The AMD Radeon RX 6900 XT is here! This is the company's new flagship graphics card, targeted at maxed out gaming at 4K UHD resolution with raytracing enabled. The Radeon RX 6900 XT was announced alongside the RX 6800 XT and RX 6800 back in October—today is the launch. The new RX 6900 XT is based on the same 7 nm Navi 21 silicon, but maxes it out with all its shaders enabled, the highest clock speeds among the three cards, and the highest possible overclocking headroom. AMD in its October announcement for the RX 6000 series stunned the gaming community by announcing that its latest cards offer competitive performance with NVIDIA—the RX 6800 XT is in the same performance league as the RTX 3080, and the RX 6900 XT gets close to the RTX 3090 while beating the RTX 3080.
For the most part, the RX 6800 XT and RX 6800 lived up to their hype, with the RX 6800 beating the RTX 3070 and the RX 6800 XT trading blows with the RTX 3080, but only in the majority of our game tests that lack real-time raytracing. With raytracing enabled, the RX 6800 series cards perform closer to previous-generation high-end "Turing" models, such as the RTX 2080 and RTX 2080 Ti. Still, there's enough for AMD to claim a return to the high-end graphics segment after many years. The new RX 6900 XT being launched today offers the very best from this generation and is targeted at enthusiasts or gamers who want the best AMD has to offer for 4K gaming.
The Radeon RX 6900 XT is being launched at an SEP price of $999, which lets it sit in the vast pricing gorge between the $700 RTX 3080 and $1,500 RTX 3090. If AMD is claiming that the card trades blows with the RTX 3090, it must beat the RTX 3080 to justify the $300 higher price and get close enough to the RTX 3090 to lure in buyers with the $500 lower price.
The RX 6900 XT is based on AMD's new RDNA2 graphics architecture, which debuted earlier this year with the GPU that powers the PlayStation 5 and Xbox Series X/S. This is important for AMD's RX 6000 series, as game developers build their game engines around consoles first since that's where the money is, and it minimizes effort for them to optimize their games for the RDNA2 architecture on the PC. The biggest feature addition with RDNA2 is full DirectX 12 Ultimate API support, which includes real-time raytracing using the DXR API, Mesh Shaders, Variable Rate Shading (both tier-1 and tier-2), and Sampler Feedback. AMD has also bolstered the card's feature set with an updated Radeon Anti-Lag, support for DirectStorage API, and resizable BAR (Smart Access Memory).
In this review, we take a very close look at the AMD Radeon RX 6900 XT. In its launch presentation, AMD claimed that the RX 6900 XT performs in the same league as NVIDIA's RTX 3090, but this claim came with a big asterisk—the Smart Access Memory and Radeon Rage mode overclocking features were enabled. In this review, we test the RX 6900 XT out of the box, at stock settings and on our regular VGA test bed, as well as add a data point measured on a Ryzen 9 5900X-powered machine with SAM enabled to show if the RX 6900 XT is capable of competing with the RTX 3090.
|Radeon VII||$680||3840||64||1802 MHz||N/A||1000 MHz||Vega 20||13230M||16 GB, HBM2, 4096-bit|
|RTX 2080||$600||2944||64||1515 MHz||1710 MHz||1750 MHz||TU104||13600M||8 GB, GDDR6, 256-bit|
|RTX 2080 Super||$690||3072||64||1650 MHz||1815 MHz||1940 MHz||TU104||13600M||8 GB, GDDR6, 256-bit|
|RTX 3060 Ti||$400||4864||80||1410 MHz||1665 MHz||1750 MHz||GA104||17400M||8 GB, GDDR6, 256-bit|
|RTX 2080 Ti||$1000||4352||88||1350 MHz||1545 MHz||1750 MHz||TU102||18600M||11 GB, GDDR6, 352-bit|
|RTX 3070||$500||5888||96||1500 MHz||1725 MHz||1750 MHz||GA104||17400M||8 GB, GDDR6, 256-bit|
|RX 6800||$580||3840||96||1815 MHz||2105 MHz||2000 MHz||Navi 21||26800M||16 GB, GDDR6, 256-bit|
|RX 6800 XT||$650||4608||128||2015 MHz||2250 MHz||2000 MHz||Navi 21||26800M||16 GB, GDDR6, 256-bit|
|RTX 3080||$700||8704||96||1440 MHz||1710 MHz||1188 MHz||GA102||28000M||10 GB, GDDR6X, 320-bit|
|RX 6900 XT||$1000||5120||128||2015 MHz||2250 MHz||2000 MHz||Navi 21||26800M||16 GB, GDDR6, 256-bit|
|RTX 3090||$1500||10496||112||1395 MHz||1695 MHz||1219 MHz||GA102||28000M||24 GB, GDDR6X, 384-bit|
The RDNA2 Architecture
For AMD, a lot is riding on the success of the new RDNA2 graphics architecture as it powers not just the Radeon RX 6000 series graphics cards, but also the GPU inside next-generation game consoles designed for 4K Ultra HD gaming with raytracing—a really tall engineering goal. AMD was first to market with a 7 nm GPU more than 15 months ago, with the original RDNA architecture and Navi. The company hasn't changed its process node, but implemented a host of new technologies, having acquired experience with the node. At the heart of the Radeon RX 6900 XT is the 7 nm Navi 21 silicon, which has been fondly referred to as "Big Navi" over the past year or so. This is a massive 519.8 mm² die with 26.8 billion transistors, which puts it roughly in the same league as NVIDIA's 8 nm "GA102" (28.3 billion transistors on a 628.4 mm² die). The die talks to the outside world with a 256-bit wide GDDR6 memory interface, a PCI-Express 4.0 x16 host interface, and display I/O that's good for multiple 4K or 8K displays due to DSC.
