What is SLI? How does it work? Is it better than a single card?

Scalable Link Interface Everything We Know

Once, long ago in the dark ages of tech, getting more video card power meant having to buy a card more powerful (and probably more expensive!) than your current card. Wouldn’t it be great if you could just buy a duplicate of the card you have now, instead of spending all that money on something newer and fast?

Well, thankfully Nvidia and their multi-GPU technology “SLI” gives us just the solution if you’re using one of their cards.

More: Best Graphics CardsBest Graphics Card Deals

Watch Our Video Rundown

Scalable Link Interface (SLI)

So what is SLI? And what does SLI do, or more specifically: how does it help me run multiple graphics cards to improve my gaming? Scalable Link Interface (SLI) is a little gem given to use from the masterminds over at Nvidia in 1998.

This technology was first used by the company 3dfx on their Voodoo2 graphics card line in 1998, but later the company was bought by Nvidia and the technology was shelved for a while.

In 2004, Nvidia re-released SLI with updates to allow it to work with then-current technology of PCIe buses.

Although there are compatibility concerns (discussed below), associated with running multiple GPU setups, SLI will allow the use of two to four cards at once. It achieves this by using a parallel processing algorithm, which takes the information, breaks it down into smaller pieces and allows for multiple pieces to be worked on at once by multiple cards.

SLI Modes

Nvidia graphics cards will work together in SLI in three main modes: SLI anti-aliasing, SFR, and AFR. You can select and change these modes in the Nvidia Control Panel, which should have been installed onto your computer when you installed your video card drivers.

Split Frame Rendering (SFR)

This method takes the frame that needs to be rendered, analyzes it, and then splits up the frame horizontally into as many pieces as there are cards, and sends the workload between them. It can split the frame how it sees fit based on what’s in the frame.

For example, if the frame contains a mostly empty, static sky in the top half of the frame (less resource intensive), but moving characters in the bottom half (more intensive), then it will determine that one card would be doing more work if it is merely split 50/50. So, the card responsible for the upper portion of the screen may receive about 60% of the screen real estate while the card responsible for the bottom will only have to process the bottom 40%. (Note: these percentages are just vague examples, not concrete numbers.)

Alternate Frame Rendering (AFR)

In this mode, the frames are queued up and each card renders a full frame. So, if you are running 2 cards, then one would be responsible for even numbered frames while the other would be responsible for odd numbered frames. This is often considered more effective than SFR for achieving higher frame rates, but it can also cause issues such as micro stuttering.

Keyword: Micro Stuttering – This it when you see visual stutters on your screen caused by a discrepancy in frame rendering times between cards. If the cards are not synced up properly, or they don’t have enough information to guess the next frame that needs to be rendered, then you will see a delay.

SLI anti-aliasing

This mode allows for a better image quality, as opposed to pushing higher frame rates. It works to split the anti-aliasing workload between the cards, increasing the overall quality of your image.

Using this mode can unlock higher anti-aliasing modes within games (such as SLI 8x, 16x, and even 32x on quad-SLI systems).

Keyword: Anti-aliasing – Computer pixels are made of squares, which can result in round images looking jagged around their edges. Anti-aliasing helps fix this by sampling pixels around the problem area and then filling in the best approximate guess.

Compatibility and Issues

Alright, so all you need to do is buy two graphics cards? Do I just slap any two Nvidia cards into my motherboard and call it a day? Unfortunately, it’s not that simple. But then again, it never is…

Two (or more) of a Kind

The first thing you need to know is what graphics cards will pair together; not just any two cards will work. They have to be the same GPU, and they have to have the same amount of video RAM.

For example, if you had a GTX 1070 TI 8GB made by Asus, and another of the exact same specs, but made by MSI, it would still work. A GTX 1070 and a GTX 1080 are not compatible, as they are not the same GPU.

You will also need to make sure that the graphics cards are SLI compatible themselves, as cards without the SLI port on top may not be compatible. Also, it is worth noting that Nvidia killed support for 3 and 4 card SLI for the GTX 10 series.

Slots and Plugs

Next on the list is your motherboard and power supply. These will both need to be SLI compatible.

This means that your motherboard will need enough PCIe x16 slots to match your number of GPUs and your power supply will need enough PCIe connectors to power the amount of cards you will be using.

Seeing Double

A common misconception about SLI is that you can get double, triple, or even quadruple video RAM with more graphics cards. Unfortunately, Nvidia SLI only uses the RAM from one card, as each card needs to access the same information at the same time.

EVGA GEFORCE GTX

What is the SLI Bridge?

The last thing you will need in order to run your SLI setup is an SLI bridge. Nvidia uses a physical connector to bridge the graphics cards together, allowing them to communicate with each other without using precious bandwidth on the PCIe slots.

You will need one of two SLI bridges: either the standard bridge (for less powerful cards), or a high-bandwidth bridge (for the more powerful cards).

SLI Bridge Series

BridgePixel Clock SpeedMax BandwidthRecommended Monitor Support
Standard Bridge400 Mhz1GB/s1920x1080 through 2560x1440 @60 Hz
High-Bandwidth Bridge650 Mhz2GB/sUp to 5k and surround sound

If you have more powerful cards (like say the GTX 1080), you can use a standard bridge, but it won’t allow for the full performance of the cards. The high-bandwidth brides would allow for this.

Performance

So we know what SLI is, how it works, and what you need to use it. But should you rush off to buy that second and third GTX 1080 TI, and have to tell your wife that the whole baby will have to wait?

