Almost every desktop PC has 1. They have billions of transistors, can utilize hundreds of watts of power, and can price over a thousand dollars. They are masterpieces of electronic engineering and generate extremes in product loyalty and disdain... and nevertheless the number of things they commonly exercise can be counted one just 1 hand. Welcome to the earth of graphics cards!

Graphics cards are among the PC components to which we dedicate the most time on TechSpot, so clearly they need a proper dissection. Allow'due south get together the team and head for the operating table. Information technology's time to delve into the innards that make up a graphics carte du jour, splitting up its various parts and seeing what each flake does.

Enter the dragon

Ah, the humble graphics card, or to give information technology a more fulsome championship: the video dispatch add together-in expansion card. Designed and manufactured by global, multi-billion dollar businesses, they're frequently the largest and singularly well-nigh expensive component inside a desktop PC.

Correct now on Amazon, the top ten best selling ones range from but $53 to an eye-watering $1,200, with the latter coming in it 3.3 pounds (1.five kg) in weight and over 12 inches (31 cm) in length. And all of this is to do what, exactly? Utilize silicon chips on a circuit board to, for most users, create 2nd & 3D graphics, and encode/decode video signals.

That sounds like a lot of metal, plastic, and coin to practice only a few things, so we better go stuck into one and run into why they're similar this. For this anatomy piece, we're using a graphics card fabricated by XFX, a Radeon HD 6870 like the one we tested way back in October 2022.

Outset glances don't reveal much: it's quite long at about 7 inches (22 cm) only most of it seems to be plastic. Nosotros tin come across a metal bracket, to hold it firmly in place in the computer, a big cherry-red fan, and a circuit board connector.

Permit's flip it over and see if in that location'south more to exist seen on the other side:

The first thing you notice is another metal bracket, framing a whole host of electronic components. The residuum of the circuit board is reasonably empty, but we tin see lots of connecting wires coming from the middle section, and so there's obviously something of import there and it must be pretty complex...

It'due south hot as heck in hither!

All chips are integrated circuits (ICs, for short) and they generate estrus when they're working. Graphics cards accept thousands of them packed into a small volume, which means they'll become pretty toasty if the oestrus isn't managed in some fashion.

As a consequence, nigh graphics cards are subconscious underneath some kind of cooling organisation and ours is no exception. Out with the screwdrivers and in with a blatant disregard for warranty alert stickers... this is a pretty old unit anyway:

This cooling system is mostly known as a 'blower' type: it sucks air from inside the computer case, and drives it across a big clamper of metal, before diggings it out of the case.

In the epitome in a higher place, we tin can see the metal block, and sticking to it are the remains of a runny substance more often than not known every bit thermal paste. Its chore is to fill up in the microscopic gaps betwixt the graphics processor (a.one thousand.a. the GPU) and the metal block, so that oestrus gets transferred more efficiently.

We can also see three blue strips called thermal pads, equally the soft, squishy material provides a amend contact between some parts on the excursion board and the base of the cooler (which is also metal). Without these strips, those parts would barely, if at all, bear upon the metal plate and wouldn't be cooled.

Let'southward have a closer look at the master chunk of metal -- in our example, it's a cake of copper, with three copper pipes coming off it, into multiple rows of aluminum fins. The name for whole affair is a heatsink.

The pipes are hollow and sealed at both ends; inside there is a small quantity of water (in some models, information technology'south ammonia) that absorbs heat from the copper cake. Eventually, the h2o heats upward so much that it becomes vapor and it travels abroad from the source of the heat, to the other stop of the pipe.

There, information technology transfers the heat to the aluminum fins, cooling dorsum down in the process and go a liquid again. The inner surface of the pipes is rough, and a capillary action (besides chosen wicking) carries the water across this surface until it reaches the copper plate again.

Heatpipes, every bit they are commonly called, don't appear on every graphics card. A budget, low-end model won't produce much heat and so won't need them. The same models often don't apply copper for the heatsink, to salvage money, as shown in the image below.

This Asus GeForce GT 710 shows the typical approach for the low-end, low-power graphics cards market place. Such products rarely use more 20W of electrical power, and fifty-fifty though near of that ends up as heat, it'due south merely not enough to bother the chip hiding under the cooler.

