Disclosure: All information is from AMD’s livestream. AMD India does not maintain any proper communication with us at the behest of our readers, their buyers. But virtually non-existent communication does not influence this writeup.
AMD Radeon RX 6000 series is something to look up to considering the current GPU market moving slowly already moved towards DirectX Raytracing (what Nvidia calls it as RTX). This is the first series that support Ray-Tracing on the Radeon platform. As you’d expect, it needs to compete with the current RTX 30 series- The RTX 3080, the RTX 3090 and the recently launched RTX 3070.
Naturally, these will be compared with GeForce offerings. But it is also important to know about its actual performance, power consumption and pricing. On paper, it sounds impressive like many things from Intel and Nvidia. It is still best to take things with some grain of salt but AMD revealed some information that’s worth to talk about.
The heart of the RX 6K Series- RDNA2 Architecture
Make no mistake, The RX 6000 series uses the same TSMC 7nm process that RX 5000 family enjoyed. Just like the AMD RYZEN 5000 series CPUs, these have certain optimizations and features. It is hard to emphasize on architecture advancements in layman’s terms. But AMD simply said their primary goal was to have a 50% performance-per-watt improvements compared to RDNA 1. So just like the AMD Ryzen 5000 series, they were able to do this with architecture optimizations in the same process. Some CPU and GPU manufacturers were able to pull off significant optimization and performance or performance per watt boost. Some were not. We’ll only get to see its effectiveness when RX 6000 series GPUs are rolled out.
Not a lot of information about the RX 6000 series was revealed. Based on its presentations, energy-efficient Compute Units was one of the catalysts. AMD talked vaguely about fine-grained clock gating, aggressive pipeline rebalancing and redesigned data paths. While gaming consoles are not something we talk about, this is more significant and appreciated in it. Sony PlayStation 5’s GPU has 36 CUs and clocked up to 2.23 GHz. Same RDNA2 formula is naturally inherited in discrete graphics cards.
Of course, having higher clock speeds at same power draw is also a driving need. Together with that, acceptable heat output and effective driver support (which AMD Radeon is well known not to have compared to GeForce WHQL driver).
Beyond DX-R: The support for DirectX 12 Ultimate
AMD showcased some of the upcoming DXR enabled games, including Dirt 5 and Far Cry 6. Sure there are other games that support DirectX RayTracing through Nvidia RTX. But some of those titles also includes DLSS 2.0 support.
AMD Radeon isn’t just catching up with DX-R, but the whole DX12U package. As AMD specified it, it supports all four DX12U features: Ray Tracing, variable-rate shading, mesh shaders and sampler feedback. To recap, Ray Tracing is a rendering method which works by having light bounce over surfaces and display a certain level of lighting and reflections which pretty how it is in the real world. With this implemented in-game, ray-tracing in support games provides real-time rendering in games. Rate shading, mesh shaders provide better graphics performance in-game geometry.
In these cards, its ray-tracing rendering ill be using an accelerator in each CU, which will be using Infinity Cache.
AMD Smart Access Memory
Out of all the things AMD showcased, its Smart Access Memory piqued my interest. Unlike Nvidia and Intel (for now), AMD designs its own CPU, chipset and GPUs. It is only now AMD now took advantage of this with its upcoming CPU and GPU releases.
While being inconveniently vague, AMD is using a way to improve its memory performance in a system with AMD Ryzen 5000 series CPU, the 500 series chipset motherboard (B550 and X570) and the Radeon RX 6000 series. The best part is that this feature can be enabled via BIOS rather than using its Radeon software.
