With AMD's launch of its new Radeon RX 6000 Series GPUs, the company's Radeon RX 6800 XT ($649) finds itself in the unenviable position of diving into shark-infested waters.
With its sights set squarely on Nvidia's GeForce RTX 3080 Founders Edition, AMD has set the goal for itself of remaining competitive with all takers; Intel on the CPU side, and Nvidia on GPUs.
While AMD's CPU division has pummeled Intel over the past year and a half, its graphics card arm still finds itself on defense versus Nvidia.
And that narrative doesn't change much with the Radeon RX 6800 XT (dubbed during its development "Big Navi").
At times, it's the fastest GPU we've tested to date in its price range.
At others, it returns inconsistent results across some modern AAA titles, legacy AAA games, and synthetic benchmarks.
At $50 less than the GeForce RTX 3080 Founders Edition, the Radeon RX 6800 XT does present an enticing value proposition—and, unlike, the latest GeForce RTX "Ampere" cards, may prove to be made of something other than unobtainium.
(Maybe you'll be able to buy one in the hours after its launch, or maybe not?) But the early drivers and firmware behind it, in its launch state, make us withhold our seal of approval.
With substantial overclocking headroom and plenty of new features to show off, the Radeon RX 6800 XT holds a lot of promise.
But until its drivers can regularly deliver on that potential, early adopters might want to hold off.
(We'll keep an eye out for firmware updates and re-test and re-evaluate our score as necessary.)
AMD Radeon RX 6800 XT Design: Inspired by the Classics
The first thing I noticed about the AMD Radeon RX 6800 XT as I pulled it out of the box was just how big it was.
Now, at 10.5 inches long, it's actually the same length as its nemesis, Nvidia's GeForce RTX 3080 Founders Edition.
But it's thick, too.
The 2.5-slot width and the sheer weight remind you that this is a top-end, big-iron GPU.
But in contrast to the RTX 3080 Founders Edition, where the total length of the card is made up of only about 50% PCB (with the rest dedicated to the heatsink and cooling system), the PCB of the Radeon RX 6800 XT runs the full length of the card.
And it features many of the traditional design elements we've seen on reference versions of AMD cards in the past.
You'll see some refinements here and there, but none is nearly so drastic as the changes Nvidia made between its GeForce RTX 20 Series and RTX 30 Series of GPUs.
On the back of the card we find a die-cast aluminum backplate, which is the same material found throughout the construction of the rest of the card, including the shroud and the frame.
The reference model of the AMD Radeon RX 6800 XT uses three fans in a push-pull configuration, which draws in heat in from the bottom of your case, passes it along the aluminum heatsink, and expels waste heat through the backplate as well as through vents on the side and rear port of the card.
Just as AMD kept things familiar with the design of its PCB and cooling system, the traditional dual eight-pin power connector makes an appearance on the RX 6800 XT, delivering 300 watts of power.
The AMD Radeon RX 6800 XT slims down its port offerings from previous AMD cards, offering up two DisplayPort 1.4b ports, one HDMI output, and one USB Type-C output all told.
Like the 30 Series, that HDMI output has been upgraded to the HDMI 2.1 spec, up from 2.0 in previous RDNA cards.
The AMD Radeon RX 6800 XT: To Infinity Cache, and Beyond!
So, onward to the specs.
Let's see how the RX 6800 XT shapes up against the rest of the new Radeon RX 6000 "Big Navi" line (the RX 6800 non-XT, and the forthcoming RX 6900 XT), as well as some preceding AMD cards and the two most relevant Nvidia ones (click the tab for the Nvidia cards)...
Understanding the differences among these cards is best broken down into a series of separate considerations.
Let's get into them.
RDNA vs.
RDNA 2
RDNA ("Radeon DNA") is AMD's preceding card architecture, debuting in 2019's Radeon RX 5700 line, headed by the Radeon RX 5700 XT.
The story of RDNA moving to RDNA 2 is one of refinement, innovation, and refinement again.
AMD's engineers have taken almost every element of RDNA and honed it down, chipping off the odd ends and focusing in on what works best.
Every GPU announced so far in the RDNA 2-based RX 6000 Series line utilizes some version of the same Navi 21 die (the 6800 XT is fittingly powered by the "Navi XT" variant), complete with 26.8 billion transistors spread across a 519mm2 die.