Well, that’s where the waters get a bit murky. While Nvidia claims that twice the performance is possible with twice the cards, that’s rarely ever the case.

The problem is, games don’t natively know how to properly use multiple video cards, and SLI profiles have to be added into them by the developers. This is not an easy task, and can be very time consuming, so most developers don’t take the time.

This means that unless you’re playing a AAA game, it’s possible that there will be no performance increase (in fact, there’s even the possibility of a decrease in performance).

Performance for each game will vary, though. So that doesn’t mean it will be completely useless for you, just that you should be aware of this before investing in the technology or any specific games. You can see a list of SLI certified games here, just in case.

But don’t let us be the final voice on the matter. Is SLI a viable proposition for you, or are you going to just save up for a completely new graphics card?

Sources

https://en.wikipedia.org/wiki/Scalable_Link_Interface

FAQs

What is the benefit of SLI?

SLI helps you achieve slicker, smoother, more altogether mind-boggling graphics in your gaming. It’s a technology that allows you to use two or more video cards in the same set-up, using their potential as an aggregated whole, rather than just letting them each act to the maximum of their own performance.

You can connect up to four GPUs via SLI, though you have to be aware of diminishing returns and a host of potential complications with heat and bugs. The more GPUs you add, the more points at which the system can fail and the more generated heat you have to mitigate.

Is NVLink better than SLI?

NVLink and SLI are differently arranged, but whether either is necessarily better is a more complex question.

The chief difference between them is that in SLI, there’s a master-slave relationship. One card governs and directs the workload of the others. In NVLink, there’s no such hierarchy, communication is bi-directional and each card can access the memory of all the others.

That has an important impact when it comes to available memory. The hierarchical relationship binds the maximum available memory to that of the master card – so if you SLI two 8GB GPU cards, you won’t have 16GB available to you.

On the other hand, if you NVLink the cards, the lack of that binding means you can legitimately double the amount of memory you have available.

So in theory, sure, NVLink is better than SLI. However, it’s only actually better if you can make use of it in a gaming environment – and at the moment, game developers have yet to make games that take advantage of a multi-FPU set-up.

So yes, it could be better. And it may well be better in the future. But for right now, there’s not much noticeable difference between the two set-ups.

Does SLI increase performance?

Yes and no. Technically, it can improve performance and frame rate. But first of all, you need games that take advantage of it if, and there are very few that do. And secondly, it seems as though SLI as a technology may be on its way towards wide-scale redundancy. As of January 2021, all GPU support for SLI will need to be explicit – meaning games will have to support the mode directly, rather than, as previously, being able to implement the mode in-driver.

So, technically, yes, it can improve performance, but you need games to take advantage of it, and game developers are unlikely to make games that can use it because of an increasingly marginal demand.

Can you run 2 GPUs without SLI?

You can, though often the ultimate question becomes why you would.

Whereas an SLI will allow you to use both GPUs on a single screen, if you use two GPUs without an SLI, you will usually need two monitors attached. Then you can create a kind of ‘multi-room’ experience – watch a movie on one, play a game on the other, simultaneously. But they won’t be working in tandem, you’ll just be using two GPUs to run two separate graphic outputs. The point of which is dubious.

Can you boost the output of a single graphical channel, like a game, using multiple GPUs, without an SLI? Again, the answer’s less than wholly satisfactory – you can if you use something like NVLink or a sophisticated software package. An NVLink is sort of cheating on the spirit of the question, because it’s another version of an SLI, only connecting GPUs non-hierarchically. And the software that’s available to help you run multiple GPUs to the same output is essentially mimicking the action of an SLI.

Can you SLI 2 different brands?

Think of it this way. If Nvidia makes a round peg and a square peg, and AMD makes a square peg and a round peg, you can use two round pegs, irrespective of who makes them. Or you can use two square pegs, irrespective of who makes them. But if you try and use one round peg and one square peg, you’ll waste a lot of money and be very, very unhappy.

The reason is straightforward. SLI exists to yoke cards of identical character – internal architecture, core clocks, shaders, buses, you name it, the SLI is looking for uniformity of structure. If that uniformity of structure exists in GPUs made by different companies, the SLI does not care – it can link them. If you use two different models, even though they’re made by the same manufacturers, all that internal consistency is missing or mismatched. The result? Malfunction.

So – same model, different brand? Thumbs up. Different model, any brand? Thumbs down.

Does SLI bridge brand matter?

In the vast majority of cases, the answer to this is a simple no. Some unit-builders claim they notice a brand-specific difference, but not until they reach very high levels of modification. If you start using all your cards to the maximum of their potential, some builders claim that some brands will give you access to higher frame rates per second. There’s also an argument over the superiority of rigid SLI bridges over flexible ones, but this is largely subjective.

Overall, there’s little enough difference between bridge brand effectiveness to discount it as a factor in your overall set-up. In fact, its biggest impact is likely to be on the overall aesthetic of your set-up, rather than anything especially measurable in performance terms.

How do I test if SLI is working?

Handily, there’s a visible on-screen bar with the giveaway text “SLI” that pops up on the left of your screen if your SLI is activated and working. If you’re in full-screen mode, that’s reduced to a non-intrusive green square. No image? No SLI.

To double-check or definitively switch on your SLI, open up your 3D settings, and select Set SLI Configuration. If you have no on-screen indicator, again in 3D Settings, click on Show SLI Visual Indicator.