Date: Tuesday , September 04, 2018
We have been trekking down a long path of GPU testing and comparisons. We are looking into the past at the generational upgrades in performance experienced in the last five years from NVIDIA and AMD. Our journey started with our article NVIDIA GPU Generational Performance Part 1. If you have not read this, go read this now. The first page will explain in detail what this series of reviews is all about. You can then follow with NVIDIA GPU Generational Performance Part 2. Finally, you can end the NVIDIA articles with NVIDIA GPU Generational Performance Part 3.
We are now moving away from NVIDIA and are now going to start our multi-part reviews on AMD GPUs. We are using the same premise and exploring the same goals as we did with the three NVIDIA articles. Similar to the NVIDIA articles this will be in multiple parts. We have to separate GPUs by price and intended market to correctly compare generational jumps in performance.
We are going to focus on what AMD has in the high-end GPU department for GPUs. This is where it gets tricky though because the high-end is not only made up of new Graphics Core Next (GCN) versions, but also refreshes and missing GPUs. There is a long period of time in the last five years that AMD "forgot" to make super-high-end GPUs.
The way we have to think about this is to look at the official GPU launches over the last five years. Focusing in on the highest priced single-GPU video cards meant to deliver the fastest performance for the enthusiast high-end market. We also have to look at the natural chronological order of GPU releases from year-to-year. Remember, our goal is to show you the "generational" jumps.
With NVIDIA this was easy, we went from Kepler to Maxwell to Pascal. With AMD it’s more complicated because all the video cards use the same core architecture called Graphics Core Next (GCN.) The only difference is that there have been changes or iterations of GCN that give some GPUs new functionality or improvements.
AMD does not have a clear naming scheme for these iteration changes. Therefore, we are going to assign a decimal system to it. For example, we will call the first iteration of GCN 1.0 for the Tahiti and Pitcairn GPUs. Then we will call the next one GCN 1.1 for Hawaii and Bonaire. Then we will call GCN 1.2 for the Tonga GPU (R9 285) which is a midrange card. Then we will call GCN 1.3 for the Fiji GPU (R9 Fury X) and GCN 1.4 would therefore be Polaris. Finally, GCN 1.5 would be for Vega10.
If any of this is confusing, back in 2015 we made a table comparison (scroll down) that compares GCN iterations. While we do not have GCN 1.4 or 1.5 on there, you can see up to GCN 1.3 what the updates to the architectures are.
In this way we can clearly understand what GPU "generation" is being compared. When we break it down like that, the natural progression of GPUs over time becomes very clear as to what we should compare. Between the years 2013 and 2018 we should compare Hawaii GCN 1.1, Fiji GCN 1.3 and Vega10 GCN 1.5 to each other. These are the "high-end" GPUs at the high-end pricing.
What this translates to in our testing is this comparison:
AMD Radeon R9 290X (Hawaii GCN 1.1)
AMD Radeon R9 390X (Hawaii GCN 1.1)
AMD Radeon R9 Fury X (Fiji GCN 1.3)
AMD Radeon RX Vega 64 (Vega10 GCN 1.5)
But wait, R9 390X is Hawaii GCN 1.1 just like R9 290X! Yes, it is. Radeon R9 390X is a refresh of R9 290X. However, we feel it is relevant and important to include in our testing for several reasons laid out below.
Firstly, The Radeon R9 390X was released almost a full two years after the Radeon R9 290X. The Radeon R9 390X is what the Radeon R9 290X should have been from the start, without the thermal GPU temperature problems with the reference card design. It was faster than the R9 290X, but also slower than the R9 Fury X, squeezing in right between.
The Radeon R9 390X went on to be a very popular and favored video card. It had a price reduction compared to R9 290X that made it very appealing. Here was a video card that was cheaper than R9 290X, but also performed faster and had a lot of potential for overclocking. This video card was adopted quickly and a lot of people went on to upgrade to this new video card.
It is important we have it in there so we can see the real-world advantage this upgrade had over R9 290X in almost two years of time. Then we can also see how much faster R9 Fury X was over R9 390X as an upgrade. When it comes down to it, despite it being a refresh, it was a very important card and well received. We must have that comparison in there because we bet there are still people out there now running this video card to this date. Now you can find out how much upgrading to Vega 64 will give you as an upgrade in performance.
Let’s quickly go over the specifications of each GPU as a refresher course to what we are looking at.
