The launch of AMD’s new Radeon RX 6700 XT graphics card ($479.99) is a bit of a rerun, but of a so-so mid-1970s cop show: a story we’ve seen before, in an episode we were hoping we wouldn’t see again. To quote our earlier Radeon RX 6800 XT review: “This launch has been anything but smooth. Benchmarking took multiple days…just to come to the same conclusion: AMD’s driver support continues to be the company’s main weak spot. As an architecture, RDNA 2 holds a lot of promise both in PC and console, but as of today, that’s all we saw in the RX 6800 XT…a whole lot of promise, without consistent-enough results to back it up just yet.”
Replace one number in that quote (“7” for the “8”), and you’ve got your synopsis of the RX 6700 XT. Even though it’s been roughly four months since the launch of RDNA 2 (AMD’s latest architecture for Radeon gaming video cards), AMD’s robust, competitive hardware remains hampered by inconsistent driver support for certain games. This prevents the Radeon RX 6700 XT from attaining greatness in the midrange alongside Nvidia’s five-star GeForce RTX 3070 Founders Edition.
Admittedly, the RTX 3070 is a scary competitor, and when the stars align, the RX 6700 XT is a fine card for the money for play at 1440p and 4K (depending on your detail level). But that’s only if the pieces fall into place for the specific games you play. That isn’t enough for us to strongly recommend the RX 6700 XT at its launch price (or anything higher, given where aftermarket video-card prices are headed these days), though with today’s card-supply situation, anything is better than nothing…right?
Specs Compared: AMD Radeon RX 6700 XT vs. the World
The new AMD Radeon RX 6700 XT is based on the company’s RDNA 2 architecture, and it features TSMC’s 7nm lithography on a 335mm2 “Navi 22 XT” GPU die.
The card features 12GB of GDDR6 RAM, which looks pretty substantial compared with Nvidia’s GeForce RTX 3070 Founders Edition and its comparatively measly 8GB of GDDR6. In advertising this card, AMD is keen to point out particular games, such as Horizon Zero Dawn and Red Dead Redemption 2, which both can utilize more than 8GB of VRAM when set to their highest detail levels.
This is perhaps a bit of flak to distract you from the areas where the RX 6700 XT trails the RTX 3070, like in ray-tracing cores, memory bandwidth speeds, bus size, and power consumption. On MSRP pricing, the RX 6700 XT snipes the RTX 3070 by $20…though, of course that’s a bit of a moot point in today’s scalper-dominated marketplace for video cards. Prices are essentially fluid unless you luck into a rare card available for an eyeblink on a reseller’s site.
We tested the reference version of the RX 6700 XT, direct from AMD. Third-party sellers will have different designs. The dual-slot, dual-fan RX 6700 XT is 10.5 inches long, and it falls right within the compact tier of cards that are at the limit of what Mini-ITX PC builders might be able to fit in their next build. The RX 6700 XT maintains many of the same design elements we first saw when we reviewed the Radeon RX 6800 XT, while also bringing a few new accents to the table like a silver-on-red finish, and some more contouring around the fan openings.
The 230-watt power requirement is within expectations for the midrange these days, as is the twin-power-connector layout found on top of the card. (You will need a six-pin and an eight-pin connector from your PSU.) AMD recommends at least a 650-watt PSU with the card.
The backplane of the RX 6700 XT card contains what’s quickly become the standard lineup in the reference-card and Founders Edition card field: three DisplayPort 1.4b slots, one HDMI 2.1 port, and no VirtualLink to be found.
Let’s Get Testing: DirectX 11? Never Heard of It!
PC Labs ran the AMD Radeon RX 6700 XT through a series of DirectX 11- and 12-based synthetic and real-world benchmarks. Our PC Labs test rig is Intel-based and employs a PCI Express 3.0, not 4.0, motherboard. It’s equipped with an Intel Core i9-10900K processor, 16GB of G.Skill DDR4 memory, a solid-state boot drive, and an Asus ROG Maximus XII Hero (Wi-Fi) motherboard. All cards below were tested on this rig. Given our tests with the Core i9-10900K and recent Ryzen 9 CPUs, this rig is the best reasonable configuration of the moment in 2021 to cut the CPU out of the equation for frame rates. (Read more about how we test graphics cards.)
For our testing, we focused some of the effort on the esports aspect of the card with games like Counter-Strike: Global Offensive (CS:GO) and Rainbow Six: Siege. We also ran the card through the rest of our standard benchmark regimen, which tests a card’s abilities to handle AAA games at the highest possible quality settings, as well as how it rides during synthetic benchmarks that stress the card in a variety of ways.