Through new design methodologies and component-level optimization throughout the silicon, along with new power-management features, AMD claims to have achieved two breakthroughs that enabled it to double the compute unit counts over the previous generation while staying within a reasonable power envelope. Firstly, the company managed to halve the power draw per CU while adding a 30% increase in engine clocks, which can both be redeemed for performance gain per CU.
The RDNA2 compute unit is where a bulk of the magic happens. Arranged in groups of two called Dual Compute Units, which share instruction and data caches, the RDNA2 compute unit still packs 64 stream processors (128 per Dual CU) and has been optimized for increased frequencies, new kinds of math precision, new hardware that enable the Sampler Feedback feature, and the all-important Ray Accelerator, a fixed-function hardware component that calculates up to one triangle or four box ray intersections per clock cycle. AMD claims the Ray Accelerator makes intersection performance up to ten times faster than if it were performed over compute shaders.
AMD also redesigned the render backends of the GPU from the ground up, towards enabling features such as Variable Rate Shading (both tier-1 and tier-2). The company has doubled ROP counts over Navi by giving the chip 128 ROPs. The RX 6800 XT and RX 6900 XT enjoy all 128 ROPs, while the RX 6800 gets 96.
Overall, the Navi 21 silicon has essentially the same component hierarchy as Navi 10. The Infinity Fabric interconnect is the link that binds all the components together. At the outermost level, you have the chip's 256-bit GDDR6 memory controllers, a PCI-Express 4.0 x16 host interface, and the multimedia and display engines (which have been substantially updated from RDNA). A notch inside is the chip's 128-megabyte Infinity Cache, which we detail below. This cache is the town square for the GPU's high-speed 4 MB L2 caches and the graphics command processor, which dispatches the workload among four shader engines. Each of these shader engines packs 10 RDNA2 Dual Compute Units (or 20 CUs) along with the updated render backends and L1 cache. Combined, the silicon has 5,120 stream processors across 80 CUs, 80 Ray Accelerators (1 per CU), 320 TMUs, and 128 ROPs.
The Radeon RX 6900 XT maxes out the Navi 21 silicon by enabling all 80 RDNA2 compute units. This works out to 5,120 stream processors, 80 Ray Accelerators, 320 TMUs, and 128 ROPs. The card comes with 16 GB of GDDR6 memory running at 16 Gbps (GDDR6-effective), across the chip's 256-bit wide memory interface, which works out to 512 GB/s of memory bandwidth. The Infinity Cache runs at the highest possible 2 TB/s data-rate; while AMD has increased the overdrive engine clock speed slider limit to 3.00 GHz (up from 2.80 GHz on the RX 6800 XT).
Infinity Cache, or How AMD is Blunting NVIDIA's G6X Advantage
Despite its lofty design goals and a generational doubling in memory size to 16 GB, the RX 6900 XT has a rather unimpressive memory setup compared to NVIDIA's RTX 3090. That is, at least on paper, with just a 256-bit bus width and JEDEC-standard 16 Gbps GDDR6, which works out to 512 GB/s raw bandwidth. NVIDIA has increased bus width to 384-bit on the RTX 3090 and innovated 19.5 Gbps GDDR6X memory to go with its cards, offering bandwidth rivaling those of 4096-bit HBM2 setups. Memory compression secret sauce can at best increase effective bandwidth by a high single-digit percent.
AMD took a frugal approach to this problem, not wanting to invest in expensive HBM+interposer based solutions, which would throw overall production costs way off balance. It looked at how AMD's "Zen" processor team leveraged large last-level caches on EPYC processors to significantly improve performance and carried the idea over to the GPU. About 20% of the Navi 21 silicon die area now holds what AMD calls the "Infinite Cache," which is really just a new L3 cache that is 128 MB in size and talks to the GPU's four shader engines at 1024 bits per pin, per cycle. This cache has an impressive bandwidth of 2 TB/s and can be used as a victim cache by the 4 MB L2 caches of the four shader engines.
The physical media of Infinity Cache is the same class of SRAM as for the L3 cache on Zen processors. It offers four times the density of 4 MB L2 caches, lower bandwidth in comparison, but four times the bandwidth over GDDR6. It also significantly reduces energy consumption, by 1/6th for the GPU to fetch a byte of data compared to doing so from GDDR6 memory. I'm sure the questions on your mind are what difference 128 MB makes and why no one has done this earlier.
