What’s better than one 1080Ti? Two of them. Well, maybe. Under certain conditions. For older games. Sometimes.
Welcome to Esk Computers, I’m Scott and today on the desk of Esk, let’s run two 1080Ti’s in SLI at 4K and see if is worth the trouble, and the power draw.
I’ve been mentioning SLI in every 1080Ti video on the channel, today I finally get that out of my system by throwing another 1080Ti into the system we built last week. I know, I know, there arguably wasn’t much point doing this in the past and there’s certainly no point doing this anymore, as has been proven by many a YouTube video, but, well this one is mine.
I’m old enough to remember the before-times when SLI used to mean scan line interleave, as the 1% of the late nineties and early two thousands could watch their two 3DFx Voodoo cards each render out alternate horizontal lines of the scene. Video rendering pipelines have changed a lot since NVidia bought 3DFx, so it now means Scalable Link Interface, and splits the rendering in more complex ways, sometimes at the cost of introducing microstuttering in games. So higher framerates, but a less even frame pacing, which might not be a great compromise. This may be in part why Nvidia sunset this flavour of the technology back in 2020, but we’ll test that in due course.
I didn’t film plugging the card in, but there’s nothing too visually exciting there that you’ve missed. I also didn’t film the first power on, but all you’ve missed there is me yelping like an overstimulated puppy. As a reminder, this is an i9-9900K system, currently running with 16GB of 3000MHz DDR4 RAM.
The great thing about this EVGA 1000W PSU is that I can run four separate power cables to the 1080Ti’s which is reassuring given that the cards will be pulling about 250W each. Much better split four ways than two, as there’s enough heat issues already.
They’re rather closely packed, so that’s not doing the airflow to the top card a lot of favours. However, with both running in Cinebench, one card settles out in the mid to low seventies and the other in the low sixties, so that’s bearable. However, after extended playtesting sessions I’ve seen it eventually getting up past eighty degrees and close to the card’s thermal limits, which I think is because it’s raising the ambient room temperature that much. It’s basically acting as a 600W space heater, so it is an airflow issue, but less with this case and more with this room. I’d need to address this if it was a permanent installation, but it is not, so I’ll live with it in the short term.
These results come after doing a very slight bit of tuning to both in MSI Afterburner, adding 250Mhz on the memory clocks and 75Mhz to the core speeds, which should be over this model’s stock boost clock speeds of 1683Mhz and actual memory speed of 1378MHz. Like all 1080Ti’s, it sports 3584 CUDA Units, 11GB of GDDR5X on a 352-bit memory bus, on a PCI Express x16 3.0 slot. These models give you two DisplayPort 1.4 ports, two HDMI 2.0 ports and a DVI port.
I’m going to test twenty odd games going back twenty years to see how effective SLI was, and how ineffective it now is. To keep things spicy today’s testing is going to be at 4K resolution, so we can also test if a single 1080Ti had what it takes to hang at that resolution in the past, and for how long that lasted. First up though, a few synthetics to make sure it’s working at all.
Cinebench 2024’s GPU test had a single card score of 5570 points, with the two cards we get 10329, an 85% improvement.
Fire Strike gets a slightly disappointing 29% bump with SLI, going from 25159 to 32352 points.
Time Spy is much better, increasing 67% from 10612 to 17680 points.
Steel Nomad is kind enough to warn us beforehand that it doesn’t support dual card setups, and indeed it gets much the same scores of 2333 with one card and 2329 in SLI.
So, that’s pretty much the gamut of possible outcomes, from close to 2x the performance, to modest improvements that are arguably not worth the hassle, to no benefit whatsoever, and indeed a slight performance hit from the overheads. Still, let’s get into the games, and again these were all at a crisp 4K resolution, although of course you’re looking at a 1080p capture and video render.
Nothing ages a man like realising how old some old favourites are. 2006 brought us The Elder Scrolls IV: Oblivion, this of course being the original release and not the recent remaster. At Ultra settings, and with every draw distance slider maxxed, with one card it’s getting 177.1 average fps in game, with 1% lows of 63.7 and 0.1% lows of 37.5 fps. In SLI, it’s actually 14% slower we get 152.6 average fps in game, with 1% lows of 50.8 and 0.1% lows of 17.4 fps.