Also remember that almost every test we run (aside from the esports titles) is done at the highest possible quality preset or settings. If you have a higher-hertz gaming monitor and you’re worried your card might not make the frame-rate grade, it could still be possible with the right card and a combination of lower settings. Not only that, but some of these titles (including Death Stranding, Shadow of the Tomb Raider, and F1 2020) have both DLSS and FidelityFX CAS with Upscaling integrated directly into the game. This can mean boosts of up to 40% more performance on top, depending on the setting and the card you’re playing with.
And so, onward to our test results. Note: If you want to narrow down our results below to a specific resolution (say, the resolution of the monitor you plan to game on), click the other two resolution dots in the chart legends below to suppress them and see a single set of results. Our list of AAA titles includes a mix of recent AAA titles like Red Dead Redemption 2 and F1 2020, as well as some older-but-still-reliable pillars of the benchmarker’s toolkit, like Shadow of the Tomb Raider and Far Cry 5.
Testing Results: Synthetic Benchmarks
Synthetic benchmarks can be good predictors of real-world gaming performance. UL’s circa-2013 Fire Strike Ultra is still a go-to as an approximation of the load levied by mainstream 4K gaming. We’re looking only at the test’s Graphics Subscore, not the Overall Score, to isolate the card performance. Meanwhile, we also ran 3DMark’s Time Spy Extreme test, which is a good test of how well a card will do specifically in DirectX 12 games at 4K resolution. We also run 3DMark’s Port Royal, strictly a test for GeForce RTX cards at the moment, measuring how well they handle ray-tracing tasks. (Thus the absent bars there for the AMD Radeon cards.) Also here are a pair of GPU-acceleration tests (Furmark, LuxMark); more details on those at the “how we test” link.
The Radeon RX 6700 XT struggled to keep pace with the Nvidia RTX 3070 Founders Edition in these benches, not finding even a single win in any of the tests we run. Not only that, but it was substantially slower than the RTX 3070 in our Furmark run, a trend mirrored by the results of the RX 6800 and RX 6800 XT in the same test.
Testing Results: Recent AAA Games
Now on to the real-world game stuff. The following benchmarks are games that you can play. We typically used in each case (for the AAA games) the highest in-game preset and, if available, DirectX 12. The multiplayer-focused and esports titles (such as CS:GO and Rainbow Six) were set slightly below top detail settings to maximize frame rates.
Straight out of the gate with our AAA games testing, we’re seeing a trend similar to that of the Radeon RX 6800 XT and RX 6800. As a DX11 game, Assassin’s Creed: Odyssey had similar stalling problems in 1080p and 1440p that we saw in that review; however, things picked up as we went down the line into more recent AAA titles based on DX12.
In quite a few instances, the RX 6700 XT took a slight lead over the RTX 3070 Founders Edition, like in Shadow of the Tomb Raider in 1080p and almost every resolution of Total War: Warhammer II, as well as some resolutions of Far Cry 5, Metro Exodus, and Death Stranding. Overall, the card keeps pace with the RTX 3070 Founders Edition, trading blows in these new or new-ish AAA games, while also keeping enough space between itself and the RX 6800 to justify the gap in cost.
This is when the RX 6700 XT is at its best: when it’s playing games that AMD has paid attention to. The hardware of RDNA 2 is clearly capable; however, as we’ll discover later, it’s hampered by AMD’s semi-scattershot approach to game optimization as a whole. For its price, the RX 6700 XT’s main problem is that the Nvidia RTX 3070 Founders Edition simply exists, and doesn’t drop a note in its symphony of rock-steady performance. The RTX 3070 is, for lack of a better phrase, just about perfect. So anything that tries to compete with it will be looked upon with even greater scrutiny, and in this examination, the RX 6700 XT stumbles once certain DirectX 11 games (even major AAA releases like Odyssey) are thrown into the mix.
A Testing Interlude With SAM
Then there’s SAM, which stands for “Smart Access Memory.” This is AMD’s version of the Resizable BAR feature that’s been available in motherboard BIOSes for years, but only recently rose to prominence as Nvidia and AMD announced their plans to include it, under two names, in their new lines of GPUs last year. (For a full breakdown of the tech, check out the link above.)
