Radeon RX 6800 & RX 6800 XT Review: AMD’s Back With Big Navi

For the previous few weeks, PC players have been eagerly anticipating the launch of AMD’s Radeon RX 6000 sequence, a.okay.a. “Large Navi”. NVIDIA shook up the GPU market over the previous few months with its Ampere-based GeForce RTX 30-series playing cards, which provide considerably extra efficiency and worth than their predecessors. However, restricted availability of GeForce 30-series playing cards has left a big contingent of fans within the lurch, with a pile of money burning a gap of their pockets, itching to improve. So, when AMD’s Radeon RX 6000 sequence was formally unveiled in late October and it’s was revealed that the Radeon RX 6800, RX 6800 XT, and flagship Radeon RX 6900 XT would full straight with one of the best NVIDIA needed to supply, numerous these fans realized a viable different on the high-end of the gaming GPU market on which to spend these {dollars} could be arriving quickly.

The efficiency numbers AMD showed at its event actually painted an fascinating image. AMD claimed the Radeon RX 6800 would beat the GeForce RTX 2080 Ti (and by extension the GeForce RTX 3070). The Radeon RX 6800 XT would do battle with the GeForce RTX 3080. And the mighty Radeon RX 6900 XT would go face to face with the highly effective GeForce RTX 3090. Contemplating how dominant the GeForce RTX 30 sequence playing cards have been once they hit the scene, AMD’s claims have been intriguing to say the least.

Right now, we get to learn how the Radeon RX 6800 sequence’ efficiency shakes out on our personal. We’ve acquired the Radeon RX 6800 and Radeon RX 6800 XT in-hand and have a number of information to share with you on the pages forward (the Radeon RX 6900 XT might be arriving in just a few extra weeks). Earlier than we get right down to brass tacks although, let’s take a fast tour of the playing cards and discover out what makes them tick…

AMD Radeon RX 6800 & Radeon RX 6800 XT Specs & Options

The Radeon RX 6800 and RX 6800 XT are constructed across the identical Navi 21 RDNA 2-based GPU, which is manufactured on TSMC’s 7nm course of node. The GPU is comprised of roughly 26.8 billion transistors and has a die dimension of 519mm². That’s a a lot larger chip than what powers the present Radeon RX 5700 XT (which is 10.3 billion transistors with a 251 sq. millimeter die), however there’s much more too it than simply extra transistors.

In accordance with AMD, the overarching purpose of its RDNA 2 structure was to assist drive efficiency management and propel AMD again right into a aggressive place on the high-end of the GPU market. That’s a lofty purpose contemplating AMD hasn’t been capable of really compete on this area with a single-GPU for a number of generations. Although Polaris, Vega, and first-gen Navi had their deserves, AMD hasn’t had a single GPU that might dangle with NVIDIA’s finest on the high of the stack for quite a few years.

To drag off this feat, RDNA 2’s compute items supply double the efficiency (or extra) of the CUs within the Radeon RX 5700 XT, with an approximate 50% performance-per-watt uplift as nicely. As well as, RDNA 2 affords a modern characteristic set that features help for DirectX 12 Final and all that comes with it, together with Variable Price Shading, Mesh Shaders, Sampler Suggestions, and ray tracing. Microsoft’s DirectStorage API for sooner recreation loading direct from SSD to GPU reminiscence might be supported as nicely, although it should take some time earlier than help for DirectStorage arrives in PC video games.

Each RDNA 2 CU has built-in ray tracing accelerators able to 4 Field or 1 Triangle intersection per cycle, which provide 10X higher intersection efficiency than software-only, non-accelerated implementations. Although, AMD by no means enabled DXR help in its drivers for present Radeons, so we are able to’t simply make a direct comparability there. Traversal of the Bounding Quantity Hierarchy (BVH) and shading of the ray outcomes is dealt with by shader code working on the CUs, and the brand new “Infinity Cache” (extra on that in a second), is giant sufficient to carry a big quantity of the BVH near the CUs, to maintain latency down. We must also point out that AMD is engaged on a brand new super-sampling and scaling mode to boost efficiency when ray tracing is enabled in future titles, considerably akin to NVIDIA’s DLSS, however it’s nonetheless within the early phases of improvement at this level and it’s not prepared for prime time.

AMD additionally outfitted RDNA 2 with higher-bandwidth, extra vitality environment friendly information paths, and incorporates fine-grain clock gating all through the GPU. RDNA 2 has a brand new cache hierarchy as nicely, which options a big, 128MB pool of quick cache reminiscence, dubbed “Infinity Cache”. In accordance with AMD, when leveraging Infinity Cache alongside a 256-bit GDDR6 reminiscence interface, bandwidth delivered to the GPU is greater than doubled. The Infinity Cache in the end improves absolute efficiency and bandwidth per watt, as a result of it could possibly feed information to the graphics pipeline at decrease latency and with a a lot increased cache hit price, which additional enhances total effectivity. Though the 256-bit reminiscence interface on Navi 21, coupled to 16Gbps of GDDR6 reminiscence affords up 512GB/s of native bandwidth, when the Infinity Cache is being absolutely leveraged, efficient bandwidth jumps to 1,664GB/s.