We’ll see this in a few earlier games, where it is spreading the rendering load across two cards, but doesn’t wind up improving anything because one 1080Ti alone is enough to max the engine out, and all we’re getting is more overhead and worse 1 and 0.1% lows.
But can it run Crysis hahahaha? Yes, it can, shut up. 2007’s system melter, again, not the remaster, at Very High settings with one card gets 64.9 average fps, with 1% lows of 54.6 and 0.1% lows of 5.9 fps. In SLI, it’s 94% faster, with 125.8 fps, with 1% lows of 90.3 and 0.1% lows of 5.9 fps. Now, that’s what we’re talking about.
Continuing the Crytek theme, sorta, with 2008’s Far Cry 2 – let’s not get too deep into engine customisation discussions – here at Ultra settings, with one card we get 130 average fps, with 1% lows of 94.7 and 0.1% lows of 87.5 fps. In SLI, it’s 1.9% slower, with 127.5 average fps, with 1% lows of 85.4 and 0.1% lows of 46.7 fps. Disappointing, but when it’s getting over 100 fps it’s hard to be too upset.
Another from 2008, just because, it’s Fallout 3 at Ultra settings where with one card we get 101.8 average fps, with 1% lows of 47.1 and 0.1% lows of 29.6 fps. In SLI, it’s 1.6% slower, with 100.2 average fps, with 1% lows of 45.4 and 0.1% lows of 32.0 fps. Same issue as with Far Cry 2, one 1080Ti’s more than enough for this game.
2009’s Resident Evil 5 was the sort of difficult second album to RE4’s reinvigoration of the franchise, and it seems to be capped at 120 in game but has an uncapped benchmark. At High settings, which is as high as it goes, with one card we get 243.1 average fps, with 1% lows of 196.2 and 0.1% lows of 115.4 fps. In SLI, it’s 30.0% slower, with 170.1 average fps, with 1% lows of 132.8 and 0.1% lows of 65.5 fps. That’s weirdly bad, but again, one 1080Ti’s plenty for this game.
Also from 2009, this is the excellent Batman: Arkham Asylum. It’s at Very High settings, and although it swears VSync is off, it either isn’t or there’s an engine framerate cap. With one card we get 61.8 average fps, with 1% lows of 47.6 and 0.1% lows of 24.6 fps. In SLI, it’s essentially performing identically, with 61.9 average fps, with 1% lows of 52.0 and 0.1% lows of 42.2 fps.
2010 brings us the demanding Metro 2033, again, the OG, not the later remaster. This is under DirectX11 at Very High settings, and with one card we get 36.6 average fps, with 1% lows of 28 and 0.1% lows of 26 fps. In SLI, it’s 101.1% faster, with 73.6 average fps, with 1% lows of 62.4 and 0.1% lows of 21.9 fps. You can’t ask for better scaling than that, and it’s made a borderline 4K experience into a good one. Should really stop here, we’re not going to get a better result than this.
Battlefield 3 from 2011 comes close, though. At Ultra settings, with one card we get 93.4 average fps, with 1% lows of 83.5 and 0.1% lows of 83.2 fps. In SLI, it’s 75.7% faster, with 164.1 average fps, with 1% lows of 139.3 and 0.1% lows of 102.3 fps. It’s hampered only by the one card being good enough, but a particular highlight here is how the percent low framerates get a huge bump too, which is normally not the case.
Also from 2011, the original release of the many times rereleased Skyrim. The internet tells me it had a rough launch, SLI wise, but in 2026 with one card at Ultra settings, we get 155.9 average fps, with 1% lows of 116.1 and 0.1% lows of 87.5 fps. In SLI, it’s 2.8% slower, with 151.6 average fps, with 1% lows of 108.8 and 0.1% lows of 29.4 fps. What’s confusing me a little while working through the data was I read this OG release engine had a 60 fps limit, and I’ve not modded it or anything. Weird, but I’ll take it.
I still don’t really get the Borderlands games. I mean, in a way I get them, eventually, when they go on sale hard enough, but I don’t get the appeal of them is what I mean. This is 2012’s Borderlands 2, with everything as high as it can go, and with one card we get 117.7 average fps, with 1% lows of 88.6 and 0.1% lows of 70.6 fps. In SLI, it’s 22.3% faster, with 144.0 average fps, with 1% lows of 82.6 and 0.1% lows of 51.8 fps. A noticeable improvement in the averages but with an equally noticeable drop in the percent lows, so not sure this is actually a win, overall.
