AMD RX 6800 graphics card review
Quick verdict: The RX 6800 is a significant step forward for AMD graphics cards, but it's not quite enough to take the crown from Nvidia.
- Finally delivers comparable performance to Nvidia's high-end graphics cards
- Smart Access Memory is capable of delivering considerable performance boosts
- Infinity Cache has potential to outperform equivalent Nvidia cards
- Lacklustre ray-tracing performance
- Needs an answer to Nvidia's DLSS tech
- Pricier than Nvidia's more versatile RTX 3070
For a good long while now, Nvidia has dominated the upper echelon of PC gaming with its high-end graphics cards. AMD had seemingly conceded the domain to its competitor, focusing instead on delivering value-oriented graphics cards at the budget to mid-range price tiers.
The RX 6000 series marks a shift in this philosophy. AMD has set its sights high, aiming to unseat Nvidia's popular RTX 3070 and RTX 3080 graphics cards by prioritising power and supporting future-facing technologies like real-time ray-tracing.
First in AMD's ambitious line-up is the RX 6800. Designed for 4K and high-refresh-rate gaming, it goes head-to-head with Nvidia's RTX 3070 in terms of price and promised performance. Let's see how it fares, shall we?
Tested on Mwave PowerCube gaming PC
- CPU: AMD Ryzen 7 5800X
- Motherboard: MSI MAG X570 Tomahawk Wi-Fi AM4 ATX
- RAM: 32GB Corsair Vengeance RGB Pro DDR4 3200MHz
- SSD Storage: 1TB Samsung 970 EVO M.2 NVMe SSD
- HDD Storage: 2TB Seagate Barracuda 7200RPM HDD
- CPU Cooler: Cooler Master ML240R RGB
- PSU: be quiet! Pure Power 11 700W
- Case: Tecware VXR Dual Chamber Mid Tower
- Positioned to compete with Nvidia's RTX 3070
- Triple-fan cooling solution replaces old blower approach
- Dual-slot design requires 2 x 8-pin power connectors
As one of the first graphics cards in AMD's new RX 6000 series product line (alongside the RX 6800 XT and RX 6900 XT), the Radeon RX 6800 marks AMD's return to high-end gaming. To this end, the RX 6800 boasts a raft of new features and technologies courtesy of AMD's RDNA 2 architecture. As the successor to the original RDNA architecture powering the RX 5000 series graphics cards, RDNA 2 promises up to 54% more performance per watt than its predecessor. That's a hefty increase, one that's sorely needed for AMD to bridge the gap with Nvidia's popular RTX 30 series graphics cards.
With an MSRP of $949 for the reference board (ignoring current stock and price fluctuations), the RX 6800 sits in between Nvidia's RTX 3070 with its $809 MSRP and the RTX 3080 with its $1,139 MSRP. That might seem odd at first when comparing the specs of the three cards. The RX 6800 boasts a 1.815GHz base clock speed and a boost clock speed of 2.105GHz, considerably higher than the RTX 3070's 1.5GHz/1.73GHz and the RTX 3080's 1.44GHz/1.71GHz. It also packs more memory, with 16GB of GDDR6 VRAM compared to 8GB on the 3070 and 10GB on the RTX 3080, though the latter is of the faster GDDR6X variety.
Dig a little deeper though and you'll see the comparison isn't so clear cut. While the RX 6800 outpaces the RTX 3070 with memory speeds of 16Gbps versus 14Gbps, the RTX 3080's larger 320-bit bus allows it to hit 19Gbps. And though the RX 6800's core clock speeds are higher, both the RTX 3070 and RTX 3080 deliver higher performance in floating point operations (FLOPs) – one of the most common graphical computing calculations and a useful metric for gauging raw performance. Here, the RX 6800 delivers 16.17 TFLOPs of single-precision compute performance compared to the RTX 3070's 20.3 TFLOPs and the RTX 3080's 29.8 TFLOPs.
In order to compete with Nvidia's cards, the RX 6800 requires more power than AMD's previous product lines. Typical power draw is rated at 250W, with AMD recommending your PC have at least a 650W power supply. That seems to be a fairly generous estimate, as the card never pulled down more than 205W throughout my testing.
In comparison, Nvidia cites 220W as the typical power draw for the RTX 3070 and also recommends a 650W PSU for use with the card. The RTX 3080 is considerably more power-hungry, drawing 320W and requiring a 750W PSU. There's also the matter of power delivery. While the RTX 3070 only needs a single 8-pin power connector, both the RX 6800 and RTX 3080 require two to satisfy their energy demands.