1 trouble with the blower type of cooler is that the fan used to nail air across the metal fins (then out of the PC case), has got a tough job to do and they're usually no wider than the graphics card itself. This means they demand to spin pretty fast, and that generates lots of noise.

The common solution to that problem is the open libation -- here the fan simply directs air down into the fins, and the remainder of the plastic/metallic surrounding the fan just lets this hot air into the PC example. The reward of this is that the fans can be bigger and spin slower, generally resulting in a quieter graphics card. The downside? The card is bulkier and the inside temperature of the computer volition be higher.

These 2 fans do their best to cool down the 225W MSI Radeon RX 5700 XT MECH OC

Of course, you lot don't have to utilize air to go all Netflix and arctic with your graphics card. Water is really good at absorbing heat, before it rises in temperature (roughly iv times better than air), so naturally you can buy h2o cooling kits and install them, or sell a kidney or 2, and buy a graphics menu with i already fitted.

The carte above is EVGA's GeForce RTX 2080 Ti KiNGPiN GAMING - the blower libation is but for the RAM and other components on the circuit lath; the principal processing chip is water cooled. All yours for a snip at $1,800.

You may exist surprised to know that the RTX 2080 Ti has a 'normal' maximum power consumption of 250W, which is less that the RX 5700 XT we mentioned a picayune earlier.

Such excessive cooling systems aren't in that location for everyday, run-of-the-mill settings though: you'd become with one of these setups, if y'all desire to raise the voltages and clock speeds to ludicrous levels, and spit fire like a existent dragon.

The brains behind the brawn: Enter the GPU

Now that we've stripped off the cooling system from our graphics card, let's see what we have: a excursion board with a chunky chip in the middle, surrounded past smaller blackness fries, and a plethora of electrical components everywhere else.

No affair what graphics card you have, they all take the same kind of parts and follow a similar layout. Even if we become all the style back to 1998, and wait an ancient ATi Technologies graphics menu, you lot can yet see roughly the same matter:

An ATi 3D Charger, sporting a Rage IIC chip.
Image source: VGA Museum (pay these guys a visit - it's a slap-up site!)

Like our disassembled HD 6870, at that place'southward a large fleck in the middle, some retention, and a agglomeration of components to keep it all running.

The large processor goes past many names: video adapter, 2D/3D accelerator, graphics chip to proper name just a few. Merely these days, nosotros tend to call it a graphics processing unit or GPU, for short. Those three messages take been in use for decades, simply Nvidia will claim that they were start to use it.

Not that it matters today. All GPUs take pretty much the same structure inside the chip we come across:

This processor was designed by AMD and manufactured by TSMC; it has codenames such equally TeraScale 2 for the overall architecture, and Barts XT for the chip variant. Packed into 0.iv square inches (255 mm2) of silicon are i.7 billion transistors.

This heed-extraordinary number of electronic switches brand upwardly the different ASICs (application specific integrated circuits) that GPUs sport. Some simply practice gear up math operations, like multiplying and adding 2 numbers; others read values in memory and convert them into a digital betoken for a monitor.

GPUs designed to lots of things all at the same time, so a large percentage of the scrap's construction consists of repeated blocks of logic units. You tin can see them conspicuously in the post-obit (heavily processed) epitome of AMD'due south electric current Navi range of GPUs:

See how there are 20 copies of the same pattern? These are the principal calculating units in the chip and do the bulk of the work to make 3D graphics in games. The strip that runs roughly downward the middle is mostly cache -- loftier speed internal retentivity for storing instructions and data.

On the edges, tiptop and lesser, are the ASICs that handle communications with the RAM chips on the card, and the far right of the processor houses circuits for talking to the residuum of the computer and encoding/decoding video signals.

You can read up nearly the latest GPU designs from AMD and Nvidia, besides equally Intel, if you want a amend understanding of the guts of the GPU. For now, we'll just bespeak that if you desire to play the latest games or smash through machine learning code, you're going to need a graphics processor.

Simply non all GPUs come up on a excursion board that you jam into your desktop PC. Lots of CPUs have a little graphics processor congenital into them; below is an Intel press shot (hence the rather blurry nature to it) of a Cadre i7-9900K.

The coloring has been added to identify diverse areas, where the blueish section on the left is the integrated GPU. Equally you can see it takes upwardly roughly one tertiary of the entire chip, but because Intel never publicly states transistor counts for their chips, it'south difficult to tell just how 'big' this GPU is.

