AMD Radeon RX 6000 series: Everything you need to know

Powered by the same RDNA 2 microarchitecture found in Microsoft’s Xbox Series X and Sony’s PlayStation 5 consoles, AMD finally has finally returned to the high-end PC graphics card market with its Radeon RX 6000 series GPU. The 6000 family competes well against Nvidia’s Ampere-based 30-series cards, with a lineup that consists of the 6700 XT, 6800, 6800 XT, and 6900 XT.

The Radeon RX 6800 is designed as a competitor to last-gen’s Nvidia RTX 2080 Ti flagship, while the Radeon RX 6800 XT takes on the RTX 3080 powerhouse. For enthusiast gamers, data scientists, and creative professionals who need even more performance, the Radeon RX 6900 XT goes up against Nvidia’s BFGPU, the GeForce RTX 3090.

At the heart of the Radeon RX 6000 family is the RDNA 2 microarchitecture, which is also known as Navi 2x and colloquially referred to as “Big Navi” by gamers. RDNA 2 is the second generation of AMD’s RDNA architecture, and like the move to RDNA from the older GCN architecture, RDNA 2 delivers significant performance uplifts.

In addition to a new architecture, the 6000 series brings ray tracing, blistering clock speeds, and previously unseen amounts of GDDR6 memory. Here’s everything you need to know about AMD’s latest GPUs.

Price and availability

The RX 6800 and 6800 XT launched on November 18, 2020, for $579 and $649, respectively. The 6900 XT arrived slightly later on December 8 for $999. AMD recently announced a new addition to the lineup: The 6700 XT. This $479 card is set to arrive on March 18.

Compared to Nvidia’s competing Ampere GPUs, the 6800 matches up against the RTX 3070 ($499) while the 6800 XT edges toward the 3080 ($699). Similarly, the 6900 XT is positioned as a competitor to the 3090 ($1,500) while the 6700 XT should perform slightly better than a 3060 Ti but worse than a 3070.

The issue is availability. Despite poking fun at Nvidia’s stock issues during the 6000-series announcement, AMD is experiencing issues of its own. The entire GPU market is in a frenzy in 2021. You won’t find any current-gen cards in stock for more than a few minutes, and even most last-gen options are sold out online. There are a number of reasons why — U.S. tariffs, manufacturing issues, crypto mining, and just good old scalping, to name a few — but the end result is the same: It’s hard getting a GPU right now.

And the street price doesn’t match MSRP. The 6800 and its XT variant are nearly twice their MSRP at retailers like Newegg, and the secondhand market is even worse. In March 2021, we found listings on eBay asking over $2,000 for a 6800 XT. When and if stock normalizes, you should still expect to pay more than the launch price.

The exception is the upcoming 6700 XT. It, too, will likely experience stock issues and price gouging, but the card isn’t out yet. AMD is launching it for $479, and based on the rest of the GPU market, it will likely be in the $600 to $700 range after launch. That’s the unfortunate state of affairs right now.

Radeon RX 6000 and the RDNA2 architecture

The Radeon 6000 brand was confirmed during AMD’s October 28 event, where AMD unveiled its RDNA 2 architecture. “I’m really excited to show you the new RDNA 2 architecture and our full stack of Radeon RX 6000 series products,” Su stated in her opening remarks, noting that this is the most powerful series of graphics card that her company has ever built. “Many of you have asked us to compete at the highest end of gaming GPUs as well. And today, we’re ready to show you exactly what we’ve been working on.”

With the Radeon RX 6000 family, AMD is promising an aggressive performance-per-watt improvement of at least 50% compared to the first-generation RDNA family, more power efficiency when compared against rival Nvidia’s competing GPUs, and architectural changes that bring better bandwidth, reduce bottlenecks, and improve speeds. In total, these changes help AMD deliver on its promise of 4K gaming at 60 fps.

