Date: Thursday , September 20, 2018
The reason to focus on this level of video cards is mainly because of AMD’s Polaris GPUs. AMD’s Polaris GPUs were very important for AMD as they heralded in large upgrades over previous iterations of the GCN (Graphics Core Next) architecture. Until Vega launched, Polaris was AMD’s fastest GPU for a long while when AMD had nothing new in the high-end to compete. Therefore, examining performance leading up to Polaris, and finding out the performance improvements since 2013, is important.
If you are new to this series we have been working on, and seeing that this is Part 2 please check out our AMD GPU Generational Performance Part 1 evaluation to learn what we are doing. In Part 1 we explored the high-end AMD GPUs of Radeon R9 290X, Radeon R9 390X, Radeon RX Fury X, and Radeon RX Vega 64.
Comparing AMD GPUs over the last five years is tricky, it was for the high-end comparison and it is also the case in the mainstream. The issue is that there have been many re-brands and refreshes over the course of time here to contend with.
Throughout this evaluation we are going to use two different terminologies, so we need to define them. We will be using the words "re-brand" and "refresh." These mean two different things, even though they sound familiar. The definition we are using is such: A "re-brand" is a strict re-branding only of the name, no architectural or performance differences between video cards. A "refresh" is a re-branding plus the addition of architectural improvements or performance differences, yet still ultimately based on the DNA of previous generations. In a perfect world we want to see neither of these and progress to whole new architectures.
Similar to the high-end there are different GPUs that represent different GCN (Graphics Core Next) iterations or evolutions. For example, the first iteration of GCN is Tahiti and Pitcairn GPUs and is called GCN 1.0. Then the next GPUs Hawaii and Bonaire are GCN 1.1. Then the next GPU Tonga is GCN 1.2. Then the next GPU Fiji is GCN 1.3 and the next GPU Polaris is GCN 1.4 and finally the latest Vega would be GCN 1.5.
In our high-end Part 1 evaluation we looked at Hawaii GCN 1.1, Fiji GCN 1.3 and Vega GCN 1.5.
Today, in our Part 2 evaluation we will be looking at Tahiti GCN 1.0, Tonga GCN 1.2 and Polaris GCN 1.4.
The GPUs we will be evaluating are Radeon R9 280X (Tahiti GCN 1.0), Radeon R9 380X (Tonga GCN 1.2), Radeon RX 480 (Polaris GCN 1.4) and Radeon RX 580 (Polaris GCN 1.4.) These are the major releases from 2013 to 2018 in the $200 mainstream video card range.
Where the re-branding and refreshing comes in starts with the Radeon R9 280X. It is in fact a re-brand of the Radeon HD 7970 GHz Edition video card. It is simply the Radeon HD 7970 GHz Edition renamed, but with the same specs and clock speeds. The Radeon HD 7970 GHz Edition was launched in 2012.
Therefore yes, we are starting with a re-branded video card, but it is the official release for 2013 even though it is the same as the Radeon HD 7970 GHz Edition. It is still GCN 1.0 Tahiti, and represents the first iteration of GCN. The Radeon HD 7970 GHz Edition launched at $499. Therefore, the Radeon R9 280X is basically bringing this previously high-end card for 2012 down to a lower price range ($299) for 2013 and starting the whole "280X-380X-480-580" mainstream lineup up to now.
The next refresh comes with the Radeon R9 380X. Its history is a little more complicated. You see, the Radeon R9 380X is really a refresh of the Radeon R9 285 which is a refresh of the Radeon R9 280X but also a cut-down version of the R9 280X. The Radeon R9 285 started a new iteration of the GCN architecture called Tonga and is GCN 1.2. It has improvements architecturally over Tahiti GCN 1.0 and even Hawaii GCN 1.1. The Radeon R9 285 launched in 2014 at $249. However, it is a cut-down version of R9 280X in the sense that it has less stream processors, less texture units, a lower clock speed and memory speed and only has 2GB of VRAM.
The reason we are skipping it though, and going with the launch of the Radeon R9 380X in 2015 is because the R9 380X is actually the fully realized version of the Radeon R9 285. You see the Radeon R9 285 was a cut-down version of what Tonga as a GPU could offer, it had less stream processors, less clock speed, less memory speed, less texture units and only 2GB of VRAM. It was a new iteration of GCN, but not the fully realized and capable version of that GPU.
It was Radeon R9 380X in 2015 that refreshed the Tonga GPU and made a fully realized version of it with full stream processors and clock speeds and 4GB of VRAM. It is the same Tonga GCN 1.2 architecture, just with all the stream processors and units turned on. Therefore, if we want to test the full benefits of Tonga GCN 1.2 over Tahiti 1.0 (R9 280X) then we have to use the Radeon R9 380X. Otherwise, if we used the R9 285 it would for sure be slower than the Radeon R9 280X since it is a cut-down version.
The next GPU after Radeon R9 390X is of course Radeon RX 480, Polaris released in 2016. That GPU marks a big improvement because of the new 14nm FinFET process. It is also a big GCN upgrade in terms of architectural improvements, so it is neither a refresh or a re-brand, it is truly a big upgrade. However, the next upgrade after it, launched in 2017 the Radeon RX 580 is very much a small refresh. The Radeon RX 580 only improved upon the clock speed of the Radeon RX 480. The RX 580 runs at 1340MHz versus the RX 480’s 1266MHz. Therefore, the RX 580 is very much a refresh product, almost a re-brand in fact with such minor changes.