We can approximate, though, and compare the largest and smallest GPUs that AMD, Intel, and Nvidia offering out of the latest architectures:

Manufacturer AMD Intel Nvidia
Architecture RNDA Gen 9.5 Turing
Chip/model Navi ten Navi 14 GT3e GT1 TU102 TU117
Transistors count (billions) x.three 6.4 Some Fewer 18.6 four.seven
Die size (mmtwo) 251 158 Effectually lxxx? 30 or so 754 200

The Navi GPUs are congenital on a TMSC 7nm procedure node, compared to Intel'southward ain 14nm and a specially refined 16nm node that TMSC offers for Nvidia (it'south chosen 12FFN); this means y'all can't straight compare betwixt them, merely ane thing is for sure -- GPUs accept lots of transistors in them!

To give a sense of perspective, that 21 yr old Rage IIC processor shown earlier has 5 one thousand thousand transistors packed into a dice size of 39 mm2. AMD'south smallest Navi flake has 1,280 times more transistors in an surface area that'south only four times larger -- that's what 2 decades of progress looks like.

An elephant never forgets

Similar all desktop PC graphics cards, ours has a bunch of retentivity chips soldered onto the circuit lath. This is used to store all of the graphics data needed to create the images we encounter in games, and is nearly e'er a type of DRAM designed specifically for graphics applications.

Initially called DDR SGRAM (double data charge per unit synchronous graphics random access retention) when it appeared on the market, today now goes by the abbreviated proper name of GDDR.

Our item sample has 8 Hynix H5GQ1H23AFR GDDR5 SDRAM modules, running at 1.05 GHz. This type of memory is still institute on lots of cards available today, although the industry is generally moving towards the more than recent version, GDDR6.

GDDR5 and 6 work in a similar style: a baseline clock (the one mentioned in a higher place) is used to time the issue of instructions and data transfers. A separate organisation is then used to shift bits to and from the retention modules, and this is done as a block transfer internally. In the example of GDDR5, 8 ten 32 bits is processed for each read/write access, whereas GDDR6 is double that value.

This block is set every bit a sequential stream, 32 $.25 at a fourth dimension, at a charge per unit that'due south controlled by a different clock system once again. In GDDR5, this runs at twice the speed of the baseline clock, and data gets shifted twice per tick of this detail clock.

Then in our Radeon HD 6870, with 8 Hynix chips in total, the GPU can transfer upwards to 2 (transfer per tick) x 2 (double information clock) x i.05 (baseline clock) x viii (chips) x 32 (width of stream) = 1075.2 Gbits per 2d or 134.4 GB/s. This calculated effigy is known as the theoretical retention bandwidth of the graphics card, and generally, yous want this to be every bit high as possible. GDDR6 has two transfer charge per unit models, normal double rate and a quad model (double the double!).

Non all graphics cards utilise GDDR5/6. Depression-stop models oftentimes rely on older DDR3 SDRAM (such as the passively cooled example we showed earlier), which isn't designed specifically for graphics applications. Given that the GPU on these cards will be pretty weak, in that location's no loss in using generic memory.

Nvidia dabbled with GDDR5X for a while -- it has the same quad data charge per unit as GDDR6, but can't exist clocked as fast -- and AMD has used HBM (High Bandwidth Memory) on-and-off for a couple of years, in the likes of the Radeon R9 Fury X, the Radeon 7, and others. All versions of HBM offer huge amounts of bandwidth, but it'south more expensive to industry compared to GDDR modules.

AMD'due south get-go GPU with HBM (the 4 chips at the sides). Image: C. Spille, pcgameshardware.de

Retention chips have to be directly wired to the GPU itself, to ensure the all-time possible performance, and sometimes that means the traces (the electrical wiring on the circuit board) can exist a little weird looking.

Notice how some traces are direct, whereas others follow a wobbly path? This is to ensure that every electric betoken, between the GPU and the memory module, travels forth a path of exactly the same length; this helps prevents anything untoward from happening.

The corporeality of memory on a graphics card has changed a lot since the starting time days of GPUs.