Mix the advantages of the brand new structure, Infinity Cache, together with AMD’s use of optimized libraries for the 7nm course of node used to construct Large Navi, and the Compute Items on this first wave of RDNA 2-based GPUs find yourself being roughly 30% extra vitality environment friendly than previous-gen choices. The brand new structure and tuned course of, nonetheless, have additionally allowed AMD to spice up frequencies by about 30% as nicely. The web impact is a 54% – 65% uplift in performance-per-watt, relying on the actual mannequin GPU examined and about ~2X higher efficiency (give or take) versus the previous-gen.

AMD Good Entry Reminiscence And Rage Mode

The Radeon RX 6000 sequence embrace help for quite a few new options as nicely, particularly “AMD Good Entry Reminiscence” and “Rage Mode”. AMD Good Entry Reminiscence, or ASAM, is what AMD is looking its implementation of a characteristic that’s a part of the PCIe spec referred to as Resizable BAR. SAM permits AMD Ryzen 5000 sequence processors to entry your entire pool of reminiscence connected to a Radeon RX 6000 sequence GPU over its PCIe 4 interface, when used on a suitable motherboard and chipset. Processors have at all times had the power to peek into some portion of GPU reminiscence (a legacy limitation of solely 256MB), however Good Entry Reminiscence grants Ryzen 5000 processors full entry, which might enhance efficiency with some workloads. For now help is enabled on a handful of 500-series chipset-based motherboards, however it will likely be coming to extra sooner or later.

Rage Mode is branding that AMD is giving to a brand new one-click power-boosting mode that’ll be out there in its Radeon graphics drivers, that harkens again to ATI’s Rage graphics chips from years previous. Rage mode basically will increase a card’s energy restrict, which ends up in a considerably extra aggressive frequency and voltage curve, to successfully improve efficiency. It’s not in contrast to the Excessive Efficiency profile out there with previous-gen Radeons.

AMD confirmed the mixed advantages of Good Entry Reminiscence and Rage mode boosting the efficiency of a Radeon RX 6800 XT throughout and array of video games by 2% – 13% — we’ll check that shortly.

The Radeon RX 6800 And Radeon RX 6800 XT

The total configuration of the Navi 21 GPU options 80 Compute Items – that’s what might be arriving in a few weeks when the Radeon RX 6900 XT hits the scene. The Radeon RX 6800 and Radeon RX 6800 XT have the identical GPU at their core, however scaled down to deal with completely different value factors.

The Radeon RX 6800 options 60 CUs with a complete of three,840 stream processors, with 96 ROPs, and 128MB of Infinity Cache. Its recreation Recreation and Increase clocks are 1,815MHz and a couple of,105MHz, respectively, and 16MB of normal GDDR6 is connected to the GPU through a 256-bit reminiscence interface providing up complete of 512GB/s of peak bandwidth. Whole board energy is rated for 250 watts. Single precision efficiency is simply north of 16 TLFOPS – double that to only over 32 TFLOPS for half-precision.

AMD positions the $579 Radeon RX 6800 towards the GeForce RTX 2080 Ti and by extension the GeForce RTX 3070, which only recently launched.

Shifting up the stack, we have now the Radeon RX 6800 XT. The Radeon RX 6800 XT additionally has 128MB of Infinity cache and 16GB of GDDR6 reminiscence connected over a 256-bit interface, however the CU depend is elevated to 72, for a complete of 4,608 stream processors, and it has 128 ROPs. The 6800 XT’s Recreation clock can also be a lot increased at 2,015MHz and the Increase clock is a lofty 2,250MHz. The extra CUs and better clocks push the board energy up into 300W territory, but additionally lead to a pleasant efficiency uplift of over 20.7 TFLOPS of single-precision compute efficiency (41.47 TFLOPS half-precision). Pixel fillrate is clearly elevated as nicely.

At its $649 entry value, the Radeon RX 6800 XT arrives with an MSRP about $50 decrease than the NVIDIA GeForce RTX 3080.

The entire playing cards within the Radeon RX 6000 sequence have an analogous design, with a stylized cooler that includes triple axial followers, mounted inside a die-case aluminum shroud. A die-cast aluminum body and backplate add some extra rigidity and assist help in cooling as nicely. That shroud sits atop an enormous vapor chamber that runs your entire size of the PCB. The one distinction between the coolers on the playing cards is their peak – the Radeon RX 6800 is two-slots vast, whereas the Radeon RX 6800 XT (and 6900 XT) are two-and-a-half slots. The heatsink fins and shroud are merely taller on the higher-end playing cards, so as to add some extra floor space. The entire playing cards are the identical size and peak (267mm x 120mm).

Radeon RX 6000 sequence playing cards are constructed on a 14-layer PCB, with 4 2oz copper layers, and 15 energy stage phases – two of that are devoted to the reminiscence. The playing cards characteristic twin, normal 8-pin PCI Categorical energy connectors and built-in RGB lighting controls. Though maybe not fairly as elegant because the cooling resolution on GeForce RTX 30-series Founders Version playing cards, the Radeon RX 6000 sequence is well essentially the most premium AMD has produced so far.

Ports on the Radeon RX 6000 sequence embrace a single HDMI 2.1 out with Fastened Price Hyperlink (FRL) help, twin full-sized DisplayPorts (1.4), and a USB Sort-C connector. FRL is a part of the HDMI 2.1 spec and allows increased uncompressed resolutions above 4K60, along with peak bandwidth of as much as 48Gbps.

That’s sufficient nerding-out – let’s get to some numbers…