Looking at the physical board itself, the RX 6800 doesn't differ dramatically from its Nvidia counterparts. AMD's and Nvidia's reference boards are all dual-slot affairs, with the RX 6800 measuring in at 267mm long. That's longer than the RTX 3070 at 242mm but shorter than the RTX 3080 at 285mm.
Like Nvidia, AMD has ditched the "blower" fan design for the RX 6800, instead equipping it with three axial fans to keep it running cool. Blower cooling solutions tend to be louder and less efficient at dissipating heat than axial fans and this move brings the reference card in line with what you get from most third-party graphics cards.
Connectivity options are also on par with the competition. The reference board includes two DisplayPort 1.4 ports, a single HDMI 2.1 port and a USB-C port.
- AMD finally gets hardware-based ray-tracing
- Infinity Cache enables higher memory bandwidth than GDDR6 alone
- Smart Access Memory can deliver considerable performance gains when paired with a Ryzen 5000 CPU
The RDNA 2 architecture powering the RX 6800 brings plenty of new features with it, many of which are crucial for enabling gaming performance comparable to Nvidia's RTX 30 series cards.
First and foremost is the introduction of hardware-accelerated ray-tracing. Each compute unit on the RX 6800 packs a Ray Accelerator for handling ray-traced computations, enabling more realistic real-time lighting, shadows and reflections at a significantly lower cost to performance than with a software-based ray-tracing solution. It's been a long time coming, with Nvidia owning the space since the introduction of its ray-tracing RTX 20 series cards back in 2018.
But while AMD now offers hardware-based ray-tracing, it still has a way to go to compete on the same level as Nvidia. Years of experience have allowed Nvidia to hone its craft, resulting in a wider range of compatible games and higher performance across the board. And that's before factoring in the performance gains delivered through Nvidia's powerful Deep Learning Super Sampling (DLSS) technology, something that AMD has yet to offer a counter to. In short, it's very much early days for AMD ray-tracing. If you want the best ray-tracing experience, Nvidia is still the way to go.
Fortunately, hardware-accelerated ray-tracing is just one of many technologies incorporated into RDNA 2. Another is AMD Infinity Cache, and it allows the RX 6800 to outperform not only the RTX 3070 and RTX 3080 but the RTX 3090 as well, albeit in very specific circumstances.
As the name implies, AMD Infinity Cache is a caching solution that allows for faster data transfer speeds than GDDR6 and GDDR6X VRAM. It consists of 128MB of high-speed memory installed on the GPU die, occupying the middle ground between the much smaller L1 and L2 caches and the GDDR6 VRAM. Frequently-accessed data is automatically copied to the cache, drastically reducing the time taken for subsequent access operations and increasing overall effective memory bandwidth.
Performance gains from this extra cache are variable, dependant on the amount of data re-use in a particular game or application. However, in optimal situations, AMD claims it can deliver as much as 3.25 times the effective bandwidth of vanilla 256-bit GDDR6. Critically, taking advantage of the Infinity Cache requires no additional work from a game or application's developers: It's a global solution managed by the RDNA 2 architecture itself. Any game you play benefits from the tech out of the box, though as previously mentioned, the performance impact will differ from title to title.
Smart Access Memory
The other trick up AMD's sleeve is its Smart Access Memory technology. PCs equipped with both an RX 6000 series graphics card and a Ryzen 5000 series CPU can take advantage of this tech to increase the width of the data pipeline between the CPU and GPU, letting the CPU access the entirety of the GPU's VRAM at full speed. Without Smart Access Memory, PCs only permit the CPU to access a limited portion of VRAM at any one time, throttling the speed at which data can be shuffled back and forth.
By increasing potential access speeds, AMD claims Smart Access Memory can deliver up to 16% increased performance in select titles, citing examples like Forza Horizon 4, Assassin's Creed Valhalla and World of Warcraft: Shadowlands.
Testing SAM for myself, I saw significant performance gains in a couple of titles and negligible differences in others. As you can see in the Performance section below, both Forza Horizon 4 and Wolfenstein Youngblood experienced boosts of nearly 20% to their average framerates. Other games like Gears Tactics and Chernobylite saw no statistically significant improvements to their performance. I expect most games will exhibit similarly diverse results, as an individual game has to be built in such a way that it can take advantage of increased memory bandwidth, when it's available.
