The 3080 Founders Edition card we reviewed isn’t much bigger than the 2080 Ti. It has a sleek, almost office-professional aesthetic that’s a nice contrast to the gaudy third-party cards we’ll inevitably see. The cooling is also dramatically different: NVIDIA made the PCB more dense to fit a fan that blows air directly through the card, while another ejects warm air from the rear. There’s also a new compact 12-pin power connector that opens up more room on the PCB (and yes there’s a dongle in the box for two 8-pin cables). 

Under the hood, the RTX 3080 is powered by 8,704 CUDA Cores, 68 RT ray tracing cores and 272 tensor AI cores. It has more than twice as many CUDA cores as the RTX 2080 Ti, and while its RT count is the same, the new architecture is around twice as fast. 

Surprisingly, its tensor core count is dramatically lower than the 2080 Ti, just 272 compared to 544. But again, NVIDIA claims this new generation of tensor AI cores is significantly faster than before. And if all that isn’t enough, NVIDIA also stuffed in 10GB of GDDR6X RAM, a new type of memory that’s making its debut in the RTX 3000 GPUs. 

All games tested in 4K/HDR with the highest graphics quality settings and ray tracing (where available), on a rig powered by an Intel Core i7-8700K and 32GB of RAM.

All of that hardware is in the service of one thing: Making the RTX 3080 the fastest GPU we’ve ever tested (that is, until we get our hands on the 3090). It reached 82FPS in the Shadow of the Tomb Raider benchmark, while running in 4K with maxed out graphics settings and “ultra” RTX shadows. In comparison, the 2080 Ti hit only 52 FPS. It clocked in around 2,500 more points in the 3DMark Port Royale ray tracing test, as well as 1,400 more points in the TimeSpy Extreme 4K benchmark. The big takeaway: this GPU isn’t just a modest improvement over NVIDIA’s last card, it’s a huge leap forward.

The 3080 also made Control playable in 4K with all of the graphics and ray tracing settings dialed up. It sat comfortably between 53 and 60FPS. I had to turn on DLSS to reach those speeds, though. That’s NVIDIA’s technology for upscaling low-res textures with its AI cores. The 2080 Ti, topped out at around 40FPS. Running Control natively in 4K on the 3080, without DLSS’s help, I hit a middling 32FPS. Clearly, ray tracing can still crush this hardware, which is why NVIDIA is investing so much in other ways of delivering high-res gameplay.

If you’re lucky enough to have a high-refresh rate 4K monitor or TV, the 3080 will also help you get the most out of that. I was able to play Wolfenstein: Youngblood in 4K with DLSS and ray tracing settings turned on at around 119 FPS. The 2080 Ti, meanwhile, could only hit around 70 FPS. I’ve argued for a while that framerates matter more than rendering resolution — the 3080 will finally let you have both.

What makes them special?

As we explained when they were announced, the RTX Super cards are basically souped-up versions of NVIDIA’s existing lineup. They’re taking the place of the faster “Ti” cards we’re used to seeing from the company (though, confusingly enough, the 2080 Ti is sticking around as the king of the RTX hill). They’re not meant to be upgrades for people who jumped on last year’s cards — they’re more like a reward for those who waited.

The $399 RTX 2060 Super is more in line with the previous RTX 2070 spec-wise, instead of the plain 2060. It now has 8GB of RAM, 256 more CUDA cores (totaling 2,176) and a healthy base clock speed bump up to 1,470MHz instead of 1,365MHz. NVIDIA says the 2060 Super is around 15% faster than the 2060, and it almost keeps up with the original 2070. That’s a compelling amount of performance for that price, especially if you don’t mind overclocking to eke out a bit more speed. You’ll find three DisplayPort connections, an HDMI port and USB-C “VirtualLink” (a single cable to plug in future VR headsets) on both cards.

	RTX 2060 Super	RTX 2070 Super
CUDA Cores	2,176	2,560
Base Clock	1,470MHz	1,605MHz
Boost Clock	1,650MHz	1,770MHz
Tensor FLOPS	57.4 TFLOPS	72 TFLOPS
Memory	8GB	8GB
TDP (Power)	175W	215W

Spend $100 more, though, and you’ll get the even faster RTX 2070 Super. It replaces the original model, and you can think of it as a slightly slower version of the RTX 2080. The 2070 Super gets 256 more CUDA cores, 16 additional texture units, and faster clock speeds (1,605MHz base and 1,770MHz boost). All of that hardware makes it faster than the GTX 1080 Ti, NVIDIA’s last-gen cream of the crop.