The ATi Rage 3D Charger we showed earlier sported but 4MB of EDO DRAM. Today, information technology'southward a grand times greater, with 4 to six GB existence the expected norm (top cease models often double that again). Given that laptops and desktop PCs are just recently moving away from having 8 GB of RAM equally standard, graphics cards are veritable elephants when it comes to their memory amounts!

Ultra fast GDDR memory is used because GPUs need to read and write lots of information, in parallel, all the fourth dimension they're working; integrated GPUs oftentimes don't come with what's called local memory, and instead, have to rely on the system'southward RAM. Access to this is much slower than having gigabytes of GDDR5 correct next to the graphics processor, simply these kinds of GPUs aren't powerful enough to actually crave it.

I demand power and lots of it

Similar any device in a computer, graphics cards need electrical power to work. The amount they need by and large depends on what GPU the bill of fare is sporting as the memory modules simply demand a couple of watts each.

The start way that the card can get its power is from the expansion slot it'south plugged into, and virtually every desktop PC today uses a PCI Limited connection.

In the image higher up, the ability is supplied past the smaller strip of pins to the left - the long strip on the right is purely for instructions and data transfer. In that location are 22 pins, 11 per side, in the short strip only non all of those pins are there to power the card.

Almost half the 22 pins are for general arrangement tasks -- checking the carte du jour is okay, simple powering on/off instructions, and then on. The latest PCI Express specification places limits as to how much current can be drawn, in total, off the two sets of voltage lines. For modern graphics cards, it'south 3 amps off the +3.3V lines and five.5 amps off the +12; this provides a total of (3 x 3.3) + (5.5 x 12) = 75.9 watts of power.

So what happens if your card needs more than than that? For instance, our Radeon HD 6870 sample needs at to the lowest degree 150W -- double what we can get from the connector. In those cases, the specification provides a format that can be followed by manufacturers in the course of extra +12V lines. The format comes in two types: a 6 pin and an 8 pivot connector.

Both formats provide three more +12 voltage lines; the divergence betwixt them lies in the number of footing lines: three for the 6 pin and 5 for the eight pin version. The latter allows for more than current to be drawn through the connector, which is why the six pin only provides an additional 75W of ability, whereas the larger format pushes this up to 150W.

Our card has ii 6 pivot connectors, so with the PCI Express expansion slot, it can take up 75 + (2 x 75) = 225 watts of power; more than enough for this model's needs.

The next problem to manage is the fact that the GPU and retentiveness chips don't run on +three.3 or +12 voltages: the GDDR5 fries are 1.35V and the AMD GPU requires 1.172V. That means the supply voltages need to be dropped and advisedly regulated, and this task is handled by voltage regulator modules (VRMs, for curt).

We saw similar VRMs when we looked at motherboards and power supply units, and they're the standard machinery used today. They also get a bit toasty when they're working away, which is why they're too (or hopefully likewise!) buried underneath a heatsink to keep them inside their operating temperature range.

Just like with motherboard and CPUs, the number and type (read: quality) of the VRMs has an affect on how stable the GPU is, when information technology's beingness overclocked. Included in this, is the quality of the overall power controlling scrap.

This decade-erstwhile Radeon HD 6870 uses a CHIL CHL821401, which is a iv+1 phase PWM controller (so it can handle the four VRMs seen higher up, plus another voltage regulation arrangement); it tin also keep track of temperatures and how much current is existence drawn. It can set up the VRMs to change between ane of three dissimilar voltages, a characteristic that's used heavily in modern GPUs every bit they switch to a lower voltage when idling, to save power and keep heat/noise downward.

The more than ability the GPU requires, the more VRMs and the meliorate the PWM controller will demand to be. For case, in our review of Nvidia's GeForce RTX 2080, we pulled off the cooling system to wait at the excursion board and the components:

You can see a battery of ten VRMs running down the full height of the carte du jour, to the right of the GPU; the likes of EVGA offer graphics cards with nigh double that number! These cards are, of course, designed to be heavily overclocked and when they are, the power consumption of the graphics carte du jour will be well over 300W.

Fortunately, not every graphics carte out there has insane energy requirements. The all-time mid-range products out there correct now are between 125 and 175W, which is roughly in the same ballpark every bit our Radeon HD 6870.

The ins and outs of a graphics bill of fare

So far we've looked at the electronic components on the circuit lath and how power is supplied to them. Time to come across how instructions and data are sent to the GPU, and how it so sends its results off to a monitor. In words, the input/output (I/O) connections.

