AMD Radeon RX 6700 XT Review: Big Navi Goes on a Diet
The AMD Radeon RX 6700 XT has landed — or at least, it will land tomorrow at retail, and sell out 0.67 seconds later. (For a bot, that’s practically an eternity!) This also marks the debut of AMD’s smaller Navi 22 GPU, which we’ll call Little Big Navi or Medium Navi or something. Officially priced at $479, AMD pits the RX 6700 XT against both the RTX 3060 Ti and the RTX 3070, targeting the sweet spot for price and performance among the best graphics cards and landing in the upper ranks of our GPU benchmarks hierarchy. Let’s find out how it stacks up in our comprehensive review.
Architecturally, Navi 22 doesn’t add or remove features from Big Navi and RDNA2. It comes with full DirectX Raytracing (DXR) support and implements the full DirectX 12 Ultimate features list, including Variable Rate Shading (VRS), mesh shaders, and sampler feedback. What AMD has done echoes what we’ve seen in past generations of GPUs: It trimmed the fat, shrinking the die size by reducing the number of shader cores, memory controllers, and Infinity Cache. Here’s a quick overview of the specs for AMD’s latest GPUs, plus the previous generation Navi 10-based RX 5700 XT for comparison.
|Graphics Card||RX 6700 XT||RX 6900 XT||RX 6800 XT||RX 6800||RX 5700 XT|
|Architecture||Navi 22||Navi 21||Navi 21||Navi 21||Navi 10|
|Process Technology||TSMC N7||TSMC N7||TSMC N7||TSMC N7||TSMC N7|
|Die size (mm^2)||336||519||519||519||251|
|Infinity Cache (MB)||128||128||128||96||N/A|
|Game Clock (MHz)||2424||2250||2250||2105||1755|
|VRAM Speed (Gbps)||16||16||16||16||14|
|VRAM Bus Width||192||256||256||256||256|
|TFLOPS FP32 (Boost)||12.4||23||20.7||16.2||9|
AMD made some serious cuts with Navi 22 compared to Navi 21, with half the potential CUs and shader cores. That’s the biggest change, but there are others. The Infinity Cache checks in at 96MB now, 25% smaller than on Big Navi. Similarly, there are now six 32-bit memory interfaces instead of 8 interfaces, which reduces bandwidth by 25%. AMD compensates for the hefty reduction in core counts by delivering the highest official GPU clocks so far, with a Game Clock rated at 2424MHz — and like other RDNA2 chips, it can and often will exceed that in gaming workloads, with a maximum boost clock of 2581MHz (it’s actually a bit higher than that according to AMD’s drivers, but that’s what AMD lists on the official spec sheet).
The RX 6700 XT’s clock speed represents a bit of an interesting compromise. To push clocks into the 2.5GHz range, AMD had to increase the power limits. Considering the smaller die and reduced VRAM, normally we’d expect a pretty significant drop in power requirements compared to the RX 6800, but AMD lists a TBP (Typical Board Power, meaning it includes all power and not just the GPU) of 230W. That’s not bad, and it’s certainly within reach of most modern gaming PCs as it only requires a decent 500W PSU (AMD recommends 650W or higher). However, it’s slightly more than Nvidia’s competing RTX 3060 Ti and 3070 GPUs.
Navi 22 actually looks very similar to Nvidia’s GA106 used in the recent RTX 3060 12GB. It has the same amount of memory and a similar die size (slightly larger for AMD due to the Infinity Cache). Had AMD kept clock speeds closer to 2.25GHz, it probably could have shaved off a decent amount of power — certainly less than 200W is possible. But higher clocks come with performance benefits, and AMD apparently felt sacrificing some efficiency to boost performance was the better course.