Both cards also get slightly better ray-tracing performance, though we’re still unsure how NVIDIA measures that. The 2060 Super offers 6 Giga Rays — NVIDIA’s term for counting ray-tracing speeds — instead of five. The 2070 Super, meanwhile gets 7 Giga Rays. In real-world usage, those numbers mean they’ll both be able to handle things like realistic reflections, shadows and lighting — the key features of NVIDIA’s RTX real-time ray-tracing tech — without hurting your framerate as much.

When the RTX cards launched, NVIDIA was criticized for not getting enough developers to jump aboard the ray-tracing bandwagon. But now that we’ve got coming titles like Cyberpunk, Call of Duty: Modern Warfare, and Watch Dogs: Legion, you’ll actually have a reason to invest in ray tracing. The few games that support the feature now, like Battlefield V, rely on it only for simple things, like making sure the sun reflects realistically off of your guns.

But looking ahead, some developers are banking heavily on everything ray tracing offers — Remedy has been showing off the tech’s potential for years with Control. Both Sony and Microsoft have confirmed ray tracing will be in their next-generation consoles (which will be powered by AMD hardware), so it makes sense that game makers are finally starting to take it seriously.

How do they perform?

	3DMark Timespy/Timespy Extreme	Hitman 2	Destiny 2	Witcher 3	Port Royal (Ray Tracing)
RTX 2070 Super	9,706/4,474	95 FPS	120-140 FPS	90-105 FPS	5,845/27FPS
Radeon 5700 XT	8,506/3,871	92 FPS	85-105 FPS	85-90 FPS	N/A
RTX 2060 Super	8,574/4,015	83 FPS	93-110 FPS	86-90 FPS	4,974/23FPS
Radeon 5700	5,510/2,545	79 FPS	75-90 FPS	75-80 FPS	N/A
RTX 2080 Ti	12,379/5,761	108 FPS	120-145	130-150 FPS	8,601/40FPS

As we saw in the review of AMD’s new Radeon cards, NVIDIA’s hardware is unsurprisingly fast. The $399 RTX 2060 Super is about on par with the Radeon RX 5700 XT. In some cases, like 3DMark, it’s noticeably better, but NVIDIA’s GPU still lags behind in games like Hitman 2. That’s the sort of gap that can easily be cleared by driver updates and game optimization, though, so I wouldn’t worry about it too much.

What the benchmarks boil down to is this: Both the RTX 2060 Super and RX 5700 XT are solid $399 GPUs, delivering the sort of 1,440p performance you’d expect from high-end cards just a few years ago. You can’t really go wrong with either option — but of course, you’ll only get ray tracing with the RTX 2060. The importance of that feature is entirely up to you, and how much you want to future-proof your next video card. At the very least, both offer decent ray-tracing performance. In Shadow of the Tomb Raider, the 2060 Super reached 51 FPS with ultra-ray-tracing settings in 1,440p, while the 2070 Super reached 57 FPS. In my eyes, the slight FPS hit by turning on the feature is worth it for more realistic shadows and lighting.

At $499, the RTX 2070 Super is a better bet for enthusiasts. It delivers even faster framerates at 1,440p, which will give you smoother gameplay on high refresh rate monitors (especially if they support G-SYNC too). In some cases, like The Witcher 3, you’ll even get 4K speeds close to 60FPS. In all honesty, though, the whole point of getting a card like this is to run 1,440p as fast as you can, with all of the graphical bells and whistles maxed out.

NVIDIA’s dual-fan coolers on both GPUs managed heat much better than AMD’s single-fan blower design. During benchmarks and gameplay, they maxed out at 82 Celsius. The Radeon cards, meanwhile, consistently warmed up to 90C under load. While AMD will likely get some help from third-party designs, those numbers aren’t great if you’re concerned with long-term stability and overclocking.

Both NVIDIA and AMD also have $349 GPUs to consider — the plain RTX 2060 and the Radeon RX 5700 — which are great performers in their own right. I haven’t benchmarked the 2060, but when I reviewed the RX 5700, I was impressed by how effortlessly it jumped beyond 60 FPS in 1,440p. While these cards will save you some money, though, they may limit you if you’re trying to take advantage of 120Hz and 144Hz monitors.

So, what do I buy?

As usual, your budget is the main consideration here. Most gamers would be better off with the $399 RTX 2060 Super and Radeon RX 5700 XT. They’ll give you a bit more power to handle demanding next-gen games like Cyberpunk. If you’ve been dreaming of getting something powerful like the RTX 2080, then the $499 RTX 2070 Super will get you close. And if you’re intrigued by ray tracing, your only choice is to pick one of the NVIDIA cards.

