For a while, AMD has taken a back seat as NVIDIA established itself as the undisputed leader in the GPU marketplace. Now that AMD’s RDNA 2 architecture has launched, many are wondering if this is finally the move that will challenge NVIDIA and lead to an all-out GPU war where the consumers will end up being the real winners.
When AMD Navi launched back in July 2019 with the Radeon RX 5700 XT and the RX 5700 GPUs, there was genuine excitement in the air. AMD finally released something that was able to compete with NVIDIA’s products, even if it was only in the mid-range.
That enthusiasm wasn’t only about the new release; it was also about the sense that AMD was preparing to throw even greater competition at NVIDIA.
We are pleased to say that all the speculation was true. AMD really did have something up their sleeve and finally brought a competitive product to the market, one that would cause a reaction from NVIDIA.
- April 19, 2021: Reworked the entire article to be fully accurate.
- November 22, 2020: Updated the outdated info and discussed launch day issues.
- October 31, 2020: Provided the actual benchmarks and compared them with NVIDIA cards.
- October 30, 2020: Added features and additional specs, and discussed leaks and their accuracy.
- October 29, 2020: Added the exact specifications and compared them to the competition.
- October 28, 2020: Added official release dates and pricing.
- October 24, 2020: Added the leaked benchmarks.
- October 14, 2020: Removed outdated info and added new information and comparisons to the RTX 3000 series.
- September 19, 2020: Changed the image to one showing an actual AMD Radeon RX 6000 series GPU and added leaked benchmarks.
- September 18, 2020: Confirmed the launch date.
- August 10, 2020: Added the Quadro competitor rumored release date and additional memory type clarification.
- August 9, 2020: Added the speculated memory numbers as well as Navi 31 and Navi 41 rumors.
- August 5, 2020: Added additional specification rumors.
- August 3, 2020: Confirmed the 80 CUs number and clarified it.
- August 1, 2020: Added new performance leaks and the rumored release date.
- July 30, 2020: Addressed the rumors about AMD using HBM for RDNA 2.
- July 26, 2020: Completed the section about the superiority over the RTX 2080 Ti with rumors about the RTX 3080’s improvement.
This next-gen technology was referred to as ‘Navi 2X‘ and ‘Big Navi‘ by fans, with the latter name gaining so much popularity that AMD decided to adopt it.
Internally, AMD refers to it as the ‘NVIDIA Killer‘, which is a bold move in the tech industry. There are lots of examples of companies branding their new releases as ‘something killer’, but, instead of beating the competition, the product ends up being a disaster.
With the release of the RX 6000 series, however, these GPUs certainly overwhelmed some of NVIDIA’s GPUs in in terms of raw rasterization performance.
The key reason for the AMD RDNA 2 hype is that it was announced as part of the graphical architecture of the next-gen consoles: PlayStation 5 and Xbox Series X. Here are the capabilities of the graphics cards inside the next-gen consoles.
Keep in mind that this is only a glimpse of these consoles’ capabilities. We might see something even more impressive in the future, as games become increasingly well optimized for this RDNA 2 based platform.
Table of ContentsShow
One important question lingered for a while: Would AMD RDNA 2 be released before or after the arrival of the next-generation gaming consoles? With rumors surrounding the supposed original launch date, which was announced at Computex in June 2020, and with a global crisis possibly pushing back the releases of the next-gen gaming consoles, no one knew when it would actually be released.
Despite some rumors suggesting that the RDNA 2 graphics card would arrive before PS5 and Xbox Series X, these proved to be false as those cards were released about a week after next-gen consoles launched.
When Computex 2020 was canceled, it placed AMD in a unique position for their reveal. It occurred the same way that NVIDIA presented their RTX 3000 series: an informative video presentation by company CEO Lisa Su.
In late April, during a Q1 2020 AMD earnings call, Su said: “Development of our RDNA 2 GPUs continues to progress well. We are on track to launch our next-generation gaming GPUs later this year, with a 50% performance per watt increase compared to our current offerings.”
During the Bank of America 2020 Securities Global Technology Conference, AMD’s CFO David Kumar confirmed that the RDNA 2 architecture will run through their entire stack, which is very positive. The most interesting piece of news we learned from the conference was that desktop GPUs would hit the market before RDNA 2 made its debut on consoles.
We now know that the launch came after the consoles’ release, so we can only chuckle and admit that internet rumors can be very unreliable.
