At the end of last year, I was researching GPUs like a madman, trying to find the best option for price and performance and maybe also have some headroom for a future CPU upgrade. My starting point was a Ryzen 5 2600, 16 GB of 3000 MHz CL15 RAM and an AMD RX 570 with 8 GB of VRAM. A very good performance per buck machine in the summer of 2019 for 1080p gaming. It was purpose-built to be cheap with an upgrade path in the near future. However, my inner hardware enthusiast didn’t want to be content. It also didn’t help that the two games I was playing at that time performed rather poorly (which was the games fault, but you take every excuse you can get to buy new stuff).
Putting that aside, I have data of three graphics cards to compare, tested in four games at three different in-game settings – plus a custom one for two games that I used for playing. In addition to that, I have a bit of CPU overclocking as a result of troubleshooting and a RAM upgrade from a 3000 MHz CL15 kit to a 3600 MHz CL17 kit – which is running at 3400 MHz. More wasn’t possible with this motherboard and CPU. This post isn’t about the CPU overclocking though. I did that to see if the 5700 XT was limited by the R5 2600 and would perform better with a faster CPU. Well no surprise there, but as it turned out, the numbers I found were not caused by the CPU. More on that later.
This is a ton of data and the outcome is not as obvious as it should be. Before I show you any numbers, I must set the stage, so you know how to read the tables. First off are the games:
- Assassin’s Creed Odyssey
- Ghost Recon Wildlands
- Far Cry 5
- The Division 2
Note that I completely relied on the games built-in benchmarks and I also didn’t do multiple runs and then averaged the numbers. I’m not a professional tester that has the time for this. I’m an enthusiast and I want to test my hardware and see the gains or losses in numbers. It’s my way of verifying the hardware works as expected – which one card did not (foreshadowing). Regarding the hardware and overclocking I used the following configurations:
|RX 570 Stock||No overclocking of anything, 3000 MHz RAM.|
|RX 570 OC||Only GPU overclocking to 1450 MHz core and a bit of VRAM, 3000 MHz RAM.|
|RX 5700 XT||No overclocking of anything, 3000 MHz RAM.|
|RX 5700 XT CPU @4 GHz||GPU stock, CPU running at 4 GHz, 3000 MHz RAM.|
|GTX 1660 Super Stock||No overclocking of anything, 3000 MHz RAM.|
|GTX 1660 Super RAM||No overclocking of anything, 3400 MHz RAM.|
You might notice that I hadn’t tested an overclocked CPU with the GTX 1660 Super. Ultimately, those few MHz don’t do that much, and I also was too lazy to go through all the games again, restarting them over and over. Tedious.
As I wrote in the introduction, my gaming resolution is 1920×1080. I used the top three graphics settings of the games, namely “Ultra”, “High” and “Medium”. In ACO and GRW “High” is called “Very High” and “Medium” is “High”. Makes sense, right? I will refer to those settings as “High” and “Medium” throughout this post. I also had a custom setting for Assassin’s Creed and Ghost