Meet the RX 6700 XT Cards
Today, we have two RX 6700 XT cards. AMD’s reference card comes with the base level of performance and a design that echoes the other RX 6000 reference cards. Physically, the 6700 XT has the length and width of the RX 6800, but it’s slightly shorter and significantly lighter. It measures 267x110x38mm and weighs 883g, while the RX 6800 measures 267x120x38mm and weighs 1384g. The RX 6700 XT also uses a dual-fan configuration, but it has custom 88mm fans, whereas the 6800 has custom 78mm fans. Considering the similar power ratings, we’d expect the RX 6700 XT to run a bit hotter than the other RDNA2 cards, but we’ll check the details below.
The other card we have comes from Sapphire, the RX 6700 XT Nitro+. Besides a modest factory overclock, the Sapphire card uses a much larger heatsink with triple fans. It measures 313x131x48mm, so it’s a 2.5-slot width. Despite the large size, it only weighs 1020g — more than the reference card, but quite a bit less than the reference 6800 card. The two outer fans on the Nitro+ are custom 94mm diameters, while the center fan is a slightly smaller 83mm model (all measurements are mine, and could be off by about 1-2mm).
We’ll have a separate write-up of the Sapphire card soon, and we’ll include the performance results here as a point of reference for the various factory overclocked models that will likely be far more common than the reference design. We ran out of time during testing to do a full overclocking investigation, so the Sapphire card also stands in for those results. We’ll also include performance results from a Gigabyte RTX 3060 Ti card in the charts to show how custom Nvidia cards stack up — and the Gigabyte card was tested with the latest 461.72 drivers, which mostly appears to have affected performance in Horizon Zero Dawn.
Software: FidelityFX, Radeon Boost, and More
The latest AMD and Nvidia drivers are packed full of features and options. For AMD, the latest enhancements include Anti-Lag, Radeon Boost, Image Sharpening, streaming options, and more. AMD also offers other software tools, like the FidelityFX library for developers, including CAS (Contrast Aware Sharpening), ambient occlusion, and screen space reflections. AMD’s Super Resolution remains a work in progress and isn’t available yet, unfortunately. All of the FidelityFX libraries are royalty-free and open source and work with any compatible graphics card, including Nvidia’s cards.
Time constraints preclude us from testing every feature in the software options, and turning certain options on can impact our performance measurements. We ended up benchmarking using the “Standard” profile in AMD’s drivers, which turns off all extras. We’ve also encountered bugs in the past (e.g., with Enhanced Sync) that sometimes make the extras more trouble than they’re worth. Still, we appreciate having choices for tuning performance available.
As an example, Radeon Boost combines resolution scaling with CAS to improve performance. When you’re moving, and particularly turning, most games have blur effects, and it’s difficult to see all of the high quality details. At the same time, turning represents an action that can benefit from higher frame rates, as it often means you’re trying to aim at an enemy. Radeon Boost lowers your rendering resolution in such situations to increase fps, and the loss in image fidelity shouldn’t be very noticeable. Theoretically, it’s the best of both worlds… but it doesn’t work well for our apples-to-apples benchmarking.
There’s a whole series of testing we can look at, including things like Radeon Boost, DLSS, Anti-Lag, Reflex, Chill, and more. We’re looking at ways to quantify these features, but again: Time constraints. Our advice is to try some of these features and see if you like the result. If a feature potentially drops fps slightly but improves latency, that could be a net win. Or if a feature improves fps while causing a drop in image quality that you don’t actually notice, that’s another potential win. AMD and Nvidia both have a bunch of tech that’s worth further investigation, but we’ll focus on performance in like for like testing for this initial review.
We’re interested in hearing your thoughts on what features matter most as well. We know AMD and Nvidia make plenty of noise about certain technologies, but we question how many people actually use the tech. If you have strong feelings for or against a particular tech, let us know in the comments section.
TOM’S HARDWARE GPU TEST PC
Our test hardware remains unchanged from previous reviews, but we have made a few updates. Specifically, we’re now running the latest version of Windows 10 (20H2, build 19042.867), and we’ve also updated our motherboard BIOS to version 7B12v1B1, which includes beta resizable BAR support (aka, ‘ReBAR’).
