's get started.

1. I connect a load to the output of the power supply - a 12 V bulb. The PS-ON wire to ground means that I short the green and black wires of the 20-pin connector with a paper clip. The light bulb is on. The block is working.

2. I disconnect the PSU from the 220 V network. (You need to pull the power cord out of the unit!) This is important. Otherwise, electric shock and possibly death. Electricity is no joke. I turn off the analysis of SQ6105 plus 5 V - I cut the track coming from pin 3, SQ6105 (V5 Voltage input + 5V, circuit 1), and I connect pin 3 to pin 20 of SQ6105 with a jumper or a 50-200 Ohm resistor (RR5 in circuit 1). Thus, I disconnect the SQ6105 from the power supply circuit and replace the monitoring of the output 5 volts with five volts of the “duty”. Now, even if the power supply does not supply 5 V to the load, SQ6105 considers that everything is fine and the protection does not work. Ready.

I turn on the power supply to the network to check, the light should be on.

3. I disconnect the PSU from the 220 V network. I turn off the definition of SQ6105 plus 3.3 V - I cut the track near pin 2 and solder two resistors, 3.3 kOhm from pin 2 to the case (RR7 in diagram 1), 1.5 kOhm from pin 2 to pin 20 (RR6 in the diagram). I turn on the power supply to the network, if it does not turn on, it is necessary to select the resistors more accurately in order to get +3.3 V at pin 2. You can use a trimming resistor with a resistance of 10 kOhm. After each alteration, it is better to check the unit for operability. Then, in case of failure, the circle of error search will narrow.

4. I turn off the PSU from the 220 V network. I turn off the definition of SQ6105 minus -5 V and - 12 V - I solder R44 (near pin 6), and I connect pin 6 to the case through a 33 kOhm resistor, more precisely 32.1 kOhm (RR8 in diagram 1 ). I turn on the power supply to the network, if it does not turn on, it is necessary to select a resistor more accurately.

5. I disconnect the PSU from the network. I turn off the definition of 12 V. To do this, I am looking for pin 7 of SQ6105. This is a 12 V input. If there is no 12 V, the microcircuit turns off the power supply. I look at the board, from leg 7 the track goes to a resistor, usually with a value of about 100 ohms. I solder the leg of this resistor - the one farthest from the microcircuit. I solder a wire to the soldered leg, to which I will supply 12 V from another power supply. There is nowhere to take 12 V in this block, and this wire will also serve as additional protection and guarantee simultaneous operation several blocks. The project requires the simultaneous inclusion of several power supplies.

6. I solder all diode assemblies. It is most convenient to do this with a soldering iron with suction. The assemblies are soldered all together with the radiator on which they are installed. I unscrew all the assemblies from the radiator and study them. I need to dial a minimum of 80A, and always with the same assemblies. From the soldered nothing came up. But in stocks there were two assemblies of 40A per 100 V. I install both of them on the radiator and connect them in parallel. Then I connect them with wires to the pads of the 5 volt line of the power supply. Wires should be as large as possible. From 4 mm 2 suitable for assemblies and 8 outgoing. Also, all involved tracks on the board, starting from the transformer, need to be powered. Either solder the wires on top, or fill them with solder. And better than both.

7. Now you need to switch the output of the error signal amplifier and the negative input of the SQ6105 comparator. To do this, we are looking for 16 (COMP) and 17 (IN) legs of this microcircuit. (This is, in fact, the very stabilization of the output voltage).

And starting from them, I go along the printed tracks and compare the real block diagram with the one I have. I get to the resistor that connects legs 16 and 17 to 12 V and solder it (R41 in diagram 2).

Scheme No. 2

I find a resistor that connects the microcircuit to 5 volts (R40 in diagram No. 2). I drink it. Then I measure its value and solder in its place a slightly larger variable resistor. Naturally, having previously exposed it to the same resistance. I solder, of course, not the resistor itself, but the wires going to the resistor. I bring the resistor to the power supply case in a convenient place. With it, I will regulate the output voltage.

