Up till the mid-2010s, the quickest gaming PCs used a number of graphics playing cards, often two however typically as much as 4. Then, a number of the greatest gaming graphics playing cards used two GPU chips moderately than only one, which helped with consolidation. Nvidia’s SLI and AMD’s CrossFire multi-GPU applied sciences had been seen as the top of any high-end gaming PC and will elevate your gaming expertise to the subsequent stage.
Immediately, multi-GPU is a factor of the previous — virtually a relic within the pc world. The truth that most new GPUs as we speak do not even help SLI or CrossFire is actually an issue, however the recognition of multi-GPUs was dropping off nicely earlier than Nvidia and AMD successfully discontinued these applied sciences. This is the historical past of multi-GPU gaming and why it did not stand the take a look at of time.
A short historical past of multi-GPUs, from 3dfx to its decline
Whereas trendy graphics playing cards emerged within the early 2000s out of the rivalry between Nvidia and AMD, there have been many extra gamers through the ’90s. A kind of firms was 3dfx Interactive, which produced the nostalgic Voodoo line of graphics playing cards. With a view to acquire a aggressive edge, the corporate determined that two graphics playing cards may very well be higher than one, and in 1998, it launched its Scan-Line Interleave know-how (SLI). It was a reasonably genius transfer on 3dfx’s half because it inspired extra GPU gross sales and dissuaded Voodoo homeowners from switching to a different card.
Nevertheless, SLI was launched proper as 3dfx was heading towards chapter, and the corporate was finally acquired by Nvidia, which obtained the mental property rights to all the pieces 3dfx owned. Multi-GPU briefly stopped present after the 3dfx acquisition, however Nvidia reintroduced SLI (altering the official identify to Scalable Hyperlink Interface) in 2004 with its GeForce 6 sequence. It basically labored the identical means because it did earlier than: Add extra GPUs, get extra efficiency. However there have been some improvements with Nvidia’s take.
Whereas 3dfx’s previous SLI had every GPU render a line of pixels one by one (the “scan line” in SLI), Nvidia’s new SLI launched two new rendering strategies: split-frame rendering (SFR) and alternate-frame rendering (AFR). With SFR, every GPU renders a portion of a single body, not by splitting the body down the center however by giving every GPU an equally intensive chunk to render. AFR, alternatively, has every GPU produce a body in flip. Whereas SFR is nice for lowering latency, AFR tends to get essentially the most efficiency, albeit with a lot poorer body pacing and stuttering.
Equally, in 2005, ATI (quickly to be acquired by AMD) launched its personal multi-GPU know-how, known as CrossFire, but it surely was form of a large number at first. With 3dfx and Nvidia playing cards, all you wanted had been two of the identical GPU and a cable or bridge to attach them, however CrossFire required you to purchase a particular “grasp” card along with a daily graphics card. Then, as an alternative of utilizing a bridge, you used a bizarre DVI cable that plugged into each playing cards. Suffice it to say the primary technology of CrossFire was poorly executed. It did not assist that, on the time, its GPUs weren’t superb.
However CrossFire actually got here into its personal with the introduction of AMD’s (previously ATI’s) Radeon 3000 sequence, which featured the Radeon HD 3870 X2, the world’s first graphics card with two GPU chips on it. AMD went actually far with this entire dual-GPU idea; its Radeon 4000 and 5000 sequence chips had been really fairly small, so dual-GPU graphics playing cards made a lot of sense. The HD 5970 in 2009, considered one of AMD’s greatest GPUs of all time, was typically described as too quick to be possible. After this, Nvidia additionally started making its personal dual-GPU playing cards.
After this level, nevertheless, the recognition of multi-GPU started to say no. Nvidia dropped the dual-GPU idea for its mainstream GPUs after the GTX 690 in 2012, and dropped it altogether after the GTX Titan Z in 2014. Nvidia made SLI unique simply two years later to its GTX 1070, 1080, and 1080 Ti GPUs, and it additionally lowered help from 4 graphics playing cards down to 2. SLI was on life help after this, but it surely was lastly axed in 2020 with the launch of the RTX 30 sequence, of which solely the 3090 supported SLI. However that did not matter since Nvidia ceased SLI driver help from 2021 onwards.
In the meantime, AMD saved making dual-GPU playing cards for years, solely stopping with the Professional Vega II in 2019, which was an Apple Mac unique card. AMD even mentioned two RX 480s in CrossFire was a great various to Nvidia’s GTX 1080 in 2016. Nevertheless, AMD ultimately gave up on CrossFire after the launch of RX Vega in 2017, which was the final AMD card to help it. It appears AMD additionally stopped making drivers with per-game CrossFire help someday in 2017 as nicely.
The numerous explanation why multi-GPU died out
Multi-GPU gaming got here and went fairly shortly, all issues thought of. It was solely a big power after 2004 with SLI and CrossFire, however by the 2010s, it was already in a decline. Finally, it was the course the graphics business was going and the way players discovered single GPU options a lot extra interesting that rang the dying knell.
GPUs had been getting greater every technology and ultimately outgrew multi-GPU
When 3dfx launched SLI, graphics playing cards had been tiny gadgets with actually low energy draw, nothing just like the behemoths we see as we speak. Graphics chips tended to be about 100mm2 massive within the 90s and early 2000s, however this all modified when ATI launched its Radeon 9000 sequence, which featured a chip that was over 200mm2, double the dimensions of something the world had seen earlier than. This began a GPU arms race that ATI/AMD and Nvidia saved escalating with every technology.
