NVIDIA’s Ada Lovelace GPUs are wildly powerful. The RTX 4090 and RTX 4080 come with surprising generational improvements over their predecessors, the RTX 3090 and RTX 3080. Instead, the company is immediately clapping back with the new RDNA 3 architecture, the successor to RDNA 2 and the architecture used by AMD’s best GPUs.
Thank you for reading this post, don't forget to subscribe!Thank you for reading this post, don't forget to subscribe!But what exactly is new in RDNA 3? And more importantly, should you buy the RDNA 3 GPU?
1. Improved 5nm Process
Compared to RDNA 2, RDNA3 will have a bigger focus on increasing not only performance but also efficiency, for which AMD has two strategies. The first is the most obvious: a more efficient process.
While the RDNA 2 GPUs were manufactured on TSMC’s 7nm process, the new RDNA 3 architecture will instead debut on its 5nm process. This is the same process used by NVIDIA’s RTX 4000 GPU and AMD’s own Ryzen 7000-series chipsets. Smaller chip manufacturing is important, but fundamentally, 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, which include RDNA 3 GPUs, represent the first appearance of 5nm on PCs and set the stage for the next step—perhaps 3nm—in a few years.
Die shrinkage is critical to innovation. But while die shrinkage is important, they aren’t the only card played by AMD with its new generation of GPUs.
2. Brand New Chiplet Design
The real change you’ll probably care about is the introduction of a new chiplet design for the GPU. Chiplets were one of the main reasons why the AMD Ryzen chips performed so well and even managed to overtake Intel at their peak.
In essence, a chiplet design consists of multiple, separate dies within a single package working together, rather than a single die with many transistors. Why is it better to spread them out than to have them all together? Moore’s Law states that the number of transistors in an integrated circuit doubles every two years. But it’s becoming increasingly difficult to cram more and more transistors into a single package, resulting in lower yields. Chiplets are helpful in this situation.
With chiplets, you can keep adding transistors without fear of decreasing yield. After all, during chip manufacturing, it’s easier to throw away a defective chiplet than it is to throw away a good monolithic CPU package, not to mention better for everyone.
CPUs have had chiplet designs for a while thanks to Ryzen, but it has yet to make its way into the GPU space. With RDNA 3, that’s changing, with at least some GPU packages confirmed to come with a chiplet design instead of a monolithic die design. This can bring both performance and efficiency gains, with AMD claiming that RDNA 3 is bringing up to 50% better performance-per-watt than the previous generation.
3. Next-Gen Infinity Cash
The biggest improvement in the RDNA 2 GPUs was the introduction of Infinity Cache, which AMD said is responsible for a large portion of that generation’s gaming gains. The RDNA 3 will not only come with Infinity Cache but also a “next-gen” version that raises the bar even higher.
Infinity cache is, basically, a cache level that plays a role similar to the role of L3 cache in a CPU. Unlike CPUs, graphics cards usually don’t have L3 cache, and we normally only have L0, L1, and L2. AMD is taking a page from its CPUs and adding a large, internal cache that the GPU can access.
It’s not yet clear whether this “next-gen” Infinity Cache will be better than what we have on AMD GPUs right now. One rumored option is that AMD could implement its own 3D cache technology in Infinity Cache.
The 3D cache does wonders for CPU gaming performance, considering the Ryzen 7 5800X3D is even better than AMD initially and performs better on games than the Ryzen 7000 chips, so if we look at the 3D cache on AMD GPUs, So we could be in for a good collision. in performance. Other improvements could also be made to things like bandwidth to improve communication between the cache and the GPU.
Should you buy the RDNA 3 GPU?
We won’t be able to answer that question until AMD finally releases its new GPUs. Of course, we do see improvements, but everything we have detailed in this post is based on leaks and rumours. It won’t be too long until we know, though, as AMD is getting ready to announce its new GPUs, possibly launching in November 2022.
However, for now we can only speculate. It seems likely that all of these improvements will mean that the new Radeon GPUs will have enough horsepower to compete with, or even beat, Nvidia’s RTX 4000 GPUs, but again, only real-life testing. Only he would be able to explain the whole story.