I solder all unnecessary parts (electrolytes on all lines except 5 V, chokes of a 3.3 V magnetic amplifier, if the details of the -5V and -12 V lines interfere) and the wires coming from the board instead of them, I solder two wires with a cross section of 4 mm 2 to the 5 V output and general. (In the photo these are thick acoustic wires). It is better to duplicate the output wires. 4 mm section is not enough. The wire may get hot.

8. I connect the load (light bulb 12 V 20 W) to the PSU output. I turn on the power supply. PS ON to ground. The block should work. So, I didn't add anything extra.

I measure the voltage on the light bulb with a tester and adjust the voltage with an alternator to the required value of 2.7 V or 2.9 V. Everything worked out. There is very little work left.

9. Now we need to convert the group stabilization inductor to a higher current. The cross section of the inductor core is sufficient. Insufficient wire size. Still, the rated current of the winding is 40 A and will be up to 75 A!

I solder the inductor and find a 5 V winding on it. These are two or three wires with a diameter of 1.5 mm. In my case, these are two wires.

The cross section of these two wires is 3.54 mm2. The rated current is 40 A. For a value of 80 A, the cross section must be doubled. I had a wire with a diameter of 1.77 mm in stock. In order to dial the required 7.08 mm 2, three wires are required (do not confuse the cross section with the diameter!)

I wind all the windings from the group stabilization choke. I count the number of turns of a 5-volt winding. 10 turns. I wind a new winding on the torus of the magnetic circuit with three wires at the same time. To do this, it is convenient to immediately measure the required length of wires, carefully fold them in a strip and twist the ends using two pliers. Then winding will be much easier. The turns of all three windings must be exactly the same.

During the winding process, I decided to use two such chokes for better smoothing of ripples. For the second one, I removed the choke from the dead power supply and rewound it too. In principle, this is not necessary. The original circuit uses two chokes. The second is just a few turns of wire wound around a post. The core is too small for 3 wires. So I decided to put two of the same.

I soldered the first inductor in place of the group stabilization inductor into +5 V contact pads. After it, I installed an electrolytic capacitor 4700 uF at 25 V, then the second inductor (it replaced the capacitors freed from desoldering (I also soldered them along the 5 V line, I it seemed that they were of insufficient capacity). I soldered it to the pads of the next inductor. It stood there small, inconspicuous. I removed it, drilled holes and soldered a new one. And I hung two electrolytes of 10,000 microfarads 25 V on the output of this. The current doubled, therefore and the capacitance of the electrolytes should be increased.The more the better.It is also a good idea to shunt them with ceramic capacitors with a capacity of 1-10uF.This is for better high-frequency filtering.

Electrolytes of this size on the board were not removed, and I attached them to the power supply case and connected them with wires to printed circuit board. The wires must be of a decent section. Not less than one millimeter square.

To improve cooling, I made a new cover for the power supply made of perforated steel and attached a 120 mm fan to it. He was connected to the wires supplying 12 V from the second power supply.

To control the output voltage, I wanted to make a built-in voltmeter. The easiest way for me to put the arrow head. I did not find heads with a nominal value of 4 V. Found some strange device. What he measured, I don't know. But all pointer heads are microammeters. And it is easy to make a voltmeter out of them by putting a quenching resistance. So I did. Consistently turned on the head variable at 33 kOhm. Collected: it turned out pretty well.

I connected two blocks (from the second I take 12 V for the operation of the first, otherwise the block will not start, see paragraph 5). On the second, I connected a light bulb as a load. It is not recommended to turn on blocks without load. I laid everything out on my favorite stool and realized that there was nothing to load the new superblock with. I remember physics.

According to Ohm's law I=U/R, hence R=U/I

U - Voltage, V

R - Resistance, Ohm

At a current of 75A and a voltage of 2.7 V, the load resistance should be 0.036 ohms. Ordinary multimeters cannot measure such resistances. Not calculated. Well, let's go back to physics.

R - Resistance, Ohm

ρ - Resistivity for copper is 0.0175

L - Length of conductor in meters

q - Cross section, square mm

From the wires I have a twisted pair. 24AWG. Such a caliber corresponds to a cross section of 0.205 mm 2. There are eight such wires. Four wires - 0.82 mm 2. Eight - 1.64 mm 2.