The AMD Radeon R9 280X was launched in October of 2013 at an MSRP of $299. Out of the cards we are evaluating in this article, this was the most expensive at launch. The AMD Radeon R9 280X is based on the Tahiti GCN 1.0 architecture at 28nm. This video card is a re-brand of the Radeon HD 7970 GHz Edition video card. There are 2048 stream processors, 32 ROPs, 128 texture units and a boost clock of 1000MHz (1GHz) on the GPU. It uses 3GB of 6GHz GDDR5 memory on a 384-bit memory bus producing 288GB/sec of memory bandwidth. The TDP is 250W.
The AMD Radeon R9 380X was launched two years later in November of 2015. The AMD Radeon R9 380X was more than just a re-brand of the Radeon R9 280X, it was a refresh, however it does share a lot in common with the Radeon R9 280X. What’s different about it is that it launched at an MSRP of $229, much cheaper than the R9 280X launched at. It was also based on what is called the Tonga architecture which debuted with the Radeon R9 285. This is GCN 1.2 and supports architectural improvements even over the Radeon R9 390/X which is based on the Hawaii GCN 1.1 architecture. That means, in terms of architectural improvements it is two evolutions newer than the Radeon R9 280X, and one evolution newer than Radeon R9 390X.
To learn what the architectural improvements that Tonga brought over Tahiti and Hawaii look at our table on the bottom of this page. In the column labeled "GCN Third Iteration" you will see the improvements to the architecture that Radeon R9 380X has over Radeon R9 280X. Notably improved tessellation performance, delta color compression (that’s a big one as it alleviates memory bandwidth), updated ISA instruction set, high quality scaler.
While it has an updated architecture, a lot of the base specifications are still the same as Radeon R9 280X. For example, it has the same number of stream processors at 2048, the same ROP count at 32 and the same texture unit count at 128. Oddly, the GPU clock speed was brought down and lowered to 970MHz on the Radeon R9 380X versus the 1GHz on the R9 280X. This means it actually has a lower pixel and texture fillrate. It also has a narrower memory bandwidth of 256-bit, though more VRAM with 4GB of GDDR5 but also running slower at 5.7GHz. This all adds up to reduced memory bandwidth of 182GB/sec on R9 380X versus 288GB/sec on R9 280X.
So it’s slower right? Well, no, and that is because of the delta color compression. The R9 380X can get away with a lower memory bandwidth because of the introduction of this very good delta color compression which saves memory bandwidth. Also, the slightly improved architecture can make up the rest of the difference. The end result is that we expect performance to be right at the performance of 280X, or slightly faster. What we will be looking for are scenarios where games utilize the updated architecture of the R9 380X better, and thus have improved performance over R9 280X. Still, we don’t suspect it will be a lot fester, only slightly.
The AMD Radeon RX 480 was launched one year later in June of 2016. The 8GB model had an MSRP of $239 at launch, and there was a 4GB model at $199. The AMD Radeon RX 480 was a much larger leap in architectural improvements over the R9 380X. The AMD Radeon RX 480 is based on the Polaris 10 GPU and would be GCN 1.4. What really separates it from the others is that AMD was able to use a new 14nm FinFET process which allows more of everything and much higher clock speeds.
To learn what the additional GCN 1.4 architecture improvements are check out this slide. You will see things such as improved geometry engine, compression and memory controller updates, better buffers, improved shader instructions, better async compute and other things. In terms of evolutionary upgrades in the $200 video card range, the Radeon RX 480 was a big upgrade from Radeon R9 280X and R9 380X which basically were refreshes stemming back to the Radeon HD 7970.
The Radeon RX 480 has 2304 streaming processors, 32 ROPs and 144 texture units. It runs at a base clock of 1120MHz and a boost clock of 1266MHz. It uses 7 or 8GHz GDDR5 4 or 8GB on a 256-bit memory bus giving us 256GB/sec of memory bandwidth with the 8GB model. It has a 150W TDP.
The AMD Radeon RX 580 is the latest in AMD’s arsenal in the $200 price range, and was launched in April of 2017 over a year ago. This video card can be considered a refresh of Radeon RX 480, though it edges closely to being borderline a re-brand. The only difference is a GPU clock speed boost over the RX 480, and that is it. Technically this raises the pixel and texture fillrate, giving it a different and improved performance profile, but it isn’t much as the boost in clock speed is minor.
The Radeon RX 580 launched at $229 with 8GB of VRAM and is based on the same Polaris 10 GPU as Radeon RX 480. It is also GCN 1.4 and at 14nm. At least Radeon R9 380X had some architectural improvements over R9 280X. With the RX 580 though, it literally has no upgrades architecturally over RX 480. It contains the same 2304 stream processors, 32 ROPs and 144 texture units. It has the same 8GB of GDDR5 at 8GHz on a 256-bit bus producing 256GB/sec of bandwidth. TDP is 185W.
The only difference is that the base clock is 1257MHz and the boost clock is 1340MHz on the Radeon RX 580 versus the RX 480’s 1266MHz boost clock. That’s an increase of only 74MHz. That is the only thing that separates the RX 480 from the RX 580 in performance at reference specifications. We therefore do not expect the Radeon RX 580 to be that much faster than the Radeon RX 480, just very minor differences.
The games we have chosen are, from earliest to latest: Crysis 3 2013, Tomb Raider 2013, Grand Theft Auto V 2013-2015, Far Cry 4 2014, The Witcher 3 2015, Fallout 4 2015, Rise of the Tomb Raider 2016, DOOM 2016, Deus EX Mankind Divided 2016, Battlefield 1 2016, Sniper Elite 4 2017, Mass Effect Andromeda 2017, Wolfenstein II: The New Colossus, Kingdom Come Deliverance 2018, Far Cry 5 2018.
Every video card mentioned above is being tested in every game above at 1080p. We are testing both high levels of graphics settings and low levels of graphics settings in each game.