The factor is, bigger chips require extra energy and higher cooling, and whereas elevated energy draw did not actually influence multi-GPU setups at first, it will definitely proved to be a big downside. At the same time as early because the GTX 480, graphics playing cards had gotten to the 250W mark, and two 480s in SLI consumed an unbelievable quantity of energy. Whereas AMD put vital emphasis on multi-GPU with its HD 4000 and 5000 sequence, it was actually solely as a result of it wanted one thing high-end to go towards Nvidia’s 480 and 580 since AMD’s graphics chips had been too midrange.
From the late 2000s on, nearly each flagship made by Nvidia and AMD consumed a minimum of 200W, typically 250W. It won’t be a coincidence that Nvidia’s final mainstream dual-GPU card, the 690, used two GTX 680 chips, which had a TDP of solely 195W. The easy incontrovertible fact that single GPUs had been getting greater and higher made SLI and CrossFire harder and fewer interesting to customers, who often did not need their gaming PC to even be an area heater and a jet engine.
Multi-GPU was buggy and required devs, Nvidia, and AMD to speculate assets into it
{Hardware} developments had been an issue for multi-GPU’s feasibility, and so had been software program developments. Again when SLI was first launched, video games had been a lot easier, and even 2004’s greatest video games, equivalent to Half-Life 2, are fairly unremarkable in comparison with as we speak’s video games, although we are able to respect how nice they had been after they got here out. SLI and CrossFire required Nvidia and AMD to create particular optimizations for multi-GPU of their drivers with a purpose to obtain good efficiency, and again then, this wasn’t a giant deal.
However over time, video games (and, by extension, GPUs) received extra difficult, and it turned more durable to optimize every year. Even in titles that had official help for multi-GPU, the expertise was typically subpar as a consequence of poorer-than-normal efficiency or bugs. For a short time in 2016, I had two Radeon R9 380s, and after I performed The Witcher 3, I typically noticed bizarre graphical glitches that typically even lined up essential options like cave entrances, making the sport not simply quirky however buggy to the purpose of being unplayable.
Finally, it was the course the graphics business was going and the way players discovered single GPU options a lot extra interesting that rang the dying knell.
The one glimmer of hope for higher software program help for multi-GPU was DX12 and Vulkan, which boasted such highly effective multi-GPU help that you might even use a number of GPUs from completely different distributors in a single recreation. Nevertheless, this simply offloaded the work Nvidia and AMD used to do into the fingers of builders, who did not stand to achieve something by supporting multi-GPU know-how, particularly since Nvidia and AMD had been phasing it out. So, the software program facet of issues did not pan out both for multi-GPU gaming.
Players simply did not want high-end multi-GPU setups
Even when issues on the {hardware} and software program sides of the equation labored out, multi-GPU gaming might need been doomed just because it was overkill. Even the HD 5970 was described as overkill, and that was simply with two midrange GPU chips. Nonetheless, multi-GPU was simply widespread sufficient to maintain going for years, however I feel its destiny was determined by one single occasion: the launch of the GTX 1080 in 2016.
Nvidia’s GTX 10 sequence was actually simply the GTX 9 sequence on the brand-new 16nm from TSMC, however that alone was a giant deal since Nvidia had spent three entire generations on 28nm as a result of decline of Moore’s Legislation. Going from 28nm to 16nm resulted within the GTX 1080 being over 50% sooner than the GTX 980 and 30% sooner than the GTX 980 Ti. The 1080 additionally supported SLI and its TDP was comparatively low at 180W, however the uncooked efficiency with a single 1080 was insane in 2016.
Whereas PC gaming utilizing a number of graphics is seemingly by no means coming again, the door for multi-GPU is definitely open.
This was additional improved with the GTX 1080 Ti the subsequent yr, boosting efficiency by practically one other 30%. A single 1080Ti was practically twice as quick as a 980 Ti and would have actually been a superior resolution to 2 980 Tis in SLI. No person of their proper thoughts would really need two 1080 Tis in SLI, not solely as a result of it might have been sizzling and loud but additionally as a result of twice the efficiency of a 1080 Ti would have been utterly overkill (and in addition not possible for many video games with official SLI help). Think about how loopy it might be to have two RTX 4090s in SLI.
Multi-GPU gaming may make a comeback
Whereas PC gaming utilizing a number of graphics is seemingly by no means coming again, the door for multi-GPU is definitely open. In case you’re conversant in AMD’s CPUs, you may know that its higher-end desktop chips and all its workstation and server CPUs use a number of CPU chips collectively as an alternative of 1 massive CPU. Utilizing a lot of smaller chips (also referred to as chiplets) is a know-how AMD began utilizing again in 2019, although solely in 2022 did it begin utilizing chiplets for its GPUs with the introduction of the high-end RX 7000 sequence.
RX 7000 playing cards just like the RX 7900 XTX, nevertheless, solely have a number of cache and reminiscence chiplets, and use a single GPU chip. Nonetheless, there’s motive to consider AMD would possibly begin utilizing a number of graphics chiplets since it might lower down growth and manufacturing prices whereas additionally making it easy to make new playing cards (simply add or take away a chiplet and bam, new GPU). Intel may additionally go the identical course because it, too, is transitioning to chiplets.
Whereas it appears Nvidia has completely little interest in chiplets, it might be stunning if AMD and Intel weren’t desirous about bringing again multi-GPU with chiplets. Maybe we’ll see the return of multi-GPU gaming with trendy know-how within the coming years if it may work nicely sufficient.