Immediately at 70 A, I did not dare to turn it on. Let's start with 35 A.

We expect:

I take the cross section of 4 wires, the length turned out to be 3.6 meters.

So, half lived 3.6 meters, resistance 0.0771 Ohm, current 35A.

All eight cores, 3.6 meters, resistance 0.038 Ohm, current 71 A. In general, it should be 70A. But when calculating, I rounded. Two loads come out at once.

I connect the first half load. I turn it on. The block worked. The tension eased a little. But I adjusted it with a variable. While fiddling, the wire heated up: 95 watts of heat!

Now I connect all eight: the current has reached the value of 70 A! I turn it on - everything works !!!

Once again, the tension eased a little. But this is not a problem - we have an adjustment.

Only the load is very hot - I can not conduct a long test. After 15-20 seconds, the insulation becomes soft and begins to “float”.

P.S. In my case, for some reason, the protection against the maximum current in the load (short circuit protection) did not work. I don't know the reason. But if this happens, then this protection can be adjusted. It is necessary to reduce the resistance R8. The lower the resistance, the more current the protection will operate.

The power supply is ready. And you could connect AA5 and enjoy. But... As always. Purchase from eBay not arrived yet :(

Discussion this material conducted in a special branch of our .

Hello, now I will talk about reworking the ATX power supply of the codegen 300w 200xa model into laboratory block power supply with voltage regulation from 0 to 24 Volts, and current limiting from 0.1 A to 5 Amperes. I'll post the diagram that I got, maybe someone will improve or add something. The box itself looks like this, although the sticker may be blue or another color.

Moreover, the boards of the 200xa and 300x models are almost the same. Under the board itself there is an inscription CG-13C, maybe CG-13A. Perhaps there are other models similar to this one, but with different inscriptions.

Soldering unnecessary parts

The original diagram looked like this:

It is necessary to remove all unnecessary, the wires of the atx connector, unsolder and wind unnecessary windings on the stabilization group choke. Under the throttle on the board, where +12 volts is written, we leave that winding, the rest we wind. Unsolder the braid from the board (the main power transformer), in no case do not bite it off. Remove the radiator along with the Schottky diodes, and after we remove all unnecessary, it will look like this:

The final scheme after the alteration will look like this:

In general, we solder all the wires, parts.

Making a shunt

We make a shunt, from which we will remove the voltage. The meaning of the shunt is that the voltage drop across it tells the PWM how current is loaded - the PSU output. For example, the resistance of the shunt, we got 0.05 (Ohm), if we measure the voltage on the shunt at the time of passage of 10 A, then the voltage on it will be:

U \u003d I * R \u003d 10 * 0.05 \u003d 0.5 (Volt)

I won’t write about the manganin shunt, because I didn’t buy it and I don’t have it, I used two tracks on the board itself, we close the tracks on the board as in the photo to get a shunt. It is clear that it is better to use manganin, but it works more than fine anyway.

We put the L2 choke (if any) after the shunt

In general, they need to be calculated, but if anything, a program for calculating throttles slipped somewhere on the forum.

We supply a common minus to PWM

You can not apply if it is already ringing on the 7th leg of the PWM. It’s just that on some boards on the 7th pin there was no general minus after desoldering the parts (I don’t know why, I could be wrong that it wasn’t :)

We solder a wire to the 16th PWM output

We solder a wire to the 16th PWM output, and we feed this wire to pins 1 and 5 of the LM358

Between 1 PWM leg and the plus output, solder a resistor

This resistor will limit the voltage output by the PSU. This resistor and R60 forms a voltage divider that will divide the output voltage and supply it to 1 pin.

The op-amp (PWM) inputs on the 1st and 2nd legs are used to set the output voltage.

The task of the PSU output voltage comes to the 2nd leg, since the second leg can receive a maximum of 5 volts (vref), then the reverse voltage should come to the 1st leg also no more than 5 volts. To do this, we need a voltage divider of 2 resistors, R60 and the one that we will install from the PSU output to 1 leg.