As announced on October 28, 2020, both the RX 6800 and the RX 6800 XT launched on November 18, 2020. This simultaneous launch allowed AMD to pull a great one-two punch on these cards’ competitors – RTX 3070 and RTX 3080. The enthusiast-class RTX 3090 competitor, RX 6900 XT, launched on December 8, 2020.
We’ll look at the performance and pricing comparisons soon. First, it’s interesting to note that this was just in time for the holiday season. Still, we couldn’t help but worry that AMD would run into supply issues, which could be a very big blow for its sales numbers. Even more importantly, it would mean we wouldn’t get to game on AMD cards during the winter.
With AMD’s CPUs reigning supreme over Intel, it’s safe to say that 2020 was the year of AMD’s great awakening. To further add to this and strengthen its market position, AMD announced that the RX 6000 series will enjoy a performance boost when paired up with a Ryzen 5000 CPU, but we’ll get to those details later.
There were some trusted leakers who speculated Big Navi’s October 7 launch, but it became clear that this was actually the announcement of their CPU Zen 3 architecture, which happened on October 8, 2020.
As mentioned previously, when AMD released their Navi lineup in 2019 they competed with NVIDIA, but they never really challenged their top-of-the-line products such as the RTX 2080 Ti. One of the main points of interest for the AMD RDNA 2 was the possibility that it would go for the throat and begin a proper GPU war.
The strategy of undercutting NVIDIA’s prices has proven fruitful as NVIDIA was forced to lower the prices of its Super cards at the last moment, so it’s possible that AMD will gamble on it again.
There was another interesting theory that AMD had something special in store in the ray-tracing department. A lot of speculation was based on that possibility. When RDNA 2 cards were finally fully announced, we were happy to see that AMD did in fact have something special in mind.
As announced during AMD’s official RX 6000 launch event, the prices are pretty competitive, especially taking performance into consideration. The RX 6800 will cost $579, while the RX 6800 XT will have a $649 price tag.
In order to paint a clearer picture of these prices, we will need to give you a sneak peak at the performance numbers, which we’ll examine later.
The RX 6800 is a full $80 more expensive than its chief competitor RTX 3070, but the performance numbers also indicate that it is a better card. The important question is: is it $80 better?
The RX 6800 XT, on the other hand, is $50 cheaper than the RTX 3080 Founders Edition and that should worry NVIDIA. It isn’t only the price that should worry them, but the fact that the RX 6800 XT appears to be a cheaper card and a better one.
Unfortunately for NVIDIA, the bad news keep on coming. The RTX 3090 competitor and AMD’s first foray into enthusiast-class cards in a while, the RX 6900 XT will release at $999.
At $1499, RTX 3090 is a whopping $500 more – and it doesn’t stop there. AMD showed that RX 6900 XT is actually neck and neck performance-wise with the RTX 3090 while being only two-thirds of the price. In terms of gaming, the RX 6900 XT is a clear winner.
However, we can’t ignore the fact that the RTX 3090 comes with 24GB of VRAM, which is 8GB more than the competition, making this one of the most affordable workstation/productivity cards on the market.
Specifications And Features
The biggest talking points regarding AMD’s RDNA 2 architecture are that it will feature ray tracing, and also provide the much-desired 4K gaming experience. While this has been a consistent characteristic of NVIDIA’s RTX series since 2018, AMD has been noticeably vague with its response. Until now.
AMD encouraged rumors of its next GPU series supporting ray tracing as far back as 2019 when the RX 5000 series was still relatively fresh. It turns out that it did have those capabilities planned, but the implementation might seem odd. Where NVIDIA has separate RT cores in charge of ray tracing performance, AMD tacked its ray accelerators onto each of the compute units.
It would be misleading to claim that either method is inherently superior but, based on the evidence we have seen, NVIDIA does have a better ray tracing solution. However, we need to remember that NVIDIA has had two extra years to fully develop those ray tracing capabilities.
In contrast, AMD’s solution is still pretty young. In time, AMD will likely continue updating drivers to improve their ray tracing power. Developers now also have access to these cards which means they can further optimize their games for AMD GPUs.
In other words, we will only see the full picture after some time, when we see more ray tracing games available on the market.
Another aspect that is being promoted is the support for variable rate shading and mesh shaders, although that was expected at this point.