AMD first brought attention to this existing feature of PCI Express with its RX 6800 series launch last November, dubbing it Smart Access Memory (SAM). At the time, AMD only supported the feature with the latest-generation Zen 3 CPUs, 500-series chipset motherboards, and AMD’s latest RDNA2 GPUs. Since then, both Nvidia and Intel have begun supporting ReBAR as well, and AMD has extended support to other CPUs and GPUs. As a result, it’s now possible for us to enable ReBAR on our primary test PC.
Speaking of which, we keep thinking it should be time to upgrade, but the gains from slightly faster CPUs aren’t quite to the point where we’ve felt it was necessary to swap testbeds and retest everything. That’s a daunting task. We looked at CPU scaling on the latest GPUs at the time of the RTX 3060 Ti launch, with a focus on the top-performing solutions (Ryzen 9 5900X, Core i9-10900K, and Core i9-9900K). While there were some differences, overall the net gain from swapping to a different CPU is only 1–2 percent, and the 9900K remains more than capable. Maybe we’ll swap when the 11900K arrives later this month; more likely, waiting for Alder Lake or Zen 4 seems like a better plan.
We’re using the same 13 games as well, but again with a change: Besides game patches, we’ve elected to drop DXR use in Dirt 5 and Watch Dogs Legion — sorry, ray tracing fans. There are several reasons for this. First, DXR support in Dirt 5 was provided via an early access code, and the support was and is still buggy. Second, the visual improvements from enabling ray tracing in these games are present, but they’re not huge while the performance hit can be quite significant. Finally, we wanted to include AMD’s previous-gen RX 5700 XT card in our test results, and it can’t support DXR.
We’ll be looking at a deeper investigation into the state of ray tracing in the coming days. It remains an interesting topic and it’s not going away, but we felt it wasn’t as critical a factor on the lower echelon cards. AMD’s RT performance remains generally worse than Nvidia’s RT performance, but some of that undoubtedly stems from Nvidia’s position as the first company to provide DXR hardware and the 2-year lead it had on AMD. DLSS is something we would also need to factor into testing, and then we’re back to comparing apples and AI-upscaled apples — not exactly unfair, but also not quite the same. So, stay tuned for a future article on the subject.
Radeon RX 6700 XT 1080p Gaming Performance
AMD pitches the Radeon RX 6700 XT primarily as a 1440p gaming solution, or alternatively, a 1080p solution targeting higher fps. We’ll start with the 1080p results and move up to 1440p and 4K below. Given the high-end price, we’re sticking with our ultra (max/highest without MSAA or SSAA) settings for this review. Our GPU benchmarks hierarchy includes results from ‘medium’ quality testing — or it will once we get it updated. We have the 6700 XT numbers, but need to update the text and charts.
Overall, the RX 6700 XT achieves buttery smooth frame rates at 1080p ultra — no surprise there. Across our test suite, which doesn’t include any DXR enabled games for this review, the card averaged 131 fps, about 7% faster than the RTX 3060 Ti and 1% slower than the RTX 3070. That disclaimer about not including any ray tracing tests matters, though. Watch Dogs Legion with DXR at ultra settings dropped performance to just 35 fps, while the RTX 3060 Ti managed 52 fps at the same settings — and with DLSS Quality mode enabled, the RTX 3060 Ti jumped to 70 fps.
As noted earlier, time constraints prevented us from doing a full suite of ray tracing benchmarks, and we’ll look at that in the coming days. It’s also important to remember that the enhanced visuals that come from ray tracing often aren’t that noticeable. Improved reflections, lighting, and shadows might be nice, but not necessarily at half the frame rate. Of course we’d say the same thing about most games’ ultra settings compared to high settings.
A few other comparisons are worth pointing out. The RX 6800 is (theoretically) $100 more than the 6700 XT, a 21% difference in price, and it’s only 16% faster. The RX 6700 XT is also 26% faster than the previous generation RX 5700 XT, and over 40% faster than Nvidia’s RTX 2060 Super, with a launch price that’s about 20% higher than those two cards. If we include the factory overclocked cards from Sapphire and Gigabyte, the story remains much the same: The Sapphire card outperformed the Gigabyte card by 6% on average.