How it works: let's say a variable resistor put 2.5 Volts on the second leg of the PWM, then the PWM will give out such pulses (increase the output voltage from the PSU output) until 2.5 (volts) comes to 1 leg of the op-amp. Suppose if this resistor does not exist, the power supply will reach the maximum voltage, because there is no feedback from the output of the BP. The resistor value is 18.5 kOhm.

We install capacitors and a load resistor on the PSU output

The load resistor can be supplied from 470 to 600 ohms 2 watts. Capacitors of 500 microfarads for a voltage of 35 volts. I didn’t have capacitors with the required voltage, I put 2 in series, 16 volts 1000 microfarads each. Solder the capacitors between 15-3 and 2-3 PWM legs.

Soldering the diode assembly

We put the diode assembly that was 16C20C or 12C20C, this diode assembly is designed for 16 amperes (12 amperes, respectively), and 200 volts of reverse peak voltage. The 20C40 diode assembly will not work for us - do not think to install it - it will burn out (checked :)).

If you have any other diode assemblies, make sure that the reverse peak voltage is at least 100 V, and for the current, whichever is greater. Ordinary diodes will not work - they will burn out, these are ultra-fast diodes, just right for impulse block nutrition.

We put a jumper for PWM power

Since we removed a piece of the circuit that was responsible for supplying power to the PSON PWM, we need to power the PWM from the 18 V power supply on duty. Actually, we install a jumper instead of transistor Q6.

Solder the output of the power supply +

Then we cut the common minus that goes to the body. We make sure that the common minus does not touch the case, otherwise, shorting the plus, with the PSU case, everything will burn out.

We solder the wires, a common minus and +5 Volts, the output of the PSU duty room

This voltage will be used to power the volt-ammeter.

Solder wires, common minus and +18 volts to the fan

We will use this wire through a 58 ohm resistor to power the fan. Moreover, the fan must be deployed so that it blows on the radiator.

We solder the wire from the transformer braid to a common minus

We solder 2 wires from the shunt for the op-amp LM358

We solder the wires, as well as resistors to them. These wires will go to the LM357 op amp through 47 ohm resistors.

Solder the wire to the 4th PWM leg

With a positive +5 Volt voltage at this PWM input, the regulation limit is limited at the outputs C1 and C2, in this case, with an increase at the DT input, the duty cycle increases at C1 and C2 (you need to look at how the transistors are connected at the output). In a word - stopping the output of the power supply unit. This 4th PWM input (we supply +5 V there) will be used to stop the PSU output in the event of a short circuit (above 4.5 A) at the output.

We assemble a current amplification and short circuit protection circuit

Attention: this is not full version- for details, including photos of the rework process, see the forum.

Discuss the article LABORATORY PSU WITH PROTECTION FROM A CONVENTIONAL COMPUTER

Progress does not stand still. Computer performance is growing rapidly. And as performance increases, so does power consumption. If earlier almost no attention was paid to the power supply, now, after nVidia's statement about the recommended power supply for their top-end solutions of 480 W, everything has changed a bit. Yes, and processors consume more and more, and if all this is still properly overclocked ...

With the annual upgrade of the processor, motherboard, memory, video, I have long resigned myself, as with the inevitable. But for some reason the upgrade of the power supply makes me really nervous. If the hardware progresses dramatically, then there are practically no such fundamental changes in the circuitry of the power supply. Well, the trance is bigger, the wires on the chokes are thicker, the diode assemblies are more powerful, the capacitors ... Is it really impossible to buy a more powerful power supply, so to speak, for growth, and live at least a couple of years in peace. Without thinking about such a relatively simple thing as high-quality power supply.

It would seem that it would be easier, buy the largest power supply you can find, and enjoy a quiet life. But it was not there. For some reason, all employees of computer companies are sure that a 250-watt power supply will be enough for you in excess. And, what infuriates most of all, they begin to lecture categorically and groundlessly prove their case. Then you reasonably notice that you know what you want and are ready to pay for it, and you need to quickly get what they ask for and earn a legitimate profit, and not anger a stranger with your senseless, unsupported persuasion. But this is only the first hurdle. Go ahead.