Specifications the RX 6000 Series
RX 6900 XT
RX 6800 XT
|ADDITIONAL POWER CONNECTOR||2×8 pin|
|BOOST FREQUENCY||Up to 2250 MHz||Up to 2105 MHz|
|GAME FREQUENCY||Up to 2015 MHz||Up to 1815 MHz|
|PEAK PIXEL FILL-RATE||Up to 288.00 GP/s||Up to 202.10 GP/s|
|PEAK TEXTURE FILL-RATE||Up to 720.00 GT/s||Up to 648.00 GT/s||Up to 505.20 GT/s|
|PEAK HALF PRECISION (FP16) PERFORMANCE||46.08 TFLOPs||41.47 TFLOPs||32.33 TFLOPs|
|PEAK SINGLE PRECISION (FP32) PERFORMANCE||23.04 TFLOPs||20.74 TFLOPs||16.17 TFLOPs|
|TRANSISTOR COUNT||26.8 B|
|GPU POWER||300 W||250 W|
|PSU RECOMMENDATION||850 W||750 W||650 W|
|INFINITY CACHE||128 MB||128 MB||128 MB|
|MEMORY SPEED (EFFECTIVE)||16 Gbps|
|MAX MEMORY SIZE||16GB|
|MEMORY TYPE (GPU)||GDDR6|
|MEMORY BANDWIDTH||Up to 512 GB/s|
|HDMI 4K SUPPORT||Yes|
|4K H264 DECODE||Yes|
|4K H264 ENCODE||Yes|
|DISPLAY PORT||1.4 with DSC|
|SLOT SIZE||2.5 slots||2 slots|
One of the noticeable factors of the RX 6900XT is the form factor and its power requirements. Usually, the flagship’s reference design will show some beef compared to its lesser spec variants. Not in this case. AMD isn’t using the new 12-pin power connector for any of its GPUs. But its reference version is doing just fine drawing additional power from a 2x 8-pin connector. It also has a requirement of 850w, while RX 6800 and RX 6800 XT need 750w and 650-watt. Preferably, its best to have 80 Plus Gold rated power supplies- something that neither AMD nor Nvidia bother to point out.
There are variable CU, RT Accelerators, ROPs, stream processor, texture units and transistor and hence power requirement. But what’s impressive is that all of its variants have 26.8 million transistors and 16GB GDDR6 operating up to 16 Gb/s. But all the cards operate at 256-bit memory interface.
The flagship RX 6900 XT…
The RX 6900 XT sounds and specs-wise compared to other variants looks like a flagship variant. With a price tag of $999 one can’t help but ask how its performance will be against the RTX 3090- or the RTX 3080. AMD did say Navi 21 chips are incredibly difficult to fabricate so it is not sure how many of these AMD can ship. -It has 80 CUs with game/ peak turbo clock at 2015 MHz/ 2250 MHz. AMD’s primary goal is to have these perform double than the RX 5000 series predecessor.
The mid-performer (?) RX 6800 XT
The Radeon RX 6800 XT has the same game/ turbo clock speed as the RX 6900 XT. Also, the same dimensions and additional power connector’s pin count. AMD considers this card to be 2x over the RX 5700 graphics card. It has 72 Compute Units with 128 MB Infinity Cache and the same amount of GDDR6 VRAM. As one would expect, this is supposed to compete with the RTX 3080.
The not-really-base model spec RX 6800
Out of all the Navi 21 core its releasing, the RX 6800 is the lowest of the three, with a price tag of $679. With 60 CUs and 1815/ 2015 game/ turbo clock speed. In this presentation, it compared it with the RTX 2080Ti. Of course, the RTX 3070 is something that it launched just a day before but I would not be shocked if AMD took an educated guess, or somehow it figured what Nvidia’s plans would be if they released these. But what’s also interesting is that AMD markets his for 4K gaming, a resolution that’s not really adopted widely, but prefers to not with the 1440p gaming performer even with the lowest spec of the three. It feels a little ambitious but we’ll know how it performs when the reviews roll out.
What about VR Gaming?
What’s not really mentioned it the VR support. There is a single-cable VR support on the RX 6000 series GPUs, and maybe supporting Type-C display at the same time? Not sure why AMD would skip emphasizing on this at the time VR games is slowly picking with titles like Half-Life: Alyx and also Phasmophobia.
Relying on benchmarks provided by chip makers should be taken with a bag of salt some our not-so-recent history taught us. The RX 6000 series are fated to be out by November 18th. But its best to wait until the benchmarks are released. There are still certain factors Nvidia takes the lead- DLSS 2.0 and NVENC. Nvidia also has a much better history of WHQL driver that works. But the driver is something that can be improved if AMD puts their heart and soul into it. If we see these cards for their actual performance- with or without DX-Ray tracing enabled. If these cards can match the current RTX 30 offerings while maintaining the same power consumption, this is the closest AMD gets to Nvidia till date since a very long time. Therefore one can’t help but imagine RDNA 3’s possibility. AMD did catch Intel with its pants down with Ryzen 2nd gen and 3rd gen. Nvidia is too relaxed with itself but this might wake them up. The question is if Nvidia will encourage to be more proactive with future releases and if Radeon can keep up or exceed. Interesting times.