It’s unclear what AMD’s strategy will be for the Mac, given that Apple recently announced that it will begin transitioning from Intel’s processors to its own Apple Silicon for the Mac. The RX 5000 series is available on the iMac as the Radeon Pro 5300, 5500 XT, 5700, and 5700 XT. When Apple’s transition is fully realized, it will use its own integrated GPU solution on its custom ARM-based processors.

RX 6000 coming to APUs

In addition to premium GPUs and consoles, the Big Navi platform will be headed into AMD’s APUs as an integrated GPU solution, according to AMD Chief Financial Officer David Kumar. It’s a move that’s similar to what rival Intel is doing with its Intel Xe graphics architecture.

AMD’s Ryzen 4000 APUs with integrated Radeon graphics offer enough performance to play many 1080p titles using the game’s lowest graphics settings. This would allow basic desktops — and mobile laptops — to play basic games without requiring a costly or more bulky setup with discrete GPUs. When AMD compared its Ryzen 7 4700G desktop APU to Intel’s Core i7-9700, it showed that GPU performance was up to 274% higher than its competition.

The RDNA 2-based APUs will likely provide AMD with additional boost, but Intel’s new integrated Xe architecture on its 11th-Gen processors is also has a significant jump in performance. In our testing of an early Tiger Lake laptop, we found the integrated Xe graphics more than capable of running less demanding games at 1080p with medium settings.

With Intel turning up the heat on AMD on integrated GPU performance, it will be interesting to see how a future Ryzen APU with integrated RDNA 2 graphics will compete against the new Xe architecture.

Memory

AMD is taking a different approach from Nvidia when it comes to memory. All of the 6000-series cards come with more memory than their Nvidia counterparts, with the 6800, 6800 XT, and 6900 XT featuring 16GB of memory, and the 6700 XT 12GB.

However, AMD is using the slower GDDR6 memory format, whereas Nvidia has a slight advantage with GDDR6X memory on its high-end graphics card, at least on paper. AMD was not able to use the GDDR6X standard because it is exclusive to Nvidia as part of a partnership with Micron. To combat the memory speed bottleneck, AMD is using some hardware and software tricks to extract more performance out of its 16GB of GDDR6 memory.

“We took a new approach to solving this technical challenge for gaming and deployed a graphics-optimized, high-density, high-speed cache,” Laura Smith, Radeon chief engineer, explained during AMD’s unveiling event. “We call this the Infinity Cache.” AMD’s Infinity Cache works to minimize the DRAM bottlenecks and latencies while at the same time reducing power consumption.

“Our specialized Infinity Cache, combined with a 256-bit G6 interface, delivers more than twice the effective bandwidth of a traditional 384-bit G6 solution, and a lower power consumption,” Smith added. Infinity Cache serves as a massive bandwidth amplifier, and AMD claimed that 256-bit of 16 Gbps GDDR6 memory coupled with 128 MB of Infinity Cache could lead to a 1664 GB/s bandwidth.

This compares favorably against the GDDR6X performance on Nvidia’s GPUs. For reference, the GDDR6X memory can perform up to 760GB per second on Nvidia’s RTX 3080. With Infinity Cache enabled, AMD listed memory bandwidth for all three of its graphics processors at 512 GB/s.

Previously, it was rumored that AMD’s new cards could come in HBM — or high bandwidth memory — versions, but more recent leaks reveal that those plans have been scrapped. HBM’s high cost would have likely contributed to a higher retail cost, which is something that AMD may want to avoid as it competes against Nvidia on both price and performance. With Infinity Cache, it appears that AMD may not even need to resort to HBM to compete, even if it has to make due with GDDR6 memory rather than GDDR6X.

And even though RDNA 2 makes use of the same 7nm process node as the first-generation RDNA platform, AMD said custom libraries and high-speed design flows help the new platform achieve 30% faster frequencies and up to 54% better performance-per-watt.