This is a step up from the Navi 10-based Radeon RX 5700 XT and Radeon RX 5700, which had "just" 10.3 billion transistors on a 251mm2 die.
The company is launching three cards altogether as a part of the Radeon RX 6000 Series debut: the top-of-the-line Radeon RX 6900 XT (due out December 8, for $999), the Radeon RX 6800 XT (launching today at $649), and the Radeon RX 6800 (today too, at $579).
AMD says to expect 1.3 times more throughput per compute unit at the same power in RDNA 2 compared with the original, as well as up to a 54% improvement in performance-per-watt over RDNA.
RDNA 2 vs.
Ampere
Moving over to Big Green, compared one-to-one with the "Ampere"-based Nvidia RTX 3080 Founders Edition, the Radeon RX 6800 XT wins some categories on paper, while losing others.
On power consumption, the RX 6800 XT is 20 watts under the 320-watt GeForce RTX 3070, while also sporting a faster maximum boost clock (2,250MHz) and costing $50 less at MSRP.
The card's 16GB amount of GDDR6 RAM is impressive on its own, but it looks a bit less so once the RTX 3080 Founders Edition enters the picture with 10GB of GDDR6X that can push 248GBps more traffic down its larger 320-bit pipeline, despite having less VRAM to work with.
This is down to the improvements made between GDDR6 and GDDR6X, with the latter having the potential to deliver up to twice as much bandwidth in the same amount of memory thanks to uniquely designed signaling pathways.
AMD notes, however, that it has a trick up its sleeve to compensate for this difference: something called "Infinity Cache."
Infinity Cache
One potentially major way AMD is changing up the GPU game with the launch of the Radeon RX 6000 Series is with its new Infinity Cache technology.
Featuring a whopping 128MB of memory cache for the card to work with, in theory this allows for the GPU to act more like a CPU, buffering against write and read operations to the main memory, and a safeguard for any larger operations that would otherwise traverse out to VRAM.
The company says Infinity Cache won't just assist in traditional workloads, but should also boost the individual power of ray-tracing compute units (CUs), thanks to the reduction in travel time.
We'll have to find out in our benchmarks whether or not this actually translates to increased performance on the level of what Tensor cores and DLSS 2.0 do for Nvidia cards, or if it's more of an incremental improvement that sees general boosts across the board.
Ray-Tracing Silicon: Now on Radeon
Speaking of ray-tracing, AMD's got that now too! The AMD Radeon RX 6800 XT comes with 72 ray-tracing CUs onboard, which AMD claims should be enough to keep titles like Far Cry 6 and Dirt 5 above the coveted 4K/60fps threshold with all ray-traced effects turned on.
Image: AMDWhen combined with VRS (below), the feature will only be supported by five titles at launch: Far Cry 6, Rift Breaker, WoW: Shadowlands, Dirt 5, and Godfall, though AMD has stated plans for at least 60 titles that should support both features sometime in the future.
Variable Rate Shading (VRS)
Introduced as a part of Microsoft's DirectX 12 VRS API, AMD has also added support for Variable Rate Shading on the RX 6000 Series.
Image: AMDVariable rate shading is a new tool for developers that allows for individualized areas of an image to be rendered at different shading rates to increase performance without losing visual fidelity.
The technology made its debut in "Turing"-based Nvidia GPUs, and as a part of Intel's Gen11 graphics architecture in its 10nm "Ice Lake" CPUs.
For example, in the case of Dirt 5 above, you can see that the car, the stands, and background elements of the bridge (the green patches) receive more shading power than the road or people in the crowd.
This technique of selectively applying power where it's needed most allows developers to increase the quality of certain elements, while reducing the load on the card at the same time.
VRS offers a clever solution to the problem of power distribution that's still in its infancy, but as a part of the open-source OpenGPU infrastructure, it could play a vital role in increasing frame rates for cards like the RX 6800 XT in the years to come.
Smart Access Memory (SAM)
Last but not least, there's Smart Access Memory.
We saved this for the end because, unlike the others above, there's been some recent controversy as to just how unique the feature is to AMD, and whether or not the company is actually using the same version of a technology that's already widely available and attempting to rebrand it as something else.
In its marketing materials, the company suggests that its "Unique to AMD" Smart Access Memory will only be compatible with AMD-branded GPUs in the RX 6000 Series, and even then will only work in tandem specifically with Ryzen 5000 Series AMD CPUs.