But can it run Crysis 3 hahahaha? Yes, it can, shut up. 2013’s system melter, again, not the remaster -whoa, deja vu – where at Very High settings, with one card we get 42.4 average fps, with 1% lows of 34.6 and 0.1% lows of 24.4 fps. In SLI, it’s 47.6% faster, with 62.6 average fps, with 1% lows of 36.0 and 0.1% lows of 32.3 fps. As with Metro 2033, it’s made a borderline 4K experience into a good one. As an aside, doesn’t this still look great at thirteen years old?
This is 2014’s patent office bothering Middle-earth: Shadow of Mordor, at Ultra settings, and with one card we get 98.0 average fps, with 1% lows of 85.4 and 0.1% lows of 77.3 fps. In SLI, it’s technically 0.7% slower, with 97.3 average fps, with 1% lows of 80.5 and 0.1% lows of 51.9 fps. This was supposed to perform better in SLI, from my pre-testing reading, but this is what I got. Such is life.
Another from 2014 with Dragon Age: Inquisition which is much more in line with my hopes and a pleasant surprise. At Ultra settings, with one card we get 62.3 average fps, with 1% lows of 52.8 and 0.1% lows of 42.3 fps. In SLI, it’s 69.3% faster, with 105.5 average fps, with 1% lows of 74.6 and 0.1% lows of 58.3 fps. This was supposed to perform disappointingly in SLI, from my pre-testing reading, but this is what I got. Such is life.
I’m sure we’re all anticipating Grand Theft Auto VI after the 11 year wait, but 2015’s original non-enhanced flavour of Grand Theft Auto V is still fun, and looking great at 4K. At Very High settings, with one card we get 96.3 average fps, with 1% lows of 75.0 and 0.1% lows of 68.4 fps. In SLI, it’s 47.4% faster, with 141.9 average fps, with 1% lows of 98.2 and 0.1% lows of 50.2 fps.
Let’s look at a few more of the big hitters 2015. This is the unenhanced, unremade The Witcher 3: Wild Hunt, running in DirectX 11. Annoyingly my saves are trapped in the Enhanced edition, so in the opening section at Ultra settings, with one card we get 59.0 average fps, with 1% lows of 51.2 and 0.1% lows of 22.9 fps. In SLI, it’s 38.6% faster, with 81.8 average fps, with 1% lows of 59.9 and 0.1% lows of 51.6 fps
Closing out 2015’s games with Fallout 4, at Ultra settings, with one card we get 94.0 average fps, with 1% lows of 61.2 and 0.1% lows of 54.7 fps. In SLI, it’s only 7.2% faster, with 100.8 average fps, with 1% lows of 60.9 and 0.1% lows of 31.8 fps. Not worth the effort, really.
This is 2016’s Rise of the Tomb Raider, at Very High settings in DirectX11, as I couldn’t get DIrectX12 to launch with SLI enabled. With one card we get 68.8 average fps, with 1% lows of 63.1 and 0.1% lows of 62.9 fps. In SLI, it’s a pleasing 67.2% faster, with 115.0 average fps, with 1% lows of 71.0 and 0.1% lows of 48.5 fps.
2017 brings us Ghost Recon Wildlands, where at Ultra settings, with one card we get 40.0 average fps, with 1% lows of 37.9 and 0.1% lows of 37.6 fps. In SLI, it’s 21.8% faster, with 48.7 average fps, with 1% lows of 40.7 and 0.1% lows of 35.4 fps.
More Ubisofting now with 2018’s Far Cry 5, where at High settings, with one card we get 63.9 average fps, with 1% lows of 54.6 and 0.1% lows of 23.5 fps. In SLI, it’s 38.5% faster, with 88.5 average fps, with 1% lows of 67.5 and 0.1% lows of 54.2 fps.
I wonder if 2019’s Strange Brigade is ever played by anyone that not benchmarking stuff. At least it’s not as bad as Ashes of the Singularity for that. At High settings, with one card we get 106.7 average fps, with 1% lows of 88.2 and 0.1% lows of 80.0 fps. In SLI, it’s 66.4% faster, with 177.6 average fps, with 1% lows of 100.8 and 0.1% lows of 96.0 fps.