DirectX 12 Ultimate
Last of the big tech additions to RDNA 2 is support for Microsoft's DirectX 12 Ultimate graphics API. DirectX 12 Ultimate is another nascent technology that has yet to see widespread implementation, with only a handful of games like Dirt 5, Godfall and Cyberpunk 2077 taking advantage of the features it enables.
The promise of DirectX 12 Ultimate is quite alluring, nevertheless. Along with increased performance for DirectX Raytracing (DXR), there's now support for Variable Rate Shading (VRS). VRS allows games to define different shading rates for different regions of a scene. This can improve performance if, for instance, a game reduces the shading rate for mostly-static regions like walls and skyboxes. Because those regions change minimally from frame to frame, reducing the rate at which their shaders are updated has minimal impact on what we see, increasing efficiency without noticeably reducing fidelity.
DirectX 12 Ultimate also introduces Mesh Shaders. These allow developers to access the low-level parallel rendering capabilities of the GPU previously hidden behind linear API functions. By taking advantage of Mesh Shaders, developers can increase the efficiency of geometry rendering, leading to better performance in scenes with a lot of objects.
To highlight the potential performance increases that Mesh Shaders offer, I ran the RX 6800 through 3DMark's Mesh Shaders feature test. With Mesh Shaders off, the average framerate clocked in at 33.08fps. With Mesh Shaders on, this shot up to 189.56fps. That's a 472.9% difference. Of course, this is an optimal scenario explicitly built to leverage Mesh Shaders, so even as game developers start implementing the technology, don't expect such enormous performance boosts to be the norm.
The last key feature of DirectX 12 Ultimate is Sampler Feedback. Without diving deep into the nitty-gritty details, Sampler Feedback lets a game access more accurate information when loading textures, allowing for more efficient texture streaming, less pop-in and reduced stutter in large, open-world games.
Beyond DirectX 12 Ultimate, RDNA 2 also touts support for Microsoft's DirectStorage API. Scheduled for release sometime in 2021, it promises more efficient data transfer from NVMe SSD drives by reducing overhead using batch processing and parallel requests. Microsoft claims it will deliver big improvements in load times and smoother performance in games that require streaming in lots of small assets in real time. We can only wait and see if those claims bear fruit.
- Solid 4K and 1440p performance comparable to the RTX 3070
- Ray-tracing performance lags well behind Nvidia
- Smart Access Memory delivers significant FPS gains in some titles
The RX 6800 is best viewed as a contender to Nvidia's RTX 3070. It promises smooth 4K at 60fps performance in the latest big-budget games as well as hitting those increasingly-common high refresh rates at 1440p and 1080p. As you'll see below, it largely delivers on its promises, though not without a few compromises along the way.
AMD is positioning the RX 6800 as capable of delivering 60fps in the latest AAA games at 4K with graphics settings turned all the way up. I'd say that's only somewhat true judging by my testing. While the RX 6800 is indeed powerful enough to maintain a minimum of 60fps in games like Forza Horizon 4 and Wolfenstein Youngblood, it struggles in the 50s and 40s in more demanding titles like Control and Deus Ex: Mankind Divided. Dialling down a few settings here and there can get you to a stable 60fps, but those are concessions you have to be willing to make.
If no-compromise 4K60 gaming is what you're after, the RX 6800 just isn't there.
On the other hand, if 1440p is your resolution of choice, the RX 6800 is a lot more attractive. It had little trouble maintaining average framerates well above 60fps at Ultra/Max settings in all the games I tested, though it did occasionally dip below 60fps in Control and Chernobylite. The latter is still in Early Access so performance is rough across the board, but the drops in Control highlight just how demanding that game can be.
1440p and high-refresh rates often go hand in hand, and it's here that the RX 6800 flexes its muscles. It held average framerates over 100fps for many of the games I tested, maintaining 120fps+ in Forza Horizon 4, Wolfenstein Youngblood and Shadow of the Tomb Raider.
With how well the RX 6800 performs at 1440p, it's no surprise that it excels at 1080p. It refused to drop below 60fps at Ultra/Max settings in every game tested, delivering averages above 100fps – far above, in many cases. This makes it well-suited to folks rocking 1080p monitors with ultra-high refresh rates, though it's largely overkill if your display is locked to 1080p/60fps.
