If you were looking for a new graphics card recently, you probably stumbled upon these terms and wondered what do they even mean? It’s okay to be confused, especially since Navi and RDNA have been used interchangeably when maybe they shouldn’t have. It’s time to get an understanding of what’s what.
As this is a rather complicated topic, I’ll cover the basics first and not get too technical.
First, we have to explain the Instruction Set Architecture or ISA which can be thought of as an abstract model of a computer. Its primary tasks are to define the supported data types, registers, hardware support for managing main memory, and the input/output model of a family of implementations of the ISA. Or simply, it’s an interface between hardware and software. And Microarchitecture is the way a given ISA is implemented in a specific processor.
Tying it all together, Computer Architecture is the combination of Microarchitecture and ISA.
For our conversation, RDNA and GCN are the codenames for both ISA and Microarchitecture for AMD GPUs, and Navi is the codename for the GPUs made with these architectures. Okay, now that we got that out of the way, it’s time to get into the specifics.
Table of ContentsShow
Graphics Core Next (GCN)
Let’s start with Graphics Core Next or GCN.
This is AMD’s codename for both a series of microarchitectures and an instruction set which was the for the first time a part of a released product all the way back in 2012.
From the start, they used PCIe 3.0 standard, but compared to today’s standards, these products aren’t as power-efficient, although that can be somewhat expected. Cards prior to GCN 4.0 were made on the 28nm manufacturing process, but those after that point used the 14nm process.
Another thing that these cards had going for them from the get-go is that they supported GDDR5 memory, and some later ones also supported HBM2 and HBM memory.
It’s worth noting that GCN is also used in graphics portions of AMD’s Accelerated Processing Units such as those in Playstation 4 and Xbox One APUs.
Radeon DNA (RDNA)
RDNA is a direct successor to GCN and it was featured for the first time on the Radeon RX 5000 series that launched in 2019, so it’s a fairly new product.
As expected, it strives to improve upon GCN’s capabilities as it has support for GDDR6 memory and PCIe 4.0 bus interface that almost doubles the bandwidth offered by PCIe 3.0. It features enhanced rendering pipeline, multilevel cache, operating primitive shaders, and a modernized display controller and is also being manufactured using the 7 nm process.
Navi is the name of the GPU series built on the RDNA architecture with FinFET’s 7nm process. This is probably where all the confusion happens, but you can consider RDNA a broader term as Navi GPUs have so far been exclusively found in the RX 5000 series.
GCN vs RDNA
As things usually work in the technology world, newer equals better and that is exactly the case here. Not saying that GCN isn’t good, it’s just that for the latest needs, it doesn’t do as good of a job as RDNA does.
The new processing units of RDNA were redesigned to offer efficiency improvement and improve upon the single-threaded performance when compared to the older GCN units and it shows. RDNA’s IPC is 1 and that is a big upgrade on GCN’s 0.25. What this means is that RDNA can perform four times as many instructions per cycle and this is really important for gaming.
One of the biggest improvements of RDNA is the width of the wavefront which is 32 compared to GCN’s 64. The math here might seem confusing, but for GCN, this meant that 64 threads were bundled in together for execution. What this does is increase single-threaded performance by decreasing clock cycles and that really comes in handy in games. AMD has done another thing to boost RDNA and that is to expand the width of their SIMDs from 16 to 32, which means that the size of the wavefront now matches the size of the SIMD.
GCN put together the shader hardware into “compute units” which contained memory access, LDS, and scalar ALUs and vector ALUs. One of these units contains 4 SIMD16s that share one path to memory. On the other hand, RDNA has a “workgroup processor” that replaces the compute unit as the base shader computation unit with a 2 to 1 ratio. In practice, this allows RDNA to have more computing power and also more memory bandwidth that can be steered at a single workgroup.
RDNA also innovated with L1 cache grading. This hierarchy lowers the cache latency for every level, increases load bandwidth two-fold, and improves upon effective bandwidth. So for the same amount of power, a double amount of work is now possible.
RDNA vs Navi
There isn’t really much to talk about here, but seeing how RDNA came out a clear winner when compared to GCN, I thought it’d be nice to mention Navi as well. As I’ve previously stated Navi refers mostly to the RX 5000 series’ use of RDNA, so these two terms are about as comparable as apples and oranges. Well, maybe apples to granny smith apples.
RDNA is clearly the most current technology here and as such is probably considered the safest buy right now
However, it is also very important to note that AMD has announced RDNA 2 release as early as late-2020. It’s been confirmed that this new technology, also dubbed “Big Navi” which will probably cause even more confusion, will be a part of Microsoft’s next-gen Xbox Series X and its Sony competitor Playstation 5.
So, if you’re patient you might be able to get your hands on even newer technology in the next couple of months.