We ran a bit of SAM testing, and overall, at least in games like Red Dead Redemption 2 or Shadow of the Tomb Raider, the uplift could be anywhere between 3% and 8%. (Because AMD hadn’t yet added support for Resizable BAR on the motherboard of our GPU testbed yet, we had to use numbers that were recorded off the MSI MEG Ace X570 PCIe 4.0 motherboard we use to test various AMD CPUs.) This boost was title- and resolution-dependent, with AAA games seeing the most benefit when running at either 4K or 1440p resolution. Some games, such as Far Cry 5, saw a slight dip, though it was within the margin of error, so we can’t attribute that drop definitively to SAM.
We’ll have a more detailed breakdown of the benefits of SAM coming up soon, so stay tuned for that article when it goes live later this week or early next. Also a quick note before we close out this testing section: With the DirectX 11 issues mentioned here and tested below, we weren’t afforded ample time to test the DirectX Raytracing (DXR) capabilities of the RX 6700 XT. Expect these numbers to hit our comparison table once we publish our follow-up pieces on SAM and DX11 performance in the coming days.
Testing Results: Multiplayer Games
Though most of PC Labs’ game tests are maxed out in graphical fidelity to push the cards to their limit, multiplayer gaming is all about maintaining the best balance between graphical fidelity and frame rate. With that in mind, we’ve kept CS:GO, Rainbow Six: Siege, and Final Fantasy 14 tuned to the best combination of necessary improvements in settings (higher anti-aliasing and lower shadows, for example), while still trying to keep frame rates for 1080p games above 144fps.
Why 144fps? That’s a coveted target for highly competitive esports gamers who have high-refresh-rate 120Hz or 144Hz (or faster) gaming monitors. For more casual players with ordinary 60Hz monitors, a solid 80fps or 90fps at your target resolution, with some overhead to account for dips under 60fps, is fine.
As a midrange card in 2021, the Radeon RX 6700 XT, like the GeForce RTX 3070, is mostly overkill unless you own a 4K 144Hz monitor (in which case an RTX 3080 or RX 6800 XT might be better suited for the job). The RX 6700 XT is plenty fast enough to push any resolution as fast as you need it to go with the right tweaks, and if you only plan on playing popular multiplayer games (again, we stress the word popular, more on that in a minute), then the RX 6700 XT is a fine pick.
Testing Results: How About Some Legacy AAA Titles?
We also ran some quick tests on some oldies-but-goodies that still offer the AAA gaming experience. These legacy tests include runs of Hitman: Absolution, Tomb Raider (2013), Sleeping Dogs, and Bioshock: Infinite, the last being a game that has no business still being as well optimized as it is here in 2020.
To quote myself from the review of the Radeon RX 6800 XT last November:
“This is where things really started to take a turn. In Sleeping Dogs, the AMD Radeon RX 6800 XT performed well under the expectations set by the results we saw in the current lineup of AAA titles. On the whole, 4K results for all the legacy AAA games we tested were substantially slower than the RTX 3080 Founders Edition, and both 1080p and 1440p results followed a similar trend.”
No need to say that in different words if the problem remains the same, right? (Just replace “6800 XT” with “6700 XT,” and “RTX 3080” with “RTX 3070,” and you’ve got the gist of it.)
Upon seeing nearly identical losses in performance as I did on the RX 6800 XT, my first instinct proved to be right. So far as we can tell, at this point in legacy testing, there don’t seem to be any problems with frame rate “capping” like we saw on Odyssey; on those tests, both the RX 6800 XT and the RX 6700 XT, with their vast power differences, scored the roughly the same in 1440p and 1080p.
Here, the RX 6800 XT maintains the right price-to-performance differential with the RX 6700 XT, but they’re still both 20% off from the mark in their respective performance tiers. It’s a repeat of the RX 6800 XT, only one step down in price this time.
So, what’s going on here? We asked AMD, and here’s a bit of what we found out…
Let’s Talk About Those Numbers for a Minute
On the subject of subpar performance in some titles based on DX11, RDNA 2’s troubles with older engines has a brief history to relate.
It’s been four months since I reviewed the RX 6800 XT, and in that review I mentioned that several of our legacy games that we test on DX11 had showed notable performance shortfalls versus the rest of the DX12-based titles we run as a part of our testing suite. AMD responded that its team was “…actively investigating these results, and [will] provide an update when available.”
That was in November of last year, and this week, I saw nearly identical performance hits in our legacy run across three different testbeds. One was our standard GPU testbed, the second was our AMD X570-based CPU testing rig (which I also used for the SAM testing), and the third (based on an X299 motherboard with an Intel Core i9-10980XE) is usually reserved for storage testing. We pulled it out of the back special for this occasion just to make sure that our testing hardware wasn’t responsible for the performance anomalies. (Spoiler: It wasn’t.) After all that, we came to the same conclusion we started with on the very first go-around with our legacy games: Despite four months of lead time, AMD still hasn’t addressed optimization of certain titles in DX11.