Looking at the individual game results, the RX 6700 XT — and AMD in general — delivered significantly better performance in Assassin’s Creed Valhalla and Borderlands 3. Those are both AMD promotional games, and they both use the DirectX 12 API. The results suggest some fine tuning of AMD’s performance, without similar effort on behalf of Nvidia. Other DX12 games like Dirt 5, Horizon Zero Dawn, and Forza Horizon 4 end up closer to a 10% lead, rather than 30-45%. If we dropped the AMD-friendly games, or added more games where Nvidia performance is better (e.g. quite a few ray tracing games), we could easily distort things to make it look closer. But that’s not the point.
The main takeaway here is that AMD’s newcomer lands right where you’d expect, in terms of traditional rasterization performance, when looking at the official launch prices from AMD and Nvidia. It’s faster than the 3060 Ti, and close to the 3070, with a price slightly lower than the 3070. If you don’t care about ray tracing, or you don’t want an Nvidia GPU, the 6700 XT looks good.
Radeon RX 6700 XT 1440p Gaming Performance
Gaming at 1440p continues to be the sweet spot in our view, combining high refresh rates with improved resolutions. The RX 6700 XT performance dropped by 23% compared to 1080p ultra, but all of the games we tested, at the settings we tested, stayed above 60 fps. Watch Dogs Legion does fall below 60 fps on minimum frame rates, as does Metro Exodus — and Valhalla just barely dips below that mark. Otherwise, 1440p ultra won’t be a problem.
The console comparison here is appropriate, considering the 6700 XT likely comes in around the same level as the Xbox Series X — XSX has 52 CUs clocked at 1.83GHz, compared to 40 CUs at 2.4GHz for the 6700 XT, so they’re pretty close. How can an Xbox manage 4K gaming while a PC can’t? Easy: Dynamic resolution scaling combined with slightly lower settings. 1440p upscaled to 4K at well over 60 fps is often what you get on consoles, and you can get the same with PCs.
Overall standings don’t change too much with the higher resolution, though some of the margins become a bit larger. The 6700 XT still came in 7% ahead of the 3060 Ti, but it’s now 3% behind the 3070. The RX 6800 lead increased to 21%, while the 6700 XT beat the 5700 XT by 30% and the 2060 Super by 48%.
In the individual games, the results varied. Looking at just the 3060 Ti and 6700 XT, AMD’s lead in Valhalla and Borderlands wasn’t quite as large as at 1080p, but Dirt 5, Far Cry 5, and Horizon Zero Dawn all have slightly bigger leads for AMD. That’s why it ends up a net wash.
Radeon RX 6700 XT 4K Gaming Performance
4K ultra tends to require extreme GPUs for 60 fps, but if you’re willing to tweak the settings a bit, or use dynamic resolution scaling or Radeon Boost, it’s certainly within reach. Across the full suite of 13 games, the 6700 XT managed 57 fps on average, with five of the games breaking 60 fps — and a few like Forza and Strange Brigade did so by a wide margin, pulling the overall average up. Memory bandwidth starts to play a bigger role at 4K, however, so the RX 6700 XT’s lead over the RTX 3060 Ti now shrinks to just 2% — and again, DLSS can easily make up for the difference and then some.
That’s actually a worthwhile point to make, considering Nvidia just announced three more games with DLSS support this morning: The Fabled Woods, System Shock, and Crysis Remastered. That brings the total number of games with DLSS 2.x support up to more than 20 currently available games, with another dozen or so in development. DLSS in quality mode generally looks just as good as native rendering with TAA, with 20-30% higher performance at 1080p and 1440p, and potentially 40-50% higher performance at 4K — more if you’re willing to use balanced or performance DLSS modes. That’s enough to swing our vote generally in favor of the RTX 3060 Ti, or at least it would be if prices weren’t in the stratosphere.