RDNA 2 is the successor to RDNA, which in turn is the successor to GCN. What allowed RDNA to be such a step-up was the alignment of wavefront size to SIMD size, but also the improvement in the efficiency department.
Part of the reason for this improvement was the switch from the 14nm process to 7nm, which people thought RDNA 2 couldn’t repeat as it would stick to the 7nm process. However, AMD managed to bring a welcome surprise to the table following the RX 6000 reveal event.
This confirmed that it managed to increase its performance-per-watt metric by 50%. In fact, they managed to improve it by 54%, but the 50% number is the improvement that RDNA achieved over GCN and also the target for RDNA 2 over RDNA.
One important thing to note is that this 54% improvement belongs to the RX 6800 XT, while the RX 6900 XT will have an even more impressive number of 65% improvement in performance-per-watt over the RX 5700 XT.
Needless to say, AMD pulled off something very impressive. We need to take a moment to truly appreciate how big of a generational leap this is, especially considering it was achieved for two generations in a row.
One thing that’s interesting to note here is that a leaked roadmap indicated that AMD’s Ryzen CPUs would introduce the 5nm process with their next Zen 4 generation. This translates to RDNA getting the same treatment, perhaps even in time for RDNA 3. The RX 6000 reveal event confirmed the existence of RDNA 3, but only said that it will be made using an “advanced node”.
One area where AMD has the high ground is the 72 computing units for the RX 6800 XT, which clearly surpasses RTX 2080 Ti’s 68. It’s interesting to note that RTX 3080 has 68 SMs (streaming multiprocessors), exactly like its predecessor.
AMD has NVIDIA beat in the mid-range as well, with RX 6800 boasting 60 CUs, while RTX 3070 has 58 SMs. These are roughly equivalent in terms of what they do. While AMD does seemingly outperform NVIDIA here, it’s worth noting that its model will cost $80 more.
Things do shift back in NVIDIA’s favor when comparing the RX 6900 XT and the RTX 3090, where the former has 80 CUs and the latter 82 SMs. However, it’s important to mention that the RTX 3090 costs $500 more, and no, that isn’t a typo.
These numbers appear to be exclusive for PC models, as the PlayStation 5 and Xbox Series X come with 36 and 56 computing units, respectively. That number is still a decent increase from the 40 computing units in the 5700 XT.
Another cool aspect of the compute units for Big Navi is that AMD claims it managed to improve its processing and minimize data movement, bringing a 30% increase in energy efficiency.
It’s interesting to note that AMD uses 16GB across three cards. The only difference can be found on the RX 6700 XT which was released slightly later with 12GB of VRAM.
This also means we are getting 16GB/s of effective memory speed via a 256-bit memory interface and up to 512 GB/s memory bandwidth.
Unlike AMD, NVIDIA handled its memory specs slightly better with only the RTX 3070 going as low as 256-bit interface and the RTX 3080 and RTX 3090 having a 320-bit and 384-bit, respectively. The latter models use the GDDR6X memory, while RTX 3070 will remain on GDDR6.
Where AMD and NVIDIA differ is the size of the memory. As mentioned, AMD has 16GB across the board except for one GPU, while NVIDIA offers 24GB for the enthusiast-class RTX 3090, 10GB for the flagship RTX 3080, 8GB for the mid-range RTX 3070, 8GB for 3060 Ti and, for some reason, the RTX 3060 comes with 12GB of VRAM.
Since NVIDIA arrived sooner on the market with the RTX 3000 Series, it’s apparent that they regret their decision to go with such low VRAM capacity, hence the 12GB on the RTX 3060. This looks like a direct response to AMD’s 12GB RX 6700XT.
For a while, there were rumors of 20GB RTX 3080 and 16GB RTX 3070 variants, but new information eventually told us that they were canceled before even being officially announced.
However, with NVIDIA, anything can happen. These GPUs with more VRAM could be reintroduced as Super or Ti variants.
Besides the memory similarity, the RX 6900 XT, RX 6800 XT and RX 6800 are almost identical. The RX 6700 XT is slightly different because it was released several months later.
The only other difference (aside from the actual chip) is their physical sizes. The RX 6800 and the RX 6700 XT will require two slots, while other two will need two and a half.
Even the actual GPUs aren’t that different.
Firstly, let’s talk about the RX 6800, as it is the most different from the first three that were released. In addition to the already mentioned 60 compute units, it has a 2105MHz boost clock with a game frequency of 1815MHz.