Instructions and data are sent and received via the PCI Express connector we saw earlier. It'due south all washed through the pins on the long section of the slot connector. All of the transmit pins are on one side, and the receive pins on the other.

PCI Limited communication is done using differential signalling, and then two pins are used together, sending i flake of data per clock wheel. Forever, pair of data pins in that location are a farther two pins for grounding, and then a full ready comprises 2 send pins, two receive pins, and 4 ground pins. Together, they're collectively called a lane.

The number of lanes used by the device is indicated by the label x1, x4, x8, or x16 - referring to 1 lane, 4 lanes, etc. Well-nigh all graphics cards utilise sixteen lanes (i.eastward. PCI Limited x16), which ways the interface tin can send/receive upwards to 16 $.25 per cycle.

The signal sending the data run at iv GHz in a PCI Express 3.0 connector, but information can exist timed for sending twice per bicycle. This gives a theoretical data bandwidth of iv GHz ten ii per wheel x 16 scrap = 128 Mbits/sec or 16 MB/sec each fashion.

It's actually less than that, because PCI Express signalling uses an encoding system that sacrifices some of bits (around 1.5%) for signal quality. The very latest version of the PCI Limited specification, 4.0, doubles this to 32 GB/due south; there are two more specifications in development that multiply this once again by 2 respectively.

Some graphics cards, similar our Hard disk 6870, have an additional connector, as shown below:

This lets you couple two or more cards together, so that they can share data chop-chop when working as a multi-GPU system. Each vendor has their ain proper noun for it: AMD calls theirs CrossFire, Nvidia does SLI. The sometime doesn't utilize this connector anymore, and instead just does everything through the PCI Express slot.

If you look support this page at the paradigm for the GeForce RTX 2080, you'll see that there are two such multi-GPU connections -- this is Nvidia'southward newer version, called NVLink. It'southward mostly targeted towards professional graphics and compute cards, rather than general gaming ones. Despite the efforts of AMD and Nvidia to become multi-GPU systems into mainstream use, information technology's not been overly successful, and these days you're just improve off getting the all-time single GPU that you can afford.

Every desktop PC graphics volition also accept at least one method to connect a monitor to it, only most have several. They do this because monitors come in all kinds of models and budgets, which means the bill of fare will need to support as many of those every bit possible.

Our stripped down Radeon has 5 such outputs:

  • 2x mini DisplayPort 1.2 sockets
  • 1x HDMI ane.4a socket
  • 1x DVI-D (digital only) dual-link socket
  • 1x DVI-I (digital and analogue) dual-link socket

Accept a look:

As well supporting as many monitor types as possible, having multiple output sockets also means you tin can adhere more than one display to the graphics card. Some of this monitor juggling gets handled by the GPU itself, but sometimes an actress scrap is needed to a scrap of betoken sleight of hand. In our card, it has a Pericom P13HDMI4 HDMI switch to do some of this:

That tiny piffling chip converts HDMI data, which contains digital video and audio streams, into the picture-simply signals for the DVI sockets. The specification of these connections is a lot more than important these days, considering of changes in how we're using our monitors.

The rise of esports has the monitor manufacture steering towards ever college refresh rates (the number of times per 2d that the monitor redraws the image on the screen) -- x years ago, the vast majority of monitors were capped at 60 or 75 Hz. Today, y'all can get 1080p screens that tin run at 240 Hz.

Modern graphics cards are likewise very powerful, and many are capable of running at high resolutions, such as 1440p and 4K, or offer high dynamic range (HDR) outputs. And to add together to the long list of demands, lots of screens support variable refresh rate (VRR) technology; a arrangement which prevents the monitor from trying to update its epitome, while the graphics card is still drawing it.

There are open and proprietary formats for VRR:

  • DisplayPort Adaptive-sync
  • HDMI two.ane VRR
  • AMD FreeSync
  • Nvidia G-Sync

To exist able to use these features (e.g. loftier resolutions, high and variable refresh rates, HDR), at that place are three questions that have to exist answered: does the monitor support it? Does the GPU back up it? Does the graphics carte du jour use output connectors capable of doing it?