Let's say you still found a powerful power supply, and here you see, for example, such an entry in the price list

Power Man PRO HPC 420W - 59 u
Power Man PRO HPC 520W - 123 u

With a difference of 100 watts, the price has doubled. And if you take it with a margin, then you need 650 or more. How much is it? And that is not all!

advertising

The vast majority of modern power supplies use the SG6105 chip. And its switching circuit has one very unpleasant feature - it does not stabilize voltages of 5 and 12 volts, and the average value of these two voltages obtained from a resistor divider is applied to its input. And it stabilizes this average value. Because of this feature, a phenomenon such as "voltage distortion" often occurs. Previously used chips TL494, MB3759, KA7500. They have the same feature. I will quote from the article Mr. Korobeinikov .

"... Voltage skew occurs due to uneven load distribution across the +12 and +5 Volt buses. For example, the processor is powered from the + 5V bus, and the +12 bus is HDD and CD drive. The +5V load is many times greater than the +12V load. 5 volts fails. The microcircuit increases the duty cycle and + 5V rises, but +12 increases even more - there is less load. We get a typical voltage distortion ... "

On many modern motherboards, the processor is powered by 12 volts, then the opposite occurs, 12 volts goes down, and 5 goes up.

And if the computer works normally in the nominal mode, then during overclocking the power consumed by the processor increases, the distortion increases, the voltage decreases, the protection of the power supply against undervoltage is activated and the computer turns off. If there is no shutdown, then the lower voltage is still not conducive to good overclocking.

So, for example, it was with me. I even wrote a note on this topic - "Overclocker's lamp" Then two power supplies worked in my system unit - Samsung 250 W, Power Master 350 W. And I naively believed that 600 watts was more than enough. Enough may be enough, but due to skew, all these watts are useless. I unknowingly strengthened this effect by connecting the motherboard from Power Master, and from Samsung a screw, disk drives, etc. That is, it turned out - mainly 5 volts are taken from one power supply, 12 from the other. And the other lines are "in the air", which increased the effect of "skew".

The best option is to purchase and use a quality power supply. But if there is no possibility and / or there is a desire to improve the block you already have, then good results can be obtained when finalizing a cheap (budget) power supply. Chinese designers, as a rule, make printed circuit boards according to the criterion of maximum versatility, i.e. in such a way that, depending on the quantity installed elements could vary in quality and, accordingly, price.

Therefore, if you install those parts that the manufacturer saved on, and change something else, you get a block of the middle price category. Of course, it cannot be compared with expensive copies, where the topology of printed circuit boards, circuitry, and all the details were initially calculated to obtain high quality.
But for the average computer, this is a perfectly acceptable option.

Everything that you will do with your PSU - you do at your own peril and risk!

If you do not have sufficient qualifications, then do not read what is written here and, moreover, do nothing!

First of all, you need to open the PSU and estimate the size of the largest transformer, if it has a tag on which the numbers 33 or higher go first and has dimensions of 3x3x3 cm and more, it makes sense to mess around. Otherwise, you are unlikely to be able to achieve an acceptable result.

In photo 1 - a transformer of a normal power supply, in photo 2 - a frank Chinese transformer.

You should also pay attention to the dimensions of the group stabilization throttle. The larger the cores of the transformer and inductor, the greater the margin for saturation currents.
For a transformer, getting into saturation is fraught with a sharp drop in efficiency and the likelihood of failure of high-voltage switches, for a choke - a strong voltage spread in the main channels.

Rice. 1 Typical Chinese block ATX power supply, there is no surge protector.

The most critical details in a PSU are:
.High voltage capacitors
.High voltage transistors
.High voltage rectifier diodes
.High frequency power transformer
.Low voltage diode rectifier assemblies

Refinement:
1. First you need to replace the input electrolytic capacitors, we change them to larger capacitors that can fit into the seats. Usually in cheap units, their ratings are 220µF x 200V, or at best 330µF x 200V. Change to 470µF x 200V or better to 680µF x 200V. These capacitors affect the unit's ability to handle short power outages and the power delivered by the Power Supply.