The former is the absolute maximum that can be achieved while the latter is the expected clock when running video games.
The other two cards each have a 2250MHz boost clock and a 2015MHz game clock, so a definite step up and, more importantly, another advantage over the competition.
The RX 6700 XT has a boost clock of 2581MHz and a game clock of 2424MHz. These frequencies are much higher than the other GPUs, but don’t let that confuse you. This GPU comes with far fewer computing units and streaming processors, so a higher clock speed can be achieved.
When discussing the Xbox Series X, Microsoft has said that the RDNA 2-powered GPU’s ability to offer up to 12TFLOPS (terra-FLOPS) of performance is an improvement over the 5700 XT’s 9.75TFLOPS, but still lags behind RTX 2080 Ti’s 14.2TFLOPS.
However, both Microsoft and Sony use their own custom silicone, so we can’t perfectly compare them to PC RDNA 2.
Then again, TFLOPS performance isn’t always the most precise indicator of gaming performance, although it does play a part.
As such, the RX 6800 will be able to have a peak single-precision performance of roughly 16.17 TFLOPs. In comparison to its competitor RTX 3070’s 17.5 TFLOPs, this is not a bad number at all.
However, the RX 6800 XT’s 20 TFLOPs simply fall short of RTX 3080’s 25, as does the RX 6900 XT’s 23 to RTX 3090’s 29.
As we said, TFLOPS numbers are not exactly precise when it comes to gaming performance. The RX 6800 has less TFLOPS, but is clearly much faster than the 3070. This same logic applies to the 6800 XT and 6900 XT.
It’s also worth noting that RX 6800 users will have to account for 250W and a recommended system PSU capacity of 650W, which isn’t much compared to the competitors.
Still, the RX 6800 XT and the RX 6900 XT both use 300W. These numbers are lower than their NVIDIA counterparts, which is probably why these GPUs run at lower temperatures.
Big Navi has two DisplayPort 1.4a with DSC, which doesn’t specifically mark it as 1.4 or 1.4a. Single HDMI and USB-C ports are present as well.
During the RX 6000 reveal event, AMD’s Senior Director Laura Smith announced this innovation as a result of AMD seeking to improve its image fidelity. She touted it as a “graphics-optimized, high-density, high-speed cache based on Zen 3 L3 cache”.
Although it could be interpreted as a slight dig at RTX 3090, Smith claimed that Infinity Cache’s 256-bit GDDR6 doubles the effective bandwidth of a 384-bit GDDR6 solution. The reason that this is something of a marketing trick is that RTX 3090 uses 384-bit bus width, but on a GDDR6X memory.
AMD will enable 128MB across all three graphics cards for Infinity Cache. As the technology itself is new and untested, we can’t really be sure of how appropriate that number is. It’s interesting to note that this cache is based on Zen 3 L3 cache and that even the top Zen 3 CPUs have 64MB of it. While that sounds very good, we will need to wait and see what the actual performance benefits are.
This feature is being touted as a one click overclocking tool that will be built into the Radeon software. It promises to take advantage of every bit of overclocking headroom that is already built in to the GPU.
While this might seem like the end of manual overclocking, there will still undoubtedly be enthusiasts who want to tinker with it for themselves. It will also be interesting to do both and see which one performs better.
Smart Access Memory
This feature was at first exclusive to gamers who use a Ryzen 5000 series processor with a Radeon RX 6000 series graphics card, which allowed these two components to interact and enable even more memory. This in turn leads to an increase in performance.
AMD said that if you pair these two on its 500 series chipset motherboards, you will see as much as a 13% performance increase, which ended up being true. Some benchmarks showed even more impressive performance increases than that.
However, this feature is no longer exclusive to Ryzen 5000 CPUs and RX 6000 GPUs. NVIDIA have been working hard to bring Resizable Bar (Smart Access Memory is just AMD’s name for it) and have been able to do so.
What’s even more interesting is that Resizable Bar now works with Ryzen 5000, Ryzen 3000 and Intel’s 10th and 11th Gen CPUs.
However, we can’t ignore the fact that NVIDIA’s gains are significantly lower than AMD’s gains from SAM. Keep in mind that further driver and BIOS updates could boost these performance increases.
195-225% Faster Than RDNA? 40-50% Faster Than RTX 2080 Ti?
There are some benchmarks from AMD that used the Ryzen 9 5900X CPU with its RX 6000 flagship and they placed the performance roughly neck and neck with the RTX 3080.