Radeon RX 6700 XT Power, Temps, Clocks, and Fans
We noted at the start that the rated TBP comes in higher than we’d normally expect, likely thanks to the higher GPU clocks. We’re not going to take the time (right now) to see how much power use drops at lower clocks, but we’ll run our normal suite of Powenetics testing and check the GPU power consumption and other aspects. We run Metro Exodus at 1440p ultra and FurMark at 900p to collect these results.
It’s good to see the RX 6700 XT come in a bit lower than the rated TBP, sitting at around 215W. As usual, FurMark required a bit more power, but it’s only 2W in this case — nothing to worry about. It looks as though AMD was a bit conservative with its power ratings. That’s better than the alternative, which we’ve seen quite a bit of lately, of cards using 5W-10W more than the rated TBP. Of course, third-party cards are free to increase the power limits, and clearly, the Sapphire card did.
Power scales with voltage and clock speed, and the 6700 XT has the highest reference clocks of any GPU to date by about 175MHz. Interestingly, the Sapphire and reference cards are basically tied on average clocks in Metro Exodus, while the Nitro+ makes use of its additional power in FurMark. Maximum clocks during gaming tend to average roughly 2.5GHz, depending on the game, while the RX 6700 XT still cruises along at 2.35GHz in FurMark.
Fan speeds directly affect temperatures, and here we see the reduced cooling capacity of AMD’s reference design. It’s not loud, but it does hit higher temperatures — not that 72C is particularly hot. The larger fans help make up for the reduced number of fans, but the triple fan cards all achieve lower temperatures. Meanwhile, Sapphire’s RX 6700 XT posts some of the lowest average fan speeds we’ve ever seen.
Lower fan speeds naturally mean lower noise levels. The noise floor of our test environment and equipment measures 34 dB(A) at a distance of 15cm from the side of the GPU. We put the SPL (Sound Pressure Level) meter close to the GPU fans to focus on their noise, rather than case fans or other noise sources. The reference RX 6700 XT measured 40.4 dB, while Sapphire’s card was only a touch above the noise floor with 36.0 dB.
Radeon RX 6700 XT Mining Performance
Unlike Nvidia’s RTX 3060 12GB, AMD isn’t even trying to stop miners from using its cards. On the one hand, that might seem like a poor decision, but we also saw how that all played out with Nvidia accidentally posting a development driver that doesn’t fully implement the Ethereum mining speed limiter. Anyway, for better or worse, cryptocurrency mining is a thing right now, so we checked the hashing performance of the RX 6700 XT using NiceHashMiner.
Before running the built-in benchmark in precision mode, we tuned for optimal Ethereum mining performance. Basically, that means finding the highest stable memory clocks and then dropping the GPU clocks (or power limit, depending on the card) until we find a good balance.
In the case of the RX 6700 XT, we settled on 50% maximum GPU clocks (around 1300MHz) with a 150MHz GDDR6 overclock (17.2Gbps effective) and fast memory timings enabled. We also set the power limit to the maximum — that appears to help AMD’s RDNA2 cards make better use of the memory bandwidth even if GPU clocks don’t improve. Actual power consumption was only about 120W with these settings.
Considering Ethash tends to favor memory bandwidth over other factors, and the RX 6700 XT has a 192-bit bus instead of the 256-bit bus on the Navi 21 6800/6900 cards, the drop in hash rate pretty much matches the drop in bandwidth.
With tuning, we’re able to get around 65MH/s out of the Navi 21 GPUs, while the RX 6700 XT maxed out at around 47.5MH/s. 25% less bandwidth, 25% lower hash rate. Maybe additional tuning would improve hash rates a bit more, but it’s unlikely the 6700 XT will get much above 48MH/s with current mining software.
It’s worth pointing out that 48MH/s is actually lower than the previous-gen RX 5700/5700 XT, again, thanks to the narrower memory bus width. It’s slightly faster than the RX 5600 XT due to the higher memory clocks and also matches the “oops, we accidentally unlocked it” RTX 3060 12GB in Ethash rates.
What does that mean for gamers? Not much, except that the other cards that can do around 48MH/s currently sell for $800 or more, which means you can pretty much count on and a very limited supply of cards priced remotely close to the official MSRP, and the RX 6700 XT selling out.
