A serious graphics card is a must if you want to play demanding games like Shadow of the Tomb Raider and Red Dead Redemption 2.
That’s also true if you want to run highly complex engineering simulations.
Sometimes, one graphics card alone can’t handle all the load, so you’ll have to find a way to double or triple your computing power.
Is it possible to have multiple graphics cards on one PC?
Is there a limit on the number?
How Many Graphics Cards Can A PC Have?
You can have as many graphics cards as the motherboard will allow.
The motherboard is the main limiting factor because it needs enough PCI express slots to mount the graphics cards.
Additionally, you need to consider your case’s size, power supply capacity, and heat dissipation mechanism.
You can have up to four graphics cards on a regular PC, and you can connect them using SLI or Crossfire, depending on the manufacturers, to add up their capacities and create one massive GPU.
You can also install multiple graphics cards without a bridge.
In this case, the cards will function independently, so you’ll use one of them at a time.
However, you can switch between the cards for different purposes, such as gaming and video editing.
Specialized computers and cryptocurrency mining rigs may include up to 256 GPUs.
And hardware designers are continuously pushing the boundaries on what’s possible.
How You Can Use Multiple Graphics Cards
When using more than one graphics card, you can configure them to work independently or pool their resources and function as a single card.
Below, we’ll talk about the second case in more detail.
We’ll only cover independent operation mode briefly because it’s not an everyday use case.
1. Connected GPUs
To integrate two or more graphics cards, you need a way to allow them to communicate and coordinate frame rendering.
The leading GPU manufacturers, Nvidia and AMD, have designed technologies that establish direct communication across GPUs without going through the chipset or CPU.
Nvidia’s product is Scalable Link Interface (SLI), while AMD calls its multi-GPU technology Crossfire.
Before we go into details, you should know that Nvidia has ended its support for SLI, but the technology is still pretty usable.
We’ll discuss the deprecation later in the article.
Now, let’s get into the specifics of each multi-GPU technology.
A. Scalable Link Interface (SLI)
SLI hit the market in 2004, although its predecessor dates back to 1998.
It allows users to theoretically double their gaming performance by running two graphics cards simultaneously—assuming their intended game or application has SLI support.
It’s been marketed as a way to achieve next-generation performance today.
However, it has two primary restrictions on the graphics cards you can together:
- SLI only works with graphics cards that have the same GPU
- The cards need to have the same amount of VRAM.
The best approach is to use identical graphics cards, but you need to use cards with identical GPUs if that’s not an option.
However, you can’t pair up dissimilar cards using SLI, no matter how hard you try.
In addition to the compatible graphics cards, you need:
- An SLI-enabled operating system such as Windows 7 and 10.
- An SLI-certified motherboard with at least two PCI Express X16 slots.
- An SLI bridge.
- A high-capacity power supply that can accommodate your two or more graphics cards.
- An effective cooling mechanism because multi-GPU setups generate excessive heat.
SLI has gone through four generations of updates, with the most recent ones being High Bandwidth SLI (HBSLI) and SLI over NVLink (which we’ll talk about later).
Note: After mounting the graphics cards, connecting the bridge, and installing the driver software, you need to enable SLI settings in your NVidia control panel to enjoy the performance gains.
B. Crossfire
Similar to SLI, AMD’s Crossfire allows you to run two to four graphics cards simultaneously and gives you the highest performance gain at high resolutions.
Similarly, it massively increases your system’s power consumption and heat output.
However, Crossfire gives you more freedom when pairing up cards.
Unlike SLI, you can connect cards with different GPUs, clock speeds, and RAM capacities as long as they’re based on the same architectural family.
Crossfire also tends to be cheaper than SLI because AMD doesn’t force board manufacturers to pay a licensing fee.
Moreover, SLI-compatible boards need to have at least PCI 8X slots, which are more expensive.
Finally, AMD-based dual GPU solutions don’t need an extra connector because the GPUs can communicate over the PCI slots.