Design

RX 6800

Unlike Nvidia’s controversial dual-axial fan design, AMD is sticking to a more traditional GPU design with its graphics card. AMD revealed that its 6000 family will be powered by three fans — all mounted on the same side. While Nvidia’s latest GPUs come with more muted aesthetics as it chases after data scientists, A.I. engineers, and non-gamers, AMD is sticking to a more gaming-forward design with black accents, silver hardware, and red trim to highlight the Radeon badge. AMD had previously confirmed that it will be moving away from a blower-style heat sink in favor of a more premium design.

Another way that AMD differs from Nvidia is that the new Radeon cards don’t make use of a new power connector, so upgrades should be relatively easy. The 6800, 6800 XT, and 6900 XT feature dual 8-pin connectors, while the 6700 XT has an 8+6-pin connector.

Also, while Nvidia uses a massive triple-slot format for its RTX 3090 GPU, AMD claims that its competing Radeon RX 6900 XT won’t be much bigger than the 6800 XT. “And you’ll see that it’s actually relatively compact for all the performance that it delivers,” Su said. “It’s actually the same size as our 6800 XT card.” Both the RX 6900 XT and RX 6800 XT utilize a 2.5-slot format, while the Radeon RX 6800 is smaller, with a 2-slot format. All three cards measure 267mm in length.

Production process

RDNA 2 is built using the same 7nm manufacturing process as the first-generation product, and AMD claims that it has packed more than 26.8 billion transistors onto its graphics chip for 30% better energy efficiency than the prior-generation cards. In terms of performance, AMD said the move from the older Vega GPUs based on the GCN platform to first-generation RDNA brought a 50% performance uplift, and a similar 50% performance-per-watt improvement can be seen from RDNA to RDNA 2.

The top-of-the-line Radeon RX 6900 XT features 80 compute units and 80 ray accelerators, with a 2.25 GHz boost clock speed, 16GB of memory with Infinity Cache, and a TDP of 300 watts. The company claims this is the card for ultimate 4K gaming performance, and it was able to achieve over 65% performance-per-watt uplift compared to the first-gen RDNA GPUs. AMD claims this card could achieve 46.08 TFLOPS of peak FP16 performance and 43.04 TFLOPS of FP32 performance. It comes with 5,120 stream processors and 320 texture units.

“It’s extremely competitive across the board, consistently delivering high frame rates and your favorite games,” Su said. “Just take a look at the performance we see against the competition in Battlefield 5, Call of Duty, and Gears of War 5. All of this is at lower power, with a smaller card form factor.”

The mainstream Radeon RX 6800 XT flagship has 72 compute units and 72 ray accelerators, a clock speed of over 2 GHz and boost speed of 2.25 GHz, 16GB GDDR6 memory with 128MB Infinity Cache, and TDP of 300 watts. This card is designed for 4K gaming at 60 fps and is competitive against Nvidia’s RTX 3080. Gamers who scale down to 1440p gaming can see 100 fps in many games. The RX 6800 XT can achieve 41.47 TFLOPS of FP16 peak performance and 24.74 TFLOPS of FP32 peak performance. It has 4,608 stream processors and 288 texture units.

Rounding out the first batch of cards is the 6800. This card is clocked at 1.815 GHz and comes with a boost speed of 2.1 GHz. It has 60 compute units and ships with 16GB of GDDR6 memory while consuming just 250 watts of power. AMD described this card as a “fantastic entry into 4K gaming” that delivers 18% “more performance than the [Nvidia GeForce RTX] 2080 Ti.” The RX 6800 has 3,840 stream processors, 60 ray accelerators, and 240 texture units. This card can reach 32.33 TFLOPS of FP16 performance and 16.17 TFLOPS of FP32 performance.