Rice. 2 Input electrolytic capacitors and a high voltage part of the power supply, including a rectifier, half-bridge inverter, 200V (330µF, 85 degrees) electrolytes.

Next, you need to put all the chokes in the low-voltage part of the power supply unit and the surge protector (place for its installation).
Chokes can be wound on a ferrite ring with a diameter of 1-1.5 cm with copper wire with varnish insulation with a cross section of 1.0-2.0 mm 10-15 turns. You can also take chokes from a faulty PSU. You also need to solder the smoothing capacitors into the empty places of the low-voltage part. The capacitance of the capacitors should be chosen as maximum, but so that it can fit in a regular place.
Usually it is enough to put 2200µF capacitors on 16V series Low ESR 105 degrees, in +3.3V, +5V, +12V circuit.

In the rectifier modules of the secondary rectifiers, we replace all diodes with more powerful ones.
The power consumption of computers in recent times has increased to a greater extent on the + 12V bus ( motherboards and processors), so first of all you need to pay attention to this module.

Typical view of rectifier diodes:

1. - Diode assembly MBR3045PT (30A) - Installed in expensive power supplies;

2. - diode assembly UG18DCT (18A) - less reliable;

3. - diodes instead of assembly (5A) - the most unreliable option, subject to mandatory replacement.

Channel +5V Stby- Change the standby diode FR302 to 1N5822. We also put the missing filter inductor there, and increase the first filter capacitor to 1000μF.

Channel +3.3V- we change the assembly S10C45 to 20C40 (20A/40V), to the existing capacitance 2200uF/10V, we add another 2200uF/16V and the missing inductor. If the +3.3V channel is implemented on a field device, then we put a transistor with a power of at least 40A / 50V (IRFZ48N).

Channel +5V- We change the S16C45 diode assembly to 30C40S. Instead of one electrolyte 1000uF/10V, we put 3300uF/10V + 1500uF/16V.

Channel +12V- We change the F12C20 diode assembly for two in parallel UG18DCT (18A / 200V) or F16C20 (16A / 200V). Instead of one capacitor 1000uF / 16V, we put - 2pcs 2200μF / 16V.

Channel -12V- Instead of 470μF/16V, we set 1000μF/16V.

So, we put 2 or 3 diode assemblies MOSPEC S30D40 (the number after D - voltage - the more, the calmer we are) or F12C20C - 200V and similar in characteristics, 3 capacitors 2200 μF x 16 volts, 2 capacitors 470μF x 200V. Electrolytes, put only low-impedance series of 105 degrees! - 105*С.

Rice. 3 Low voltage part of the power supply. Rectifiers, electrolytic capacitors and chokes, some missing.

If the power supply heatsinks are made in the form of plates with cut petals, we unbend these petals in different directions in order to maximize their efficiency.

Rice. 5 ATX power supply with modified heatsinks.

Further refinement of the PSU comes down to the following ... As you know, in the PSU, the +5 volt and +12 volt channels are stabilized and controlled simultaneously. With +5 volts set, the actual voltage on channel +12 is 12.5 volts. If the computer has a heavy load on the +5 channel (an AMD-based system), then the voltage drops to 4.8 volts, while the voltage on the +12 channel becomes equal to 13 volts. In the case of a Pentium-based system, the +12 volt channel is loaded more heavily, and everything happens the other way around. Due to the fact that the +5 volt channel in the PSU is made much better, even a cheap unit will power an AMD-based system without any problems. While the power consumption of the Pentium is much higher (especially at +12 volts) and the cheap PSU needs to be improved.
Too much voltage on the 12 volt channel is very harmful to hard drives. Basically, HDD heating occurs due to increased voltage (more than 12.6 volts). In order to reduce the voltage of 13 volts, it is enough to solder a powerful diode, for example KD213, into the gap of the yellow wire that feeds the HDD. As a result, the voltage will decrease by 0.6 volts and will be 11.6 - 12.4V, which is quite safe for the hard drive.

As a result, by upgrading a cheap ATX power supply in this way, you can get a good PSU for a home computer, which, moreover, will heat up much less.