Besides being cheaper, this lack of proprietary connector allows for extra flexibility—a hallmark of most AMD solutions and their approach to building computer hardware.
Note: Similar to SLI, you need to enable software settings for Crossfire to start working.
C. NVLink
NVLink is a modern version of SLI developed as a protocol for professional-grade GPUs used in scientific experiments and large-scale analyses.
Nvidia released a version of the technology to the consumer market with the GeForce RTX series in 2018.
Thanks to a couple of innovations, NVLink is infinitely faster than SLI.
Its first noticeable difference is the larger connector fingers, almost three times as large as the previous generation.
SLI has a major limitation in its design: one of the GPUs has to function as the master card while the remaining cards operate as slaves.
The master GPU decides how the workload gets divided among the GPUs and integrates the render results it receives from the slave cards.
The master’s speed is capped at 2 Gbps even with the highest speed SLI connector, which isn’t very high.
Moreover, since some of the master’s capacity is wasted on coordinating the other GPUs, the attainable capacity will be smaller than the sum of all the GPU capacities.
In contrast, NVLink’s architecture is bidirectional, meaning there’s no single GPU that dictates what the other ones will do.
The radically different architecture and the extra pins translate into 150 times more data transfer capacity.
You can expect the final capacity to match the sum of the individual GPUs because the overhead is minimal.
D. SLI Over NVLink
The NVLink available in RTX graphics cards is, in fact, SLI over the NVLink bus.
Therefore, you can’t expect resource pooling with these cards, and the overheads are still there.
The only difference is the data superhighway offered by NVLink’s larger connections.
Still, this version offers up to 20 percent improvement in gaming performance and eliminates micro-stuttering.
Note: Consumer NVLink bridges have fewer pins than professional-grade ones.
E. Law Of Diminishing Returns
The law of diminishing returns is a famous principle borrowed from economics.
In the context of computers, it states that doubling your hardware resources won’t double your performance.
As you keep adding extra hardware, the performance gains will become negligible.
The same principle applies to graphics cards: two cards cost you twice, but your performance won’t be twice as good.
That’s because inter-GPU communication imposes some overhead.
The GPUs need to coordinate on how to split the workload
In addition to the cost of the graphics cards, you’ll have higher power consumption and heat output, meaning you’ll have to spend extra on your power supply and cooling.
2. Independent Cards
As mentioned earlier, you don’t have to connect your graphics cards if you don’t want to pool their computing capacity.
The most likely scenario is when you want to use your computer for two different purposes, such as intensive numerical analyses and regular gaming.
In this case, you can invest in a high-end GPU, like an RTX 3080, for the analyses and buy a budget-friendly card for your regular tasks.
Another scenario is when you want to play older games.
You can install a low-end graphics card on a backward compatible PCI express slot and switch to it when you load up your game.
Admittedly, that scenario isn’t very likely, but it’s possible.
Finally, you can install multiple independent graphics cards on a computer if you want to use virtualization to turn one physical machine into multiple virtual machines.
With this configuration, you can have one shared CPU and motherboard while dedicating a single GPU to each virtual workstation.
Here’s an example of how this team built six high-power video editing workstations using one CPU and six GPUs.
Advantages Of Having Multiple Graphics Cards
When you use two or more graphics cards simultaneously, you’ll get better performance, especially at higher resolutions.
You can also play your games at higher quality settings while maintaining an acceptable frame rate.
Scientists, data analysts, and financial experts also benefit from resource pooling because they can quickly run their simulations and calculations.
Finally, some avid gamers install multiple graphics cards to show off their gaming rigs and earn bragging rights.
Even if you don’t use a multi-GPU setup, having two graphics cards could still be beneficial as you can switch between them for different purposes.
For example, you can use the more powerful one to play games and watch 4k videos while using the other one for less demanding tasks.
