Double your PC’s performance! 100% more fps in games! Major games run much better! These are claims that both of the major players in the consumer graphics industry, NVIDIA and AMD, had made in the past. They were both at one time pushing the same technology for consumers. NVIDIA called it SLI, AMD called it crossfire, but they both meant the same thing. Running multiple graphics cards in a single system, for a theoretical doubling of performance compared to single graphics card. AMD has not made a crossfire-compatible graphics card since 2018, and haven’t pushed for game developers to support the technology since long before that. But NVIDIA’s latest and greatest flagship graphics card, the RTX 3090, does technically support SLI over a new interface called NVLink. Seventeen days after the announcement of this graphics card however, NVIDIA announced that they would no longer be developing SLI profiles through drivers in order to “focus efforts on supporting developers to implement SLI natively inside the games.” Has NVIDIA kept their promise? What was so great about multi-gpus in the first place? How did it work? Why did it die? I intend to answer all of these questions in a tribute to one of the most impressive technologies of the mid 2010s.
Ever since its creation, multi GPU technology has relied on one of two different methods to split the work between the two graphics cards in the system. Split frame rendering, (SFR,) divides your screen into two pieces horizontally, so that one graphics card renders the top of the screen, and the other renders the bottom. Later, technology was introduced primarily by NVIDIA to alter where the screen is split to better even the load between the GPUs. For example, if a GPU-intensive rendering item, like foliage, was at the top of the screen, the line would be drawn closer to the top of the screen to give the GPU rendering the foliage an easier time. Oftentimes, this would result in screen-tearing, as it was challenging to sync the progress of the two graphics cards, sometimes creating a horizontal tear line in the center of the screen between frames. Alternate frame rendering, (AFR,) lets one graphics card render an entire frame, then the other graphics card renders the next frame in an alternating pattern. The primary issue with this technology is that it requires extensive driver tuning to make sure that the frames are rendered with the same time intervals in between, because if the frame times are even slightly off, stutter is perceived by the player. As has been well documented by now, both of these methods for multi-gpu rendering required an extensive amount of work either from NVIDIA, or from game manufacturers to work properly.
While AMD and NVIDIA both began by requiring both of the graphics cards to be connected by a custom “bridge” interface to allow the cards to work properly together, AMD made their claim to fame later in the multi-gpu saga by ditching the need for a bridge altogether, only requiring that the cards that are intended to be crossfire be of the same generation and product SKU. Unfortunately, this only seemed to swiften the fall of crossfire, as having many different vendors with all different kinds of motherboards integrate the support for multiple of any number of GPUs quickly became a compatibility and stability nightmare. Eventually, AMD gave up trying to tame the storm, and quietly stopped officially supporting crossfire in 2018, but the writing was on the wall long before that. Nvidia however, never one to stray away from making things more confusing for customers, produced several different “SLI Bridges” throughout the lifespan of the technology, all being nearly identical, and the compatibility of each being terribly difficult to find. The primary things to know about these bridges was that the main improvement being made each time was bandwidth. More importantly, as the bridges gained bandwidth, NVIDIA dropped SLI support for another tier of graphics cards with every generation. For example, the GTX 960 supported SLI, but the follow up, the 1060, did not. Similarly, the GTX 1070 supported SLI, while the standard next-generation RTX 2070, did not. This behavior was consistent all the way up to today, where the only SLI-supported graphics card we have is the RTX 3090. Immediately following the announcement of this card, NVIDIA made the big reveal that they would not be creating SLI driver profiles for any of their current or existing graphics cards, instead, they claimed that game developers would ‘carry on the torch’ of SLI support, and that they would focus on supporting those developers. Since this announcement, exactly zero games have been added to the list of SLI-compatible games on NVIDIA’s website, and game developers seem to have absolutely no interest in creating driver profiles for multiple GPUs in their games. But where does that leave the legacy of this tech?
Due to the continued lack of support and negative media attention, many gamers would consider the legacy of multiple graphics cards to be negative. In practice, I overall agree. Neither NVIDIA or AMD were ever able to get the technology in a state where consumers actually wanted to buy it. Even when the numbers reached a point where they might be considered “worth the extra money,” the limitations of the technology showed. Linus Tech Tips made a video in 2014 titled “SLI Performance Scaling - Gaming at 4K Resolution” that urged viewers to ignore the raw numbers when it game to SLI performance scaling, as there were other factors, such as stuttering and screen tearing, that made many of the games that performed well on paper unpleasant to play. As contrived as it may sound, the reason that I still look back on multi-gpu setups with a positive attitude is the concept and promises that were made. The idea of connecting two graphics cards together and using them to double the performance of your computer will never not inspire awe in me. And while the past ten years have shown us that our technology isn’t ready for this concept yet, I urge the community to remain open to the possibility of this technology returning in the future.