Here’s 2020’s Red Dead Redemption 2 which needs little introduction. This was another game that was supposed to give decent SLI results, but for me at least, didn’t. Under Direct X12 at Favour Quality settings, with one card we get 42.7 average fps, with 1% lows of 35.2 and 0.1% lows of 34.6 fps. In SLI, it’s 5.4% slower, with 40.4 average fps, with 1% lows of 33.7 and 0.1% lows of 31.2 fps. I thought I’d try under Vulcan, and the results we worse. More experimentation required.
2021 brings us Forza Horizon 5, and also apparently the cliff edge of good SLI results. Thankfully it’s a really well optimised game such that with one 1080Ti at Ultra settings it’s hitting 70 average fps in game, with 1% lows of 60.5 and 0.1% lows of 55.8 fps. However, it makes no use of the second 1080Ti at all, in fact taking a slight hit with 67.6 average fps in game, with 1% lows of 57.6 and 0.1% lows of 54.1 fps. It’s the same with all these newer games results, the second card might as well not be there. I’ll leave the details on screen for interest and posterity, but I’m not going to repeat myself speaking them aloud.
However, I can imagine the single card 1080Ti 4K results could be of interest, so let’s plough on with those. I’ve tried to give the 1080Ti a fighting chance, so they’re not at the highest settings possible. This is 2022’s A Plague Tale: Requiem at High settings, where we get 29.7 average fps, with 1% lows of 22.8 and 0.1% lows of 14.3 fps.
Here’s 2023’s Starfield at Medium settings and native resolution, where we get 25.7 average fps, with 1% lows of 22.1 and 0.1% lows of 19.0 fps.
I wanted at least some Unreal Engine 5 representation, so this is 2024 Black Myth: Wukong benchmark tool. At Medium settings, native resolution and with no RT, we get 19.0 average fps, with low 5th percentiles of 16.0 fps and a minimum of 10.0 fps.
Staying in 2024, Senua’s Saga: Hellblade II at Medium settings, native resolution, we get 18.7 average fps, with 1% lows of 15.4 and 0.1% lows of 13.6 fps.
2025’s Metal Gear Solid Delta: Snake Eater at Medium settings gets 14.3 average fps, with 1% lows of 8.7 and 0.1% lows of 7.1 fps.
Also from 2025, Silent Hill f at Balanced settings gets us 19.3 average fps, with 1% lows of 16.0 and 0.1% lows of 13.9 fps.
And rounding off this testing with 2026’s Death Stranding 2 at Medium settings, where we get 26.1 average fps, with 1% lows of 23.4 and 0.1% lows of 22.5 fps.
Now, I’m not going to lie, a lot of these examples had been cherry-picked to fit a narrative, and there’s plenty more examples of games even from the times when SLI was, relatively speaking, more common that don’t play nicely. I’ve tried to show this a bit, like in 2015 which had good examples like GTA V or The Witcher 3, but there’s also buggy or just not worth the hassle implementations like Fallout 4.
Now we have the numbers, I think it’s fair to say the 1080Ti is not best card for this experiment. We’ve seen plenty of earlier titles where one 1080Ti pretty much maxes out what it’s asking for, so having two cards sitting at 30% utilisation isn’t much benefit.
And, of course, we could have picked practically every game released this decade as an example of it not working, not just the ones tested. Annoyingly, a lot of those were the games that could have used an SLI-based shot in the arm to stay relevant, even if not at 4K Ultra settings. So, yes, sadly, SLI dead in 2026 reconfimed for the millionth time, dead horse beaten yet again.
Also while I’m certain this is the correct SLI bridge, Afterbuner was frequently showing the SLI link speed as being a performance limit. I didn’t have time to properly check that out, but I’m quite certain it wouldn’t make enough of a difference to change the outcome of the video.
I’m not trying to argue that you should go build an SLI setup if you want to get the best gaming performance, either now, or then, if you have access to a time portal taking you back to the early 2010’s. It is, however, a very interesting area to play around in, for me at least, and I suppose if you’ve watched this far, it is for you too.
Now I have this set up, I would like to test out one of my earlier theoretical upgrade paths and see if two GTX670s in SLI can get into the same performance ballpark as one GTX 960. This would also skirt around the “single card being too powerful by itself” stumbling block we’ve seen with the 1080Ti. So, I guess I’ll get back on eBay but in the meantime, if you have any questions or want further details please leave a comment down below, and if you enjoyed this videotronic missive then consider subscribing. Until next time, take care of yourself, and each other.