Disadvantages Of Having Multiple Graphics Cards
We’ve already mentioned that having more than one graphics card can put a dent into your pocket, and you may need a high-end cooling mechanism—not to mention the higher electricity costs of running multiple GPUs.
In addition to those, the most common problem with multi-GPU setups is micro-stuttering, also known as frame time variance.
It refers to when your FPS suddenly dips to very low levels for a split second.
In this case, the average FPS will be acceptable, but looking at the 99th percentile, FPS will reveal the problem.
The frame rate is adequate 99 percent of the time, but it drops very low in one percent of the cases, which can ruin your gaming experience.
Thanks to its stricter restrictions, SLI has managed micro-stuttering more effectively than AMD.
The final issue with multi-GPU systems is that you won’t get any performance benefits if the program or game you’re using doesn’t have native support for the technology.
Plus, if your game doesn’t support SLI, you need to turn SLI off before you can play the game.
Note: If you want to use all of your GPUs’ horsepower for analyses or simulations, you can run them in compute mode, which disables the graphics engine. However, you’ll need an extra (independent) graphics card to connect to your monitor.
End Of Multi-GPU Systems
As chips have become increasingly powerful and monitor resolutions have grown, multi-card computers have fallen out of favor among regular users because one GPU is more than enough to meet their needs nowadays.
That’s why Nvidia stopped rolling out new SLI drivers in January 2021, and AMD’s Crossfire won’t have a native driver for Windows 11.
Additionally, DirectX 12 doesn’t support SLI anymore.
According to the industry leaders, this deprecation allows game developers to dedicate their resources to building better games instead of implementing an extra layer of code that handles multi-GPU support.
Fewer games resort to SLI these days because they no longer run into performance bottlenecks with one GPU.
If you want to run your games in ultra settings, the most effective solution is to buy a single powerful GPU.
That said, many popular titles are still compatible with SLI and Crossfire, and they’re likely to stay that way for a few years at least.
Here are some of the SLI-enabled titles you can enjoy:
- Shadow of the Tomb Raider
- Rise of the Tomb Raider
- Civilization VI
- Sniper Elite 4
- Gears of War 4
- Ashes of the Singularity: Escalation
- Strange Brigade
Frequently Asked Questions
1. Are Multiple Graphics Cards Worth The Hassle?
Building a rig with multiple GPUs is not worth spending time and resources because modern graphics cards are powerful enough to render most games smoothly, even on 4K.
You should invest in a more sophisticated card instead of paying for two or more weaker ones.
The only situation where a multi-GPU system is worth the effort is if you want to run highly complex machine learning algorithms, financial analyses, or scientific simulations that require vast amounts of memory on a graphics card.
2. Can You Have Three Graphics Cards On A PC?
You can have three graphics cards on your PC if PCI express slots are available on your motherboard because each card must be mounted on one slot.
The slots need to have enough clearance to accommodate the large GPUs.
In addition, you need to have enough space in your case for three graphics cards.
And you need a high-power PSU (around 1000 watts) and an effective cooling mechanism (multiple high-power fans or water cooling).
This setup isn’t typical for regular PC users.
If you’re considering it, think about the use case you want to implement and instead purchase a single, more suitable graphics card.
3. Can You Use Two Graphics Cards With Two Monitors?
Although you can technically use two graphics cards to connect to two monitors on a single machine, the functionality isn’t available out of the box.
One solution is to use virtualization to turn the physical computer into two virtual machines and dedicate each card to one monitor.
This way, it’ll be like you have two different computers.
It’s also possible to mirror the same output on both monitors without virtualization.
However, doing so requires perseverance because the newest versions of Windows and GPU drivers don’t support that scenario.
You should start by plugging in the devices and checking your card’s control panel setting to see what functionality you have at your disposal.
Also, investigate whether your machine supports multi-seating, which lets you turn your computer into multiple workstations.
If you get lucky, the process will take a few clicks, but it’s more likely to take a few hours.
There are no guarantees.
Leave a Reply