The 4K 60fps GPU for 2018 and beyond.
Nvidia’s all-new 20-series GPUs have finally arrived, with the company announcing three models so far: the RTX 2080, the beefier RTX 2080 Ti, and the more affordable RTX 2070 (which is coming further down the road). Today I’m looking at the RTX 2080 Founder’s Edition, which is the successor to Nvidia’s popular GTX 1080 card. Built on Nvidia’s new Turing architecture, this graphics card promises increased performance over Pascal along with all-new hardware that enables real-time ray tracing and an all-new version of AI-assisted super sampling (in the future). I spent the last few days with this GPU putting it through its paces, and here’s my full report, including benchmarks in a variety of today’s popular PC games.
RTX 2080 – Design and Features
Nvidia claims its Turing architecture is the biggest technological advancement the company has made in over a decade, and considering the fascinating potential of these new graphics cards, it’s a difficult point to argue. These are the first GPUs to offer the long-awaited technology to power real-time ray tracing, where in-game lighting is cast in a wholly new, much more realistic way. Instead of having light basically painted into a game by designers, ray tracing renders lighting, shadows, and reflections just like it behaves in the real world, and is massively difficult to pull off due to the amount of calculations involved. The fact that it can be done in any manner, in real-time, is remarkable.
Sadly, as of this writing, there are no games currently running the technology for me to test. However, there’s a number of games in the pipeline, including Battlefield V and Shadow of the Tomb Raider. The GPU hardware supports it, but game software still has to follow suit, and Microsoft needs to update Windows 10 to support it as well. While the delay on seeing ray tracing in a testing environment is unfortunate, when IGN editors went hands-on at Gamescom it looked pretty snazzy.
The second big advancement with Turing is Deep Learning Super Sampling, or DLSS. This is essentially a new method of anti-aliasing that promises to offer profound performance boosts over typical methods such as TAA. It’s a complicated technology, but essentially Nvidia will be feeding “perfect” images to a neural network, which it then uses to learn how to reconstruct frames with much less overhead, freeing up the load on the GPU. Typical anti-aliasing is a real GPU killer, so having the AA work done by the Tensor cores in Turing can theoretically offer improved image quality without the typical performance penalty. Of course, one of the roadblocks of DLSS, like any new technology, is getting developers on board. Nvidia has already announced some noteworthy games supporting the technology, with more to come in the future.
As for how it’s enabled for games, Nvidia will essentially have to issue a DLSS patch for each game after their neural network has done its job, sort of like how it would offer updates for SLI support in driver updates in the past. If game developers take advantage of the tech, it could offer a big boost to GPU performance in the future, but it’s another function that could not be tested at launch.
As was the case with the GTX 1080, Nvidia has released a Founder’s Edition of the RTX 2080, which basically means its a first-party graphics card made by Nvidia. The company’s partners will release their own versions of the RTX 2080 with varying degrees of over-clocking and cooling options. The RTX 2080 Founder’s Edition does offer some improvements over the previous versions though. Namely, it’s moved to a dual-fan cooling setup instead of a single-fan “blower” design, which should provide improved cooling performance.
In terms of aesthetics, the entire card is encased in a silver metallic shroud which looks really slick. On-board lighting is limited to the green glow from an RTX 2080 logo on the card’s side though, so there’s no RGB here. The only downside to the Founder’s Editions is that they are supposed to cost a bit more than partner cards. However, due to insane demand for all the RTX cards right now, currently even partner boards are the same price as the Founder’s Edition – $799 – or higher, but that might change if demand dies down a bit.
The RTX 2080 features 2,944 CUDA cores, a 14 percent increase over its predecessor, the GTX 1080 Founder’s Edition. Total memory is 8GB of 14Gbps GDDR6, but the RTX 2080 has an improved memory clock of 7000 MHz (compared to 5000 MHz) and GPU clock of 1800 MHz (compared to a 1733 MHz clock on the GTX 1080). There’s also a whopping 13.6 billion transistors in the RTX 2080’s TU104 chip, which is even more transistors than that of Nvidia’s powerful Titan XP.
Obviously, Turing also introduces some new components to render real-time ray tracing and to utilize the new DLSS AI technology. It has 46 RT cores, and 368 Tensor cores to handle both of these new technologies. That new architecture does require a bit more power, and the RTX 2080 has a Thermal Design Power (TDP) of 225 watts; a increase of over 20 percent from the GTX 1080. Accordingly, the RTX 2080 also requires two eight-pin power connections, in the form of one eight-pin and one six-pin connector.