The AMD Radeon R9 290X was launched in October of 2013. The official MSRP was $549. This placed it below the cost of the GeForce GTX 780. This GPU is based on the GCN 1.1 architecture called Hawaii. The GPU is based on the 28nm manufacturing process. There are 2816 stream processors, 64 ROPs and 176 texture units. The GPU frequency can run up to 1GHz given appropriate cooling solutions. The memory bus was 512-bit with 4GB of GDDR5 running at 5GHz. This delivers 320GB/sec of bandwidth.
One of the main issues with the reference video card was that it ran so hot AMD had to provide a BIOS toggle switch to change the fan speed. In "Quiet Mode" the fan ran at 40% but this severely hampered performance. In "Uber Mode" the fan ran at 55% but even this low rotation rate caused the video card to underperform and not attain its actual reference clock speed. It wasn’t until custom retail cards with custom-cooling that the reference clock speed of 1GHz was achieved consistently.
It was almost a full two years later that the AMD Radeon R9 390X refresh was launched in June of 2015. The MSRP had officially been reduced to $429 for this video card, a great price for performance that was faster than R9 290X. This GPU was based on the same Hawaii GCN 1.1 architecture at 28nm.
Since this video card is a refresh of R9 290X it has the same layout, 2816 stream processors, 64 ROPs and 176 texture units. However, the GPU frequency clock speed got a bump up to 1050MHz instead of 1000MHz on R9 290X. What’s more, the R9 390X received a standard compliment of 8GB of GDDR5 versus 4GB on R9 290X. Even more, the clock speed for the memory was increased to 6GHz from 5GHz. This upped the bandwidth to 384GB/sec on the same 512-bit memory bus. TDP was 275W. Overall, the thermal problems with R9 290X were also improved with this refresh.
The AMD Radeon R9 Fury X was a very unique product for AMD. Firstly, this video card was released alongside the AMD Radeon R9 390X in June of 2015. The AMD Radeon R9 Fury X was based on the next iteration of GCN architecture called Fiji which we can call GCN 1.3. It had many improvements to the core architecture compared to Hawaii. The official MSRP was $649 and this was certainly AMD’s flagship GPU for 2015, and honestly all the way up to 2017 until Vega 64.
The Fiji GPU is based on the 28nm process. There are 4096 stream processors, 64 ROPs and 256 texture units. The GPU clock speed runs at 1050MHz. The memory is what was very unique on this video card. It contained 4GB of HBM memory which had a 4096-bit memory bus width. Operating at 500MHz/1GHz it provided 512GB/sec of memory bandwidth. Memory capacity, was therefore a big question or this video card’s longevity. However, it sure had gobs of memory bandwidth thanks to the HBM. The TDP was 275W.
The other unique feature of this video card is that it used a custom closed-loop cooling device. This video card was liquid cooled, and that kept the GPU frequency consistently at 1050MHz no matter what. Performance was exactly as it should be for the reference spec.
Finally, we come to the current flagship for AMD, the Radeon RX Vega 64 released in August of 2017. The official MSRP for the regular air-cooled version (non-Limited Edition) was $499. The Limited Edition was $599, and the Liquid Cooled Edition was $699. AMD Radeon RX Vega 64 is based on Vega10 GCN 1.5 architecture, which sports upgrades and improvements over Fiji and Polaris.
Vega10 is on the 14nm process. It has 4096 stream processors, 64 ROPs, and 256 texture units. The base clock runs at 1247MHz and the boost clock is 1546MHz. The video card sports 8GB of HBM2 memory at 2048-bit and 1.89GHz providing 485GB/sec of memory bandwidth. The TDP is 295W.
Though this video card does not have different fan profiles, if you are using the standard air-cooled reference video card, you will find it hampers performance slightly due to GPU temperature. The cooling is inadequate and keeps the GPU frequency from attaining its default reference frequency. If you use a faster fan profile on the video card, like ramping the fan up to 75%, or using a custom-cooled video card at reference specs you will find slightly higher performance. All that is happening is that with better cooling you are allowing the GPU to reach its intended frequencies.
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, Kingdom Come Deliverance 2018, Far Cry 5 2018.
Every video card mentioned above is being tested in every game above at 1440p. We are testing both high levels of graphics settings and low levels of graphics settings in each game.
An important consideration for this review is the fact that both the games and video cards have all been out for years now. This means all the games being used are patched up to their latest patches and have received all the patches and optimizations they are ever going to receive in their lifetime. Likewise, we are using the latest video card drivers, which also have now been optimized as much as they are ever going to be optimized for each video card.
That means now in 2018 we are experiencing the absolute best each video card has to offer in terms of performance in games, both the games and drivers are as optimized as they are ever going to get.