Radeon RX 6700 XT: More and Less Navi
It feels like so much has happened in just the few short months since the launch of Big Navi. In December, all we had to worry about was limited supply and scalpers hoovering up what few graphics cards there were, only to see cards show up on eBay at significant price premiums. Now we have to deal with all of that, plus coin mining fueled shortages and even higher prices.
If all of the Nvidia Ampere and AMD RDNA2 GPUs were available at prices close to MSRP, the RX 6700 XT would look a bit overpriced. It’s basically a match for the RTX 3060 Ti, without the option for DLSS and weaker ray tracing performance, at an $80 price premium. Calling this an RTX 3070 competitor is a bit too ambitious, unless you limit testing to AMD-promoted games like Assassin’s Creed Valhalla, Borderlands 3, Dirt 5, etc. But now? Now there are a lot of gamers who would happily pay $479 for an RX 6700 XT — and they’re not likely to find one in stock.
AMD’s Navi 22 chip takes over from the previous generation Navi 10 chip, packing in new features and a large Infinity Cache that results in 33% larger die. It’s not just ray tracing and cache, though, as VRS, mesh shaders, and a reworked pipeline that can hit significantly higher clocks are all part of the package. If nothing else, the clocks on AMD’s RDNA2 parts are certainly impressive, and the 6700 XT typically runs at clocks around 40% higher than RX 5700 XT. Combined with the other architectural tweaks and enhancements, that yields gaming performance that’s around 30% faster, plus 50% more VRAM to ensure you won’t hit memory limits in the most voracious of games.
Is that enough? At ‘official’ prices — which are basically just fantasyland right now — we’d call this a good card, but we’d point to the RX 6800 and RX 6800 XT as better overall picks. More memory, more cores, better performance, and not that much more expensive. We’d also give the RTX 3060 Ti and RTX 3070 an edge over the 6700 XT, due to features, performance, and pricing (in that order). Except, right now, the Navi 21 and Ampere cards are all selling at more than double their MSRPs on eBay, and finding cards at retail continues to be a problem.
There’s also the next-gen console factor to consider. We’re big proponents of PC gaming, but you can get a very good gaming experience from an Xbox Series X or PlayStation 5. In fact, the RX 6700 XT ends up as a very close match to the Xbox Series X GPU: Fewer cores and higher clocks make it a net wash. Of course, all the next-gen consoles keep selling out as well, but at least miners aren’t snapping them up, which means you can find the consoles for around $600. Given a choice between $600 for an Xbox or potentially $800 (what the RTX 3060 currently sells for on eBay) for an RX 6700 XT, sadly, we would have to say the consoles are the better deal right now from a pure gaming perspective. Unless you want to mine when you’re not gaming to offset the cost of the card, maybe?
There’s no question the Radeon RX 6700 XT represents a potent card at a reasonable price. Rather, the question is how many cards will actually be available for anywhere close to AMD’s official launch price. We suspect that, just like with the other RDNA2 cards, the answer will be “not nearly enough.” Even if mining weren’t making a bad situation worse, the latest GPUs would likely still be hard to come by, and shortages on various tech components — including GPUs, memory, substrates, and even automobile and smartphone chips — are currently projected to last until 2022.
The good news is that, as a smaller chip, AMD planned for higher sales back when it ordered Navi 22 wafers. If AMD follows the usual pattern, there will probably be twice as many RX 6700 XT cards as all the Navi 21 cards combined. Maybe the less impressive mining performance will make them less attractive to miners as well. Unfortunately, we doubt that will be the case. That means if you can find one for sale at MSRP tomorrow, you should probably just buy it — even if you don’t want it, you can almost certainly find someone who will buy it off of you at a decent markup. Or you could just be a good friend and trade it to someone at cost, since getting one will be more like winning the lotto than just navigating through the usual checkout procedures.
MORE: Best Graphics Cards
MORE: GPU Benchmarks and Hierarchy
MORE: All Graphics Content
#AMD #Radeon #Review #Big #Navi #Diet