All of this is to say: the RTX 2080 is an impressive step up from the GTX 1080 in nearly every way. And while it still remains to be seen how well new technologies like ray tracing and AI learning will play out, this card is still geared toward offering superb performance for today’s games too. And as 4K monitors—some with HDR—grow in popularity, the RTX 2080 is designed to handle it, as it’s being positioned as a GPU for 60Hz 4K gaming.
GeForce RTX 2080 – Benchmarks
Like any hardware afficianado I couldn’t wait to get my hands on this new GPU when it was announced before Gamescom, so the moment we’ve all been waiting for has finally arrived. I ran the RTX 2080 through many popular titles using built-in benchmark tools and demos when applicable, and recording my average frames per second from a typical gameplay session using Fraps when no built-in benchmarking was available. The test bench I used was built by IGN and was used in our previous GPU reviews for Pascal. It’s a Skylake system with a Core i7 7700K CPU, Asus Z270 Prime motherboard, 8GB of DDR4 memory, an Intel SSD and an EVGA power supply.
I tested this card (and all others) at 3840 x 2160, 2560 x 1440, and 1920 x 1080 resolutions with all games and benchmarks set to their highest graphical settings, but without anti-aliasing enabled. You can see the results in these charts below:
Obviously, everyone is expecting to see a marked performance improvement between the RTX 2080 with its new Turing architecture and the Pascal-based GTX 1080. What I found was a significant improvement in regards to frames-per-second, especially at 4K resolution. By and large, the RTX 2080 was 25 to 40 percent faster on both real-world game tests and on benchmark software. Those numbers carried over to 2560 x 1440 resolution, but in some cases—the Heaven 4.0 benchmark, for example—the span was even greater.
In any case, these are impressive numbers that interestingly enough are very similar in scope to the improvements we saw between the GTX 1080 and GTX 980 Ti. These aren’t the massive increases we saw between the 1080 and the 980 proper, but they’re impressive nonetheless. Coupled with the upcoming advances in ray tracing and DLSS, the RTX 2080 is a fairly future-proof card. That’s most evident when looking at Shadow of the Tomb Raider. While its 4K performance came in just a smidge under 60fps, it’s important to note that with some very, very minor tweaking this game will run well over 60Hz with the RTX 2080. None of these tests take DLSS into consideration either, which offers to boost performance even further in compatible titles. In general, if you’re looking to run games at 4K 60fps, this card will do it.
If you’re still on the 1080p, 144Hz gaming train there’s much more affordable options out there. If you’re a 1440p monitor user, this card will absolutely run like a beast for years to come.
Finally, if you’re not sure if you should pull the trigger on an RTX 2080 or a GTX 1080 Ti, I’d say given current pricing the RTX 2080 is a no brainer simply because it’s around $100 to $150 more, performs roughly the same, and has the benefit of the RTX technology. Though it’s hard to quantify what a technology is “worth,” it doesn’t seem outlandish to pay a slight premium for ray tracing and DLSS, as they both seem to offer a lot of potential.
Given the short time I had with the cards, I wasn’t able to play around with the new overclocking tools for this particular card yet. I can say that under load testing, the RTX 2080 overclocked itself all the way up to 1995MHz and hit a maximum temperature of 75C. That’s a bit warmer than I would like, but the numbers are all well within acceptable ranges—especially for a Founder’s Edition card. For context, the GTX 1080 Ti Founder’s Edition cards ran at 84C under load, so it’s safe to say this iteration does run cooler than its predecessor while hitting similar clocks.
Finally, despite the impressive benchmarks, the RTX 2080 still leaves a lot of questions unanswered. Namely, will ray tracing actually catch on as a common feature in upcoming games? And will DLSS actually lead to a new, unleashed form of graphical processing that boosts performance in a significant way, in AAA titles? That’s obviously a bit hard to discern at this point, but at least this card is situated to take advantage of future technologies while still playing today’s games beautifully at 4K resolution.
In any case, there will be third-party cards in the IGN test labs soon that will differ quite a bit from the Founder’s Edition in terms of both clock speeds, cooling, and even price. The Founder’s Edition is a great card for sure, but unless you have to upgrade right away, you might want to wait for reviews of the partner cards before upgrading.
The GeForce RTX 2080 Founder’s Edition has an MSRP of $799 and is now available (if you can find one) directly from Nvidia, or or on Amazon. IGN also put together a list of all the RTX GPUs you can order online, right here.