If you head off and buy one of the latest cards from AMD (Navi) or Nvidia (Turing), here's what output systems they back up:

Manufacturer AMD Nvidia
DVI Dual-Link Digital Dual-Link Digital
1600p @ 60Hz, 1080p @ 144Hz 1600p @ 60Hz, 1080p @ 144Hz
DisplayPort 1.4a (DSC ane.2) 1.4a (DSC 1.2)
4K HDR @ 240Hz, 8K HDR @ 60Hz 4K HDR @ 144Hz, 8K HDR @ 60Hz
HDMI 2.0b 2.0b
4K @ 60Hz, 1080p @ 240Hz 4K @ 60Hz, 1080p @ 240Hz
VRR DP Adaptive-sync, HDMI VRR, FreeSync DP Adaptive-sync, HDMI VRR, G-Sync

The above numbers don't paint the total moving-picture show, though. Sure, you could fire 4K frames at over 200 Hz through the DisplayPort connexion to the monitor, just it won't be with the raw data in the graphics carte's RAM. The output can only transport so many bits per second, and for really high resolutions and refresh rates, it's non plenty.

Fortunately, data pinch or chroma subsampling (a process where the amount of color data sent is reduced) can be used to ease the load on the display system. This is where slight differences in graphics carte du jour models tin can make a difference: i might apply a standard pinch system, a proprietary one or a variant of chroma subsampling.

xx years ago, there were large differences in the video outputs of graphics cards, and yous were often forced to make the choice of sacrificing quality for speed. Not so today... phew!

All that for simply graphics?

It might seem just a wee bit odd that so much complexity and price is needed to just depict the images we run across, when nosotros're playing Telephone call of Mario: Deathduty Battleyard. Go back to about the start of this article and look again at that ATi 3D Charger graphics card. That GPU could spit out up to 1 one thousand thousand triangles and colour in 25 meg pixels in a second. For a similarly priced carte du jour today, those numbers jump by a factor of over 2,000.

Do we actually demand that kind of functioning? The answer is yes: partly because today's gamers have much higher expectations of graphics, only also because making realistic 3D visuals, in real-time, is super hard. So when you're hacking abroad at dragons, zooming through Eau Rouge into Raidillon, or frantically coping with a Zerg Rush, just spare a few seconds for your graphics card -- y'all're asking a lot of information technology!

But GPUs tin can do more than just process images. The past few years has seen an explosion in the use of these processors in supercomputers, for complex motorcar learning and artificial intelligence. Cryptomining became insanely pop in 2022, and graphics cards were platonic for such work.

The buzzword here is compute -- a field normally in the domain of the CPU, the GPU has now taken over in specific areas that require massive parallel calculations, all done using high precision data values. Both AMD and Nvidia make products aimed at this market, and they're about always sporting the biggest and most expensive graphics processors.

Speaking of which, have you always wondered what the guts of a $two,500 graphics bill of fare looks similar? Gamers Nexus must have been curious, too, because they went ahead and pulled one apart:

If you're tempted to do the same with yours, please be careful! Don't forget that all those electronic components are quite fragile and we doubt any retailer will replace information technology, if yous mess things upwards.

And then whether your graphics card cost $twenty, $200, or $2,000, they're all fundamentally the aforementioned design: a highly specialized processor, on a circuit lath filled with supporting chips and other electronic components. Compared to our dissection of a motherboard and ability supply unit, at that place's less stuff to pull apart, but what's there is pretty awesome.

And so we say bye to the remains of our Radeon Hard disk 6870 graphics card. The bits volition go in a box and get stored in a cupboard. Somewhat of an ignominious cease to such a marvel of computing applied science, that wowed us by making incredible images, all fabricated possible through the apply of billions of microscopic transistors.

If you've got any questions about GPUs in general or the one yous're using right at present in your reckoner, and so transport them our mode in the comments section below. Stay tuned for even more beefcake series features.

Shopping Shortcuts:
  • GeForce GTX 1660 Super on Amazon
  • GeForce RTX 2060 on Amazon
  • GeForce RTX 2070 Super on Amazon
  • GeForce RTX 2080 Super on Amazon
  • GeForce RTX 2080 Ti on Amazon
  • Radeon RX 580 on Amazon
  • Radeon RX 5600 XT on Amazon
  • Radeon RX 5700 on Amazon
  • Radeon RX 5700 XT on Amazon