Turns out, though, that the underlying tech in action here is known as the "resizable bar" feature, something that's been standard as a part of the PCI Express specification for some time.
It can be manually enabled through most motherboard BIOSes.
Recently Steve Burke over at Gamers Nexus reached out to Nvidia to inquire about the company's plans to introduce a competing feature, to which the company responded:
“The capability for resizable BAR is part of the PCI Express spec.
NVIDIA hardware supports this functionality and will enable it on Ampere GPUs through future software updates.
We have it working internally and are seeing similar performance results."
The suggestion here is that the resizable bar feature is not unique to AMD, nor even to PCI Express 4.0.
Nvidia says that once implemented, the technology should work on any PCI Express-based platform, AMD or Intel, and be compatible with all Nvidia cards that receive the corresponding firmware update.
We ran some tests with SAM turned on and off across multiple benchmarks, but never noticed a gain of more than 1% in any of the titles tested.
This represents a margin so slim it could be chalked up to run-to-run benchmark variance just as easily as the effect of SAM itself.
AMD Radeon RX 6800 XT Software Features: Playing Catch-Up
FidelityFX Anti-Lag
AMD's Anti-Lag, like Nvidia's Reflex, is a feature that's designed for high-precision esports gamers who want to reduce the latency between their mouse and their screen as much as technically possible.
We weren't able to test the true efficacy of Anti-Lag from our current lab setup, as it requires a high-speed camera capable of recording up to 1,000 frames per second (along with a customized mouse).
Anecdotally, however, I've tried the feature out in the past, and it still feels just as crisp while playing with it turned on in the RX 6800 XT.
It's a bit of an odd duck, in that it's only most effective when the resolution is turned to 4K, and it's not often you find people concerned about latency also playing at high resolutions like that.
(Esports extremists tend to stick to resolutions like 1080p.) Even still, both Nvidia's Reflex and AMD's Anti-Lag are nice to have as options, and we'll continue to test both features are they're rolled out to more games and engines in the future.
Radeon Boost
Radeon Boost makes another appearance in Radeon Software, and while there haven't been any major improvements made since the last time we saw it, it should still be a consideration for competitive gamers who play any of the titles that it supports.
Image: AMDHowever, the feature is a bit less important for cards like the Radeon RX 6800 XT, which already has the raw horsepower to keep many of the games supported by Boost well above the 4K/144Hz threshold (at the right settings) where tech like this would be most impactful.
This is a feature that provides greater benefit to lower-end cards like the RX 5700 XT, and we'll have to wait and see whether or not AMD has plans to release midrange or low-end cards under the Radeon RX 6000 Series badge anytime soon.
Microsoft DirectStorage API
Like Nvidia's RTX IO, as well as both the Sony PS5 and Xbox Series X, Radeon RX 6000 Series cards will help with the task of getting data off your SSD and into the pipeline faster and more efficiently, thanks to compatibility with Microsoft's DirectStorage API.
Though we've only seen the tech in very limited game-design implementations thus far, in titles like Ratchet and Clank: Rift Apart, universally it will allow for games to load in a near-instant, depending on the number of elements it has to pull from the drive.
(Puzzle games will load faster than Assassin's Creed: Valhalla, for example.)
The DirectStorage API will be compatible with any game that chooses to support it on PC.
Given how much of a hit instant-load has proven to be with players since the release of those two consoles, we imagine it won't be long before PC developers are lining up to add it to their titles, as well.
Super Resolution
Well you can't blame AMD for cutting straight to the point.
Like Nvidia in 2018 at the launch of its GeForce RTX 20 Series cards, this year AMD has come to the stage with two features; ray-tracing and its "Super Resolution" upsampling technology.
While the former will be supported by five titles at launch, very little information has been released about AMD's plans for its DLSS competitor as of today.
So, what do we know so far? Well, from the rumor mill and confirmation of DirectML features in the Xbox Series X, we've been able to glean that in principle Super Resolution will act in much the same way that Nvidia's DLSS does, by training games on a neural net that approximates polygon positions based on various input types to increase performance with no visual loss in graphics fidelity.
Super Resolution is supposedly being developed in conjunction with Microsoft, and could utilize the company's Azure network of data centers to handle the training work for games that want to add Super Resolution support once the feature is...