Thanks for reading, I'd love to hear your thoughts and/or feedback in the replies
Ever since its creation, multi GPU technology has relied on one of two different methods to split the work between the two graphics cards in the system. Split frame rendering, (SFR,) divides your screen into two pieces horizontally, so that one graphics card renders the top of the screen, and the other renders the bottom. Later, technology was introduced primarily by NVIDIA to alter where the screen is split to better even the load between the GPUs. For example, if a GPU-intensive rendering item, like foliage, was at the top of the screen, the line would be drawn closer to the top of the screen to give the GPU rendering the foliage an easier time. Oftentimes, this would result in screen-tearing, as it was challenging to sync the progress of the two graphics cards, sometimes creating a horizontal tear line in the center of the screen between frames. Alternate frame rendering, (AFR,) lets one graphics card render an entire frame, then the other graphics card renders the next frame in an alternating pattern. The primary issue with this technology is that it requires extensive driver tuning to make sure that the frames are rendered with the same time intervals in between, because if the frame times are even slightly off, stutter is perceived by the player. As has been well documented by now, both of these methods for multi-gpu rendering required an extensive amount of work either from NVIDIA, or from game manufacturers to work properly.
While AMD and NVIDIA both began by requiring both of the graphics cards to be connected by a custom “bridge” interface to allow the cards to work properly together, AMD made their claim to fame later in the multi-gpu saga by ditching the need for a bridge altogether, only requiring that the cards that are intended to be crossfire be of the same generation and product SKU. Unfortunately, this only seemed to swiften the fall of crossfire, as having many different vendors with all different kinds of motherboards integrate the support for multiple of any number of GPUs quickly became a compatibility and stability nightmare. Eventually, AMD gave up trying to tame the storm, and quietly stopped officially supporting crossfire in 2018, but the writing was on the wall long before that. Nvidia however, never one to stray away from making things more confusing for customers, produced several different “SLI Bridges” throughout the lifespan of the technology, all being nearly identical, and the compatibility of each being terribly difficult to find. The primary things to know about these bridges was that the main improvement being made each time was bandwidth. More importantly, as the bridges gained bandwidth, NVIDIA dropped SLI support for another tier of graphics cards with every generation. For example, the GTX 960 supported SLI, but the follow up, the 1060, did not. Similarly, the GTX 1070 supported SLI, while the standard next-generation RTX 2070, did not. This behavior was consistent all the way up to today, where the only SLI-supported graphics card we have is the RTX 3090. Immediately following the announcement of this card, NVIDIA made the big reveal that they would not be creating SLI driver profiles for any of their current or existing graphics cards, instead, they claimed that game developers would ‘carry on the torch’ of SLI support, and that they would focus on supporting those developers. Since this announcement, exactly zero games have been added to the list of SLI-compatible games on NVIDIA’s website, and game developers seem to have absolutely no interest in creating driver profiles for multiple GPUs in their games. But where does that leave the legacy of this tech?
Due to the continued lack of support and negative media attention, many gamers would consider the legacy of multiple graphics cards to be negative. In practice, I overall agree. Neither NVIDIA or AMD were ever able to get the technology in a state where consumers actually wanted to buy it. Even when the numbers reached a point where they might be considered “worth the extra money,” the limitations of the technology showed. Linus Tech Tips made a video in 2014 titled “SLI Performance Scaling - Gaming at 4K Resolution” that urged viewers to ignore the raw numbers when it game to SLI performance scaling, as there were other factors, such as stuttering and screen tearing, that made many of the games that performed well on paper unpleasant to play. As contrived as it may sound, the reason that I still look back on multi-gpu setups with a positive attitude is the concept and promises that were made. The idea of connecting two graphics cards together and using them to double the performance of your computer will never not inspire awe in me. And while the past ten years have shown us that our technology isn’t ready for this concept yet, I urge the community to remain open to the possibility of this technology returning in the future.
Thanks for reading, I'd love to hear your thoughts and/or feedback in the replies