What Is RDNA 3? AMD’s Upcoming Ada Lovelace Competitor Explained
NVIDIA’s Ada Lovelace GPUs are wildly powerful. The RTX 4090 and the RTX 4080 come with amazing generational improvements over their predecessors, the RTX 3090 and the RTX 3080. But, of course, AMD isn’t sitting around and just letting NVIDIA win. Instead, the company is promptly clapping back with the new RDNA 3 architecture, the successor to RDNA 2 and the architecture used by AMD’s best GPUs to date.
But what’s new in RDNA 3 exactly? And more importantly, should you buy an RDNA 3 GPU?
1. Improved 5nm Process
Compared to RDNA 2, RDNA3 will have a big focus on increasing not only performance but also efficiency, which AMD has two strategies for. The first one is the most obvious: a more efficient process.
While RDNA 2 GPUs were fabricated on TSMC’s 7nm process, the new RDNA 3 architecture will debut on its 5nm process instead. This is the same process used byNVIDIA’s RTX 4000GPUs and AMD’s own Ryzen 7000-series chipsets.Smaller chip manufacturing is important, but basically, using a more efficient, modern process allows chip makers to save money and for processors to be both faster and more power efficient.
5nm chips have been present in the smartphone ecosystem for years, and if you look at 2022 smartphones, many come with 4nm SoCs. These new GPUs and CPUs, including RDNA 3 GPUs, represent the first appearance of 5nm on PCs and set the stage for the next step down—probably 3nm—in a couple of years.
Die shrinks are important for innovation. But while die shrinks are important, they’re not the only card being played by AMD with its new generation of GPUs.
2. All-New Chiplet Design
The real change you would probably care about is the introduction of a new chiplet design for the GPU. Chiplets were one of the main reasons why AMD Ryzen chips performed so well and even managed to overtake Intel at their peak.
In essence, a chiplet design pretty much consists of having multiple, separate dies within the same package working together rather than having one single die with a lot of transistors. Why is spreading them out better than having them all together? Moore’s law dictates that the number of transistors in an integrated circuit doubles every two years. But it’s getting increasingly difficult to cram more and more transistors in a single package, something that in turn results in lower yields. In that regard, chiplets are helpful.
With chiplets, you may keep adding transistors without fearing lowered yields. After all, during chip manufacturing, it’s way easier, not to mention better for everyone, to throw away a defective chiplet than throwing out a whole, defective monolithic CPU package.
CPUs have had chiplet designs for some time thanks to Ryzen, but it had yet to make its way to the GPU space. With RDNA 3, that’s changing, with at least some GPU packages confirmed to come with a chiplet design rather than a monolithic die design. That might bring both performance and efficiency advantages, with AMD claiming that RDNA 3 is bringing 50% better performance-per-watt than the previous generation.
3. Next-Gen Infinity Cache
The biggest improvement on RDNA 2 GPUs was theintroduction of the Infinity Cache, with AMD saying it’s responsible for a big part of that generation’s gaming gains. RDNA 3 will not only come with Infinity Cache but with a “next-gen” version that brings the bar up even further.
Infinity Cache is, basically, a cache level that takes a similar role to what the L3 cache would take in a CPU. Unlike CPUs, graphics cards don’t usually have an L3 cache, and we normally have just L0, L1, and L2. AMD is taking a page from its CPUs and adding a large, internal cache the GPU can access.
It’s not clear yet how this “next-gen” Infinity Cache will be better than what we have right now on AMD GPUs. One option that has been rumored is that AMD might implement its 3D cache tech into Infinity Cache. 3D cache does wonders for CPU gaming performance, seeing as the Ryzen 7 5800X3D is even better than AMD initially thought and performs better on games than Ryzen 7000 chips, so if we see 3D cache on AMD GPUs, we might be in for a nice bump in performance. Other improvements could also be made to things like bandwidth to improve the communication between the cache and the GPU itself.
Should You Buy an RDNA 3 GPU?
We won’t be able to answer that question until AMD finally releases its new GPUs. Of course, we see improvements, but most of what we’ve detailed in this post is based on leaks and rumors. It won’t be too long until we know, though, of course, as AMD is getting ready to announce its new GPUs, likely launching in November 2022.
Right now, though, we can only speculate. It looks like all of these improvements will mean that the new Radeon GPUs will have enough horsepower to face off, or maybe even beat, NVIDIA’s RTX 4000 GPUs, but then again, only real-life testing will be able to explain the whole story.
The RDNA 3 architecture will be available on gaming GPUs initially, and it will most certainly make it to APUs and integrated graphics down the road. RDNA 2 even made it to smartphones thanks to Samsung’s Xclipse 920 mobile GPU, which is used in theExynos 2200 chipsetsthat power Samsung Galaxy S22 smartphones in Europe. It was also used on consoles like the Xbox Series X and the PlayStation 5. So who knows? Maybe RDNA 3 will make it widely, too.
Don’t forget, it’s not just Intel vs. NVIDIA this time, either.Intel’s Arc GPUs are also nearing release, and while these will deliver performance similar to the existing NVIDIA 3000 Series and AMD 6000 Series GPUs, it’s another GPU option for gamers.
NVIDIA’s Competition is Not Backing Down
Competition is alive and well, and if NVIDIA is coming out with big guns, then AMD has big guns of its own, too. The new RDNA 3 GPUs are looking great, and we can’t wait to know more about them.
Looking for a fast graphics card for Linux gaming? You have two main options: AMD and Nvidia. Here’s what you need to know.
You can’t call this offline, Notion.
Flagship price, mid-range phone.
You’re not getting the most out of what you pay for iCloud+.
One casual AI chat exposed how vulnerable I was.
Your phone’s camera app doesn’t show this, so it’s easy to miss.