Based on the reveal event, it appears that these numbers were true and that the RX 6800 XT competes with and, more often than not, outperforms NVIDIA’s RTX 3080. Likewise, the RX 6900 XT is very close to the RTX 3090 in terms of performance, although NVIDIA’s card does stay slightly ahead. AMD can hold its head high, however, as its enthusiast card is a whole $500 cheaper.
Based on AMD’s comparisons of the RX 6800 and the RTX 2080 Ti, and what we already know regarding the 2080 Ti’s performance compared to the RTX 3070, we can say with some certainty that the RX 6800 will compete with the RTX 3070.
It seems as though AMD is firing on all cylinders as there are pretty good points regarding the 7% IPC increase. Leakers have stayed on the cautious side and actually proclaimed that the IPC upgrade will be at least 7%, with some suggesting it could go up to 10%.
These leaks are now outdated as the GPUs have been released. Let’s look at the real numbers.
The Leaks Were Surprisingly Accurate
From what we’ve already seen, it appears that the leaked performance benchmarks were surprisingly close to accurate. We should underline that this was not the case for NVIDIA as its leaks were all over the place. This probably means that AMD has some security issues but, in the end, we benefited.
There was one leak that is particularly interesting to mention. A June 2020 leak showed three slides from an internal AMD presentation regarding the RX 6900 XT. We covered this leak extensively and analyzed it, but many industry insiders dissected it and determined that it was fake.
Let’s go through it again and see exactly how accurate it was.
The first slide shows that AMD had the $999 price tag planned way in advance. The rumors stating that they would wait for NVIDIA to release their Ampere cards before setting the prices for Big Navi were inaccurate.
The compute units and TFLOPs numbers were on point and there was a welcome change in the actual reveal. This leak showed us that the RX 6900 XT was going to have 14GB of GDDR6 VRAM, while in reality we got 16GB.
The boost clock and game clock are also within the margin of error. This is something that have been fine-tuned later in production.
The next slide compares the 4K FPS performance increase over the RTX 2080 Ti and places it on average at around 30%, which might raise some eyebrows as the reveal RX 6900 XT benchmarks are more than that, although not by much.
The last slide touts AMD’s implementation of ray tracing, which describes the drop as 5-10%. However, during the Big Navi reveal event, AMD noticeably avoided talking about specific ray tracing performance.
A pessimist might say that the numbers are so bad that it would hurt the brand if they were showcased, but an optimist could say that AMD wants to announce the ray tracing numbers separately for greater impact.
RDNA 3? RDNA 4?
There have been reported sightings of a card with a Navi 31 code. Understandably, this brought everyone out to speculate.
Is this going to be the next generation, RDNA 3? That would mean AMD might be planning on releasing a new GPU series as early as 2021, and possibly continuing that model for some time. This would put major pressure on NVIDIA to retain its status as the leader in the GPU world. As the leak comes from a screenshot of AMD’s macOS 11 “Big Sur”, it appears to be legitimate, but it is still shrouded in mystery.
Another thing that is difficult to believe is a Navi 41 code having been spotted in the wild. As plans for RDNA 3 are still largely unknown, it’s obvious that RDNA 4 is even further away. Some have speculated that these are simply code names for AMD’s upcoming enthusiast-class GPUs.
During the RDNA 2 reveal event, it was also revealed that AMD does have a plan for RDNA 3, but the company has announced nothing beyond that it will be made with an “advanced node“. This makes it seem as though it will differ from RDNA’s and RDNA 2’s 7nm process, but it could also mean further refining the 7nm node.
Launch Day Issues
It appears that the world has a much greater need for new graphics cards than both AMD and NVIDIA anticipated. Much like its chief competitor, AMD had issues with the availability of their cards when the launch day arrived Unfortunately, just like NVIDIA, Sony with its PlayStation 5 and Microsoft with its Xbox Series X, AMD suffered greatly from scalpers.
Finding an RX 6000 series card is difficult both physically and digitally, but this makes matters worse for AMD.
The reason why this is happening to both NVIDIA and AMD is largely because the demand for GPUs and CPUs has considerably increased in 2020 and the same is expected for 2021.
Market researchers have claimed that the PC industry saw growth of about 11% in 2020 in comparison to 2019. The industry has not seen this kind of growth for decades.
Is AMD RDNA 2 Finally A Worthy Opponent For NVIDIA?