The recently announced 6700 XT comes with 40 compute units and 40 ray accelerators paired with 12GB of GDDR6 memory. Like the other 6000-series cards, the 6700 XT comes with a blistering 2.3 GHz base clock and 2.6 GHz boost clock. This card may deliver decent frame rates at 4K, but it’s targeted at 1440p and 1080p gamers. In games like Hitman 3 and Call of Duty Black Ops: Cold War, the card can hit around 100 fps with max settings at 1440p, according to AMD. The card has 2,560 stream processors and 160 texture units, and delivers a theoretical 13.21 TFLOPS of FP32 performance and 26.43 TFLOPS of FP16 performance.

AMD also announced a single-click Rage Mode overclocking mode that can be enabled through the Radeon software, as well as AMD Smart Access Memory, which helps to reduce game load times, a feature that’s competitive with what Nvidia is doing with its Ampere GPUs.

When both AMD features are enabled, gamers can realize up to a 13% performance boost, the company claims. And to reduce latency, AMD also unveiled its Radeon Anti-Lag and Radeon Boost technologies that work together with FreeSync displays to deliver up to a 37% latency reduction in competitive games, like Fortnite, at 4K resolution. Radeon Boost is similar to Nvidia Reflex, which will analyze your inputs — like mouse clicks — to help on-screen latency and ensure you’re ready for action. The tool is useful for action-packed games and e-sports.

Ray tracing

AMD is working with partners like Microsoft to bring ray tracing to its GPUs for the first time.

“We partnered closely with Microsoft on Direct X 12 ultimate features, including Direct X ray tracing, variable rate shading, mesh shaders, and support for Microsoft direct storage APIs,” said Smith, the Radeon chief engineer .

The company also expanded its library of Fidelity FX features to work hand-in-hand with ray tracing compute and shader effects to bring better visual fidelity to games. Denoiser and variable rate shading algorithms help to adapt image quality based upon luminance and motion, AMD said, and a suite of compute-based effects are also available to developers.

And like rival Nvidia, AMD will also use Microsoft’s Direct Storage API to help reduce game load times.

Performance

With RDNA 2, AMD has a shot at taking on Nvidia in the premium segment. Though AMD has an excellent track record in the midrange market, the RDNA 2 architecture will allow AMD to compete at all price points. While AMD’s current lineup is strong for gamers in the 1080p and 2K 1440p segments, Big Navi will bring “uncompromised 4K gaming,” according to a slide the company showed to investors.

When AMD unveiled its RX 6000 family, Su revealed that AMD’s latest Radeon graphics card delivers a 50% generational performance uplift, with the top-tier Radeon RX 6900 XT delivering a 64% performance-per-watt improvement over first-generation RDNA. And with AMD’s focus on power efficiency, the Radeon cards are also able to reduce power consumption by 30% while maintaining performance that’s competitive with Nvidia’s cards.

During the company’s presentation, AMD showed that 4K and 1440p performance on the Radeon RX 6800 XT was very competitive against Nvidia’s RTX 3080. Both cards achieved similar frame rates across a number of different game titles.

Real-world benchmarks back that up. Without ray tracing, the flagship 6800 XT trails the RTX 3080 in some games at 1440p and 4K. There are some titles that favor AMD’s card, though, such as Borderlands 3 and Dirt 5. Rasterized performance for the 6800 XT is about equal to the 3080, and the difference in frame rates between the two is really splitting hairs. Synthetic benchmarks like 3DMark Firestrike actually peg the 6800 XT as the more powerful card over the 3080.

The differences come in ray tracing. Based on testing at Forbes, the 3080 delivered 97 fps at 1440p in Shadow of the Tomb Raider with ray tracing turned on. The competing 6800 XT achieved only 77 fps with the same settings. Compare that to the rasterized performance and the difference is clear. The 3080 achieved 151 fps without ray tracing while the 6800 XT followed with 146 fps. That’s a negligible difference.

The 6000-series cards hold up well against Nvidia’s best without ray tracing. Ray tracing still favors Nvidia, but that doesn’t downplay just how powerful AMD’s latest cards are. Upgrading from a 5700 XT to a 6800 XT, for example, can yield as much as a 100% uplift in performance.

Editors' Recommendations