For all the anticipation surrounding AMD's adoption of hardware-based ray-tracing, the RX 6800's performance leaves a lot to be desired. Using Remedy's Control as a benchmark, 4K performance dropped from an average of 42.5fps to 15.6fps with ray-tracing set to High. Dropping ray-tracing to Medium bumped the average framerate to 20.2fps. As good as the real-time lighting and reflections look, it isn't worth the sluggish performance.
Even dropping the resolution to 1440p doesn't get you a solid 30fps with ray-tracing set to High. Only by running at 1080p does the RX 6800 deliver decent performance, almost managing to hold an average of 60fps.
These figures pale in comparison to what Nvidia's RTX series graphics cards deliver, and that's without factoring in the massive performance gains from Nvidia's DLSS tech. Until AMD has its own super-sampling solution with equivalent performance gains to DLSS, you're best served by sticking with Nvidia for real-time ray-tracing.
Smart Access Memory has a whole lot of potential, as the performance improvements for Forza Horizon 4 and Wolfenstein Youngblood in the chart above show. In both titles, the average framerate increased by almost 20%, a figure that translates to as much as 49fps in the case of Wolfenstein at 1080p. That's a staggering improvement for a feature that requires no additional development work or complex configuration settings.
However, such considerable gains aren't the norm, at least not according to my testing. Most titles experienced only minor bumps in performance, some likely attributable to the variance inherent in real-time benchmarking (this is almost certainly why some figures dropped after activating SAM). It's possible that future games and game engines will be built with more awareness of technologies like SAM and thus better leverage them when they're present. We probably won't see that for some time though. For now, SAM is an exciting technology that makes all-AMD systems more attractive, but not so much so that Intel and Nvidia should be quaking in their boots. Not yet, at least.
While synthetic benchmarks are less useful for gauging real-world performance, they can serve as a handy point of comparison when looking at how one graphics card stacks up against the competition.
I've used the ever-popular 3DMark benchmarking software here to assess the RX 6800.
|Test||Graphics score||Average framerate|
|Time Spy||14,570||Test 1: 97.41fps / Test 2: 81.83fps|
In the Time Spy benchmark, the RX 6800's graphics score of 14,570 places it between the RTX 3070 and RTX 3080, which at the time of writing had average benchmark scores of 12,559 and 15,886 respectively. That's a strong position in terms of raw performance, but the narrative shifts when looking at ray-tracing performance.
In the Port Royal benchmark which focuses on real-time ray-tracing, the RX 6800 racked up a graphics score of 7,477. Average scores for the RTX 3070 and RTX 3080, meanwhile, sit at 8,255 and 11,527 respectively. It's another testament to how much more mature Nvidia's ray-tracing tech is, and how far AMD needs to go to truly compete.
Power efficiency has long been one of AMD's strengths, and that remains true with the RX 6800. The 250W figure cited in the card's specs seems like a generous over-estimate according to my testing, as I never saw the card draw down more than 205W, even in the most demanding scenarios.
Temperatures were similarly subdued. Under load, I recorded an average of 62.81 degrees Celsius across all the games I tested. Maximum thermals, meanwhile, topped out at 71 degrees.
Should you buy the AMD RX 6800?
- Buy it if you want solid 4K and 1440p performance that gets better in an all-AMD system.
- Don't buy it if you're all about that ray-tracing - Nvidia is still leagues ahead.
The RX 6800 is an important step forward for AMD. Its raw performance allows it to compete with Nvidia's high-end graphics cards, and support for modern technologies like real-time ray-tracing save it from feeling like the compromise previous generations of AMD cards often were. With features like Infinity Cache and Smart Access Memory, there's finally a compelling reason to consider AMD at the high end of PC gaming.
AMD still has a way to go to match Nvidia tit for tat, however. Ray-tracing support and performance on the RX 6800 is miles behind Nvidia's RTX cards, and having no response to Nvidia's powerful DLSS tech severely limits the RX 6800's versatility. Pricing, too, is a sticking point. MSRP for the RTX 3070 is considerably lower than the RX 6800, and the RTX 3070 is often on par or outperforms the RX 6800 – and that's without DLSS or ray-tracing factored in.
While the RX 6800 might not be the RTX killer many were hoping for, it's a capable high-end graphics card that has enough tricks up its sleeve to give the RTX 3070 a run for its money.