Hot Damn, AMD Did It

Hot Damn, AMD Did It
Screenshot: AMD

AMD put on one hell of a show-stopper when it revealed all the details about its next-gen processors earlier this month. And it’s done it again with its full-reveal of its next-gen graphics cards today. This may very well be the generation of AMD hardware that finally puts it neck-and-neck with its competitors, not just slightly behind in a close second.

But not content to just talk about its upcoming graphics cards, which have some interesting architectural changes, AMD also announced a boatload of new software and features that strengthen its ecosystem — and maybe even offer a little incentive to go all AMD on your next build.

To start, here’s the good stuff:

  • Radeon RX 6900 XT: 80 compute units; 2015 MHz game clock (2250 MHz boost); 128 MB infinity cache; 16 GB GDDR6 memory; 300W TBP; $US1,000 ($1,415)

  • Radeon RX 6800 XT: 72 compute units; 2015 MHz game clock (2250 MHz boost); 128 MB infinity cache; 16 GB GDDR6 memory; 300W TBP; $US650 ($920)

  • Radeon RX 6800: 60 compute units; 1815 MHz game clock (2105 MHz boost) 128 MB infinity cache; 16 GB GDDR6 memory; 250W TBP; $US580 ($821)

Both the RX 6800 XT and 6800 will be available starting November 18, with the RX 6900 XT following on December 8.

Before I start comparing the RX 6000 cards to the RTX 3000 cards, let’s talk architecture and software. Other than making the compute units 30% more power-efficient, and increasing the frequency by 30% (which is no small feat in itself), with its RDNA2 architecture, one of the big changes AMD is ushering in with its RDNA2 architecture is what it calls Infinity Cache. It’s based on what AMD has done with high-density L3 caches in its Zen product line and adapted for its new GPUs. Essentially, this is a little bit of extra storage space, 128 MB, that the GPU can tap into on top of the already available bandwidth on the card itself, giving up to a 0.9x power boost and up to a 2.17x bandwidth boost, according to AMD.

These new GPUs will also have hardware-accelerated ray tracing, and support for variable rate shading, mesh shaders, and sampler feedback — all the same advanced graphical features Nvidia supports with its RTX 3000 series cards. In addition to supporting DirectX12 Ultimate, which makes it easy for game developers to port their games between PC and console (important for AMD to have since its GPUs are in the new PS5 and Xbox Series X/S), AMD will also support DirectStorage API in the future, which it says will improve how fast games load from SSDs. Combined with PCIe 4.0 storage and a motherboard that supports the new standard, it sounds like it’s a decent improvement, but AMD did not go into detail about how much faster it could be.

Screenshot: AMD Screenshot: AMD

AMD is bringing back Rage Mode, too, or one-click overclocking, which harkens back to the ATI GPU days. All new Radeon graphics cards will come with a software package that allows you to overclock your card with a single mouse click. It’s basically an auto-overclocking feature, so if you don’t want to mess around in your BIOS or any other tools, you don’t have to. This feature will be available at launch.

Lastly, AMD is introducing what it calls Smart Access Memory. If you build a system with an AMD Ryzen 5000 processor, plus an AMD-compatible motherboard, plus an AMD Radeon 6000 graphics card, your CPU will be able to access all of the GPU’s 16 GB of memory and vice versa. This feature will need to be enabled on the motherboard chipset, but it allows the CPU and GPU to talk directly to one another. This allows your system to rely less on its RAM for various tasks, which should improve loading times according to AMD.

Screenshot: AMD Screenshot: AMD

All that said, the RX 6900 XT, RX 6800 XT, and RX 6800 are positioned to seriously compete with Nvidia’s RTX 3000-series line-up, even without paying much attention to the gameplay stats in AMD’s presentation. First and foremost, there are some terminology differences in the way both AMD and Nvidia break down the specs of their graphics cards. CUDA cores aren’t direct equivalents to compute units, and base clocks aren’t necessarily the same thing as game clocks, it can be a little challenging to compare both companies’ cards by numbers alone. If you’ve been following both companies for at least the last two years, then this will probably be old news to you.

AMD introduced its game clock concept with the announcement of its RX 5000-series. Instead of highlighting the absolute lowest frequency of its graphics cards, it instead now highlights the lowest frequency users can expect to get while gaming. It’s kind of a marketing tactic more than anything else, but it’s important to note that game clock and base clock are not the same thing, while boost clock has the same meaning for both AMD and Nvidia. So just keep those things in mind when comparing base clock and game clock.

Screenshot: AMD Screenshot: AMD

Compute units (or stream processors) versus CUDA Cores gets a little more tricky. Because every GPU manufacturer defines its architecture in its own way, there really isn’t a comparison between these two. Even if you get down to roughly the same number of “cores,” actual performance can differ a little bit or a lot. Which is why gaming benchmarks or other performance measures are crucial when comparing two cards from different manufactures.

However, AMD says its new Radeon RX 6800 XT will get equivalent or better performance in most games than Nvidia’s RTX 3080 due to its RDNA 2 architecture and other features outlined above. That’s at 4K and 1440p, too. Same with its other 6000-series GPUs. And with Rage Mode and Smart Access Memory turned on, AMD says you’ll get up to an extra 8% frame rate bump. AMD didn’t touch on what its ray tracing performance will look like, but it could very well rival what Nvidia has been able to do. With the cost of these cards all undercutting Nvidia’s rivals, I think I’m even more excited to get my hands on these cards than I was the RTX 3080 and RTX 3070.