Wrap-up

NVIDIA’s “Super” GPUs are pretty self-explanatory: they’re just faster versions of the cards that came before. But the name also has a hint of desperation, now that AMD is nipping at its heels with a brand new Radeon architecture. All of that corporate drama means one thing: Consumers finally have a healthy selection of midrange GPUs to choose from.

What matters most, of course, is what’s under the hood: AMD’s RDNA architecture, previously known to enthusiasts as “Navi.” The Radeon 5700 and 5700 XT are built on a 7 nanometer process, which makes them more efficient than the 14nm Vega. This isn’t a first for AMD — the beefy Radeon VII, which remains its high-end option, is also a 7nm card, but one built on its older “Graphics Core Next” (GCN) platform.

The real benefits for the Radeon 5700 and 5700 XT come from AMD’s revamped computing unit design, as well as a higher bandwidth and lower latency memory. Most important is the more efficient graphics pipeline, which the company says will allow for better performance per clock, in addition to higher speeds. AMD claims that an RDNA card will be about 50 percent faster than a GCN card with the same clock speed and power consumption.

	Radeon RX 5700	Radeon RX 5700 XT
Compute Units	36	40
TFLOPs	7.95	9.75
Memory	8GB GDDR6	8GB GDDR6
Base Clock	1.47 GHz	1.6 GHz
Game Clock	1.625 GHz	1.76 GHz
Boost Clock	1.73 GHz	1.9 GHz

Both new GPUs also support PCIe 4.0, which will offer twice the bandwidth of the PCIe 3.0 standard found on modern PCs. That should help to reduce load times in games with faster NVMe drives, If you’re excited about PCIe 4.0 though, just be aware that you’ll need to snag a new motherboard to take advantage of it. And at the moment, it’s only supported in AMD’s X570 chipset with third-generation Ryzen CPUs. Not surprisingly, Intel is trying to downplay the importance of PCIe 4.0 today, since the current standard can still handle 4K/144Hz displays without any issue. At the very least, AMD is showing that it’s new platform is future proof. Sony, for example is already touting incredibly fast load times for the next PlayStation, which will also run AMD’s RDNA hardware.

The Radeon 5700 XT features 40 compute units, 2,560 stream processors and a boost clock speed of 1.9GHz. The 5700 is based on the exact same hardware, but it has four fewer compute units, 2,304 stream processors and tops out at 1.73GHz. Both cards include 8GB of GDDR6 RAM, instead of the HBM2 memory from AMD’s last few GPUs. Technically, HBM2 offers faster bandwidth, but AMD says the move to GDDR6 allows for more flexibility when designing GPUs. As for ports, both offer three DisplayPort connections and an HDMI port. The Radeon 5700 relies on a single 8-pin power connection, while the XT uses an 8+6 pin configuration.

So what are AMD’s new cards going up against? Originally, the company positioned the 5700 XT as something that could clobber NVIDIA’s RTX 2070, while the 5700 against the RTX 2060. In most cases, the new Radeons were noticeably faster. But now they’re competing with the speedier RTX Super cards, which makes things a bit more complicated.

In use

	3DMark Timespy/Timespy Extreme	Hitman 2	Destiny 2	Witcher 3
Radeon 5700 XT	8,506/3,871	92 FPS	85-105 FPS	85-90 FPS
Radeon 5700	5,510/2,545	79 FPS	75-90 FPS	75-80 FPS
RTX 2060 Super	8,574/4,015	83 FPS	93-110 FPS	86-90 FPS
RTX 2070 Super	9,706/4,474	95 FPS	120-140 FPS	90-105 FPS

On a broad level, both new Radeon GPUs deliver excellent 1,440p performance. The XT, naturally, fares better when pumping up graphics settings — I saw between 85 and 105 frames per second in Destiny 2, compared to 75 to 90 FPS on the 5700. Most importantly, they’re both able to run that game beyond 60FPS without a sweat. But if you’re a discerning gamer with a 144Hz monitor, which can handle higher frame rates, then you’ll probably want to invest in the pricier XT.

I saw a similar bump in performance with Hitman 2’s benchmark. The Radeon 5700 XT reached an average of 92 FPS with all of the graphics settings dialed up, while the 5700 clocked in at 79 FPS. In 1080p, both cards performed about the same, but that’s not a huge surprise. At that resolution, your performance relies more on your CPU than the GPU.