When asked about the problem of optimizing RDNA 2 for more modern and popular titles, AMD issued us this statement:
“AMD strives to provide the best performance to gamers with all of our products. We prioritize optimizations that benefit the widest group of gamers, ensuring those on the AMD RDNA 2 architecture will enjoy strong performance in more modern titles as well as more broadly played popular titles using both modern and legacy APIs.”
This tracks with what we were told during a 1:1 conversation, where a product manager confirmed that AMD was using data from services such as the Steam Survey to influence which games got priority treatment for optimization. In that call, the company was candid about the hard decisions it has to make, how to optimize every game and every engine to their fullest capacity. Simply put, developer resource limitations come into AMD’s RDNA 2 optimization strategy for legacy titles.
Furthermore, in a reply to us noting that we saw fewer performance falloffs with legacy titles under RDNA 1, AMD reps sent us a quote stating that…
“Performance uplift will vary on a game-to-game basis. We expect to see an amazing performance lift on RDNA 2 on a wide variety of games. However, you may experience lower performance uplift compared to first-generation RDNA GPUs in a small subset of CPU-bound legacy titles.”
Now, if uplift will vary on a “game-by-game basis,” the only question that remains is: Which games fall into which buckets? Unfortunately, right now, there’s no list, nor any kind of online tool to search whether or not your game has been optimized for RDNA 2. So until AMD offers up an internal point of reference, it’s a grab bag, and the only way to find out is to buy the card and roll the dice yourself, or read reviews. Alas, most reviewers don’t get too deep into the weeds with testing older games in the limited time they have with a card before a launch.
Now Wait, Why Does DirectX 11 Still Matter?
“But we’re already on to DirectX 12, and have been for a few years now!” you might be shouting to your screen right about now.
For a more concrete example of how poor DX11 optimization might affect gamers in a real-world scenario, let’s look at a page from Xbox Game Pass…
Now, maybe I’m old-school. But on this one page alone, I can count six games that I personally have plans to play at some point down the line. (You know, in between running F1 2020 benchmarks for the tenth time as a sanity check.) All six of them use DX11 as their primary API. Of course, this is less an advertisement about the awesomeness of Xbox Game Pass (it is awesomeawesome, though), as it is a warning to potential Radeon 6700 and 6800 card buyers: If you plan to play mostly older games on your PC, you might want to consider whether you want an RDNA 2-based GPU just yet.
Though legacy games should always be a consideration for any video-card buyer, these days the need for supporting older games well is more relevant than ever. Why? The rise of these subscription gaming services, and the access they give you to older games you may have missed the first time around. Gaming subscription services like Xbox Game Pass have become a veritable treasure trove of gaming bliss over the past few years, with Game Pass in particular applying a Netflix-style model where third-party games are continuously cycled in and out, while Microsoft and Bethesda games are a permanent mainstay of your $15 a month subscription. Services like these differ from game streaming services because you download and play every game locally on your system (where the GPU matters), rather than streaming video of a PC playing your game off in a server farm somewhere, like with GeForce Now or Google Stadia.
Let’s say you were to sign up for Xbox Game Pass and Uplay+; that’s $30 a month combined (just half the price of one AAA release). With those two services alone, you would already have access to more than 700 legacy games that run on any number of APIs, ranging from golden oldies that were released before DirectX was even a thing, all the way up to modern AAA games on DX12. So with that in mind…
For the sake of getting this review live on March 17, we won’t have an authoritative, deep breakdown of the impact of running games in DX11 until later in the week. However, just from early testing in games that let you select DX11 or DX12 (the four above), we can say with some confidence that the problem extends beyond just legacy titles, to more recent AAA releases like Shadow of the Tomb Raider, as well as F1 2020. In our benchmarks, both saw performance hits of roughly 20% across all three major resolutions when going from DX12 to DX11. These results, alongside what we saw in the “legacy games” testing (with Tomb Raider, Sleeping Dogs, Bioshock, and Hitman) tracks almost to a T.
That said, I also ran the tests you see in Total War II: Warhammer and Metro: Exodus, and there I saw no meaningful difference in DX11 and DX12 performance. Our concern, then (which should be yours, as well): How to know which DX11 games will run well? As much as we might want to (and on this particular topic, I really do want to), there’s no reasonable way to bench-test every game based on DX11. They number well into the thousands.