To answer the first question, even with just 128 MB spread across two slabs of 64 MB, each, Infinity Cache takes up roughly 20% of the die area of the Big Navi silicon, and AMD's data has shown that much of the atomic workloads involved in raytracing and raster operations are bandwidth rather than memory-size intensive. Having a 128 MB fast victim cache running at extremely low latencies (compared to DRAM) helps. As for why AMD didn't do this earlier, it's only now that there's an alignment of circumstances where the company can afford to go with a fast 128 MB victim cache as opposed to just cramming in more CUs to get comparable levels of performance, but for less power consumption—as a storage rather than a logic device, spending 20% of the die area on Infinity Cache instead of 16 more CUs does result in power savings.
DirectX 12 Ultimate and Real-Time Raytracing
Earlier this year, Microsoft finalized the DirectX 12 Ultimate API subset, which enables improved visual fidelity over conventional DirectX 12. Four features were chosen by Microsoft for GPU vendors to qualify for the new logo—support for real-time raytracing using the DirectX Raytracing or DXR API, Mesh Shaders, Sampler Feedback, and Variable Rate Shading (VRS). AMD worked toward ticking off all four features, and the Xbox Series X/S launched earlier this year became the first DirectX 12 Ultimate device powered by AMD hardware. NVIDIA's "Turing" graphics architecture from 2018 already meets all these requirements.
AMD's implementation of DirectX Raytracing is slightly different from NVIDIA's. The RDNA2 graphics architecture uses Ray Accelerators, fixed-function hardware which calculate ray intersections with boxes and triangles (4-box intersections per clock, or one triangle intersection per clock). Intersection is the most math-intensive step, which warranted special hardware. Most other stages of the raytracing pipeline leverage the vast SIMD resources at the GPU's disposal, while NVIDIA's RT core offers full BVH traversal processing via special hardware. Also, the two companies take different approaches to de-noising, an important stage of raytracing that seeks to remove the "noise" resulting from the sparsity of rays being used. Remember, we're not quite there with fully raytraced 3D scenes, but raster 3D with select raytraced elements is an option. While NVIDIA uses an AI-based de-noiser that leverages tensor cores, AMD's de-noiser leverages compute shaders. The company claims to have innovated an efficient compute-based de-noising solution.
AMD also implemented support for mesh shaders as a geometry front end, which again heavily rely on compute shaders, sampler feedback, and variable rate shading (VRS), a key feature that allows different regions of a 3D scene to have different levels of shading, letting the GPU conserve resources. RDNA2 supports both VRS tier-1 and tier-2. VRS, along with dynamic resolution, makes up much of the secret sauce that lets next-gen consoles offer 4K UHD gaming.
Smart Access Memory and DirectStorage
Your CPU can only address up to 256 MB of video memory at once, a legacy from when things were operating in 32-bit mode with 4 GB of address space. Modern graphics cards come with a lot more memory, and in cases where the CPU needs to address more VRAM, a windowing mechanism is used where the GPU keeps a 256 MB chunk of its memory as a transfer area data the CPU requests is juggled in and out of. The 256 MB aperture size was arbitrarily decided on in the 32-bit days when address space was at a premium. Video cards have plenty of memory bandwidth (compared to main memory totals), so this arrangement didn't really bottleneck anything. AMD is already running a tight ship with its relatively narrow memory bus and Infinity Cache. As such, it sought to change this by using the resizable base address register (BAR) capability standardized by the PCI-SIG, which AMD and NVIDIA hadn't leveraged until now.
AMD simply branded resizable BAR "Smart Access Memory." This feature requires an AMD Ryzen 5000 series processor, an AMD 500-series chipset motherboard, and a UEFI firmware update that toggles the feature, but in theory any modern chipset will support it. With Smart Access Memory, the CPU is able to access the full VRAM as a continuous block of memory. AMD claims that in specific game engines that rely on heavy CPU access of video memory, Smart Access Memory can improve performance by up to six percent. We put these claims to the test later in this review, where we separately test the RX 6900 XT on a Ryzen 9 5900X-powered machine with SAM enabled.
AMD is also introducing support for the DirectStorage API, which can accelerate game-loading times by giving the GPU direct access to game resource data from an NVMe SSD in its native compressed format and performing the decompression on the GPU, leveraging compute shaders. Since the GPU isn't performing much 3D rendering during level-loading scenes, you won't feel its impact on frame rates, but loading times will be cut down.
Display and Media
AMD substantially updated the display and multimedia engines with RDNA2. The Radeon RX 6800 series cards come with two DisplayPort 1.4 connectors, one HDMI 2.1 port, and a USB-C port with DisplayPort and USB 3.1 Gen 2, along with up to 27 W USB-PD. Thanks to HDMI 2.1 and Display Stream Compression (DSC), the card now supports 8K at up to 120 Hz resolution, along with support for FreeSync and Variable Refresh Rate. The multimedia engine now has AV1 hardware decode at up to 8K resolution and HEVC hardware encode at up to 8K, along with H.264 B-frame support.
Our Patreon Silver Supporters can read articles in single-page format.