This is the key question. However, it’s difficult to simply state that one is better than the other. We don’t want to mislead you in any way, shape or form, so we’ll discuss the announced options for each performance/budget bracket.
RX 6800 vs RTX 3070
Although we briefly touched upon the in-game performance numbers, we’ll take a closer look. Let’s start at the bottom and talk about NVIDIA’s GeForce RTX 3070 and AMD’s Radeon RX 6800.
The RX 6800 is $80 more expensive than the RTX 3070, but is on average 10% faster at 1080p. The difference grows if you bump up the resolution to 1440p. At 4K, the 6800 is about 15% faster than the 3070.
Whether that performance increase of 10% or 15% is worth $80 is up to the buyer.
There are a few other factors that need to be considered. Let’s start with ray tracing.
Currently, ray tracing performance on the entire RX 6000 range is subpar. As mentioned previously, NVIDIA have had much more time to refine their ray tracing capabilities while developers have been optimizing games for RTX cards.
If you want the best ray tracing performance right now, you should opt for the RTX 3070 instead of the RX 6800.
In terms of power consumption, they are quite similar, with nothing significant to separate them.
If you are pairing your RX 6800 with a Ryzen 3000 or Ryzen 5000 CPU, you could get much better performance over the 3070 by enabling Smart Access Memory.
Based on these numbers (and plenty of others), this might look like an easy AMD win, but remember the negatives: the lack of solid ray tracing performance and a higher price tag.
RX 6800 XT vs RTX 3080
The benchmark scores for NVIDIA’s GeForce RTX 3080 and AMD’s Radeon RX 6800 XT are pretty straightforward, so let’s look at them.
In Borderlands 3, Call Of Duty: Modern Warfare, Doom Eternal, Gears of War, Shadow of the Tomb Raider and Wolfenstein: Young Blood, the RX 6800 XT marginally outperforms the RTX 3080. In The Division 2 and Resident Evil 3, the RTX 3080 pulls ahead. The outliers are Battlefield V and Forza Horizon, where the RX 6800 XT significantly outperforms the RTX 3080.
Note: These comparisons were measured at 1440p and not 4K.
In addition to the better performance, it’s important to mention that the RX 6800 XT achieved this with 30W less power consumption. What cements this as an AMD win is that the RX 6800 XT is $50 less at launch for what appears to be a better product. Don’t forget to include SAM in the equation.
RX 6900 XT vs RTX 3090
The RX 6900 XT outperforms the RTX 3090 in Death Stranding, Watch Dogs: Legion, Borderlands 3, Assassin’s Creed Valhalla, Call of Duty: Modern Warfare, Forza Horizon 4 and Gears of War 5, while performing equally on Tom Clancy’s Rainbow Six Siege.
An interesting aspect here is AMD’s sort of admittance of defeat because they do show that the RTX 3090 performs better than the RX 6900 XT in Doom Eternal, Resident Evil 3 and Wolfenstein: Young Blood. Still, this shows that, in the end, AMD is able to compete with the RTX 3090, despite costing a considerable $500 less.
This super high-end “attack” on NVIDIA was the breath of fresh air that the GPU world needed for a while. It remains to be seen if NVIDIA will slash prices on the RTX 3090 in response, but this feels like an overall AMD win.
RX 6700 XT vs 3070/3060 Ti
Finally, the newest member of the AMD family, the RX 6700 XT enters the market as an affordable option in the mid-range GPU sector.
Originally, this GPU was thought to be a competitor for NVIDIA’s RTX 3060 Ti, but the performance numbers place it closer to the 3070 instead. However, it does cost $479, which is $80 higher than the RTX 3060 Ti.
In any case, let’s see what the benchmarks can tell us about this card.
The RX 6700 XT wins in several games against the 3070, most notably in Death Stranding, Godfall, Assassin’s Creed Valhalla, and a few others. Despite these wins, on average, the RX 6700 XT is about 4% to 5% slower than the 3070.
This is a win for NVIDIA but this is also a cheaper GPU, so giving up on 5% for $30 less is not a bad deal.
If you are interested in ray tracing, the 3070 is the clear winner here as AMD’s competitor is considerably lacking in this aspect.
In conclusion, RDNA 2 appears to be a better piece of technology than Ampere, and the RX 6000 series generally does better than the RTX 3000 series, which is exactly what was needed from AMD.