In early testing, thus far, it’s an even 50/50 split; two games that lose 20% performance in DX11, and two games that don’t, and when you add that to what we saw in our four-game legacy testing mini-suite, it tilts in the wrong direction for AMD. None of these results could be replicated on the RTX 3070 Founders Edition (results were almost always 1:1 between DX11 and DX12 modes), so right now, the most we we can say with certainty that if you play F1 2020 in DX11 mode or any of the legacy titles tested in the section above, you can expect the $479 RX 6700 XT to perform behind a $329 GeForce RTX 3060. Factor in the inflated prices for all these cards in today’s card-buying environment, and we’re talking several hundred dollars worth of difference for possibly near-identical performance, based on the games we’ve had the time to test so far.
We’ll have a more detailed telling of our RDNA 2 testing in DX11 soon, so keep an eye out for that as we delve deeper into what could be an Achilles’ heel for AMD’s latest RDNA-2 based graphics cards.
Overclocking and Thermals: A Little Hot, But High Headroom
Onward to temperature-testing and overclocking the card. We ran a 10-minute stress test in 3DMark Port Royal on the Radeon RX 6700 XT, and the card peaked at a temperature of 78 degrees C.
This is right around expectations set thus far by the rest of the RDNA 2 lineup. However, it’s well above the temperatures we saw on the RTX 3070 Founders Edition, with its new and improved cooling design—just 60 degrees C.
When it came time to overclock the card using AMD’s Radeon WattMan utility, I achieved a small, but stable, boost of 150MHz on the boost clock and 275MHz on the memory clock. Much like our experience with the RX 6800 XT, this actually translated to some noticeable uplift in lightly threaded games like Rainbow Six: Siege, anywhere between 4% and 10% depending on the game and the resolution. (Note: Playing at 4K in OC mode wasn’t as favorable as 1440p in Red Dead Redemption 2.)
A final note: We tend to save our thermal and overclocking tests for last when reviewing a card, and it was during the thermal run that we noticed our card sample had started to whine. This was after around 10 hours of testing, and the noise would kick up during times of intense load. It wasn’t a fan bearing. (I tried manually stopping both spinners to see if it helped, but the whine continued.) It was coming from somewhere inside the card. Even as I write this review, I can hear the Radeon RX 6700 XT emitting a low straining sound as it tries to finish up some final tests. Your sample may vary.
The Verdict: RDNA 2’s Growing Pains Recur
When the RX 6800 XT was launched last year, we gave AMD a pass for what, at the time, we called “first-day driver jitters.” The company had achieved a lot in a short amount of time on both the CPU and GPU front, and a few games getting some odd results wasn’t going to spoil the rest of the launch of RDNA 2 as far as we were concerned.
Four months later, though, we’re at the second launch of an RDNA 2-based card, the RX 6700 XT, and DX11 performance remains a grab bag. Now, if you only play modern AAA DX12 titles or well-populated multiplayer games at the top of the Steam Survey Charts (or run most of the known benchmarks), you can expect the RX 6700 XT to perform as a well priced competitor to the RTX 3070 Founders Edition. Admittedly, that card received a rare perfect score from us when it launched last year, so the RX 6700 XT is up against a Terminator-grade adversary here.
Step outside the garden of games that AMD has prioritized for driver optimization, and you may see solid performance, or a solid performance hit. As noted above, we’ll dig deeper into this in the coming days.
If your gaming diet is made up of lots of older or DX11-reliant games, tread carefully. Tip-top performance on the Radeon RX 6700 XT, and the other two RDNA 2 cards released thus far, come with no promises. AMD is working as quickly as it can, yes, but if your favorite older game doesn’t get optimized, you might as well buy a card a fraction the cost of the RX 6700 XT and call it a coupon for the same level of performance in DX11 or below. The card whine of our test card is also concerning, though this can happen with any card sample (and has in the past). If you’re not dead-set on buying on launch day, we’d look at early reviews and customer ratings to see if it’s a widespread issue.
Until AMD’s engineers can optimize this card for all, or at least more, games (not just those at the top of this month’s Steam Survey), the Radeon RX 6700 XT should remain a second choice to Nvidia’s stellar lineup of GeForce RTX 30 Series cards, and the killer RTX 3070 Founders Edition in particular.
In any other year, we would wrap up here and say, “Just get an RTX 3070!” But with the video-card stock situation being what it is here in early 2021, who are we kidding? You’ve already tried clicking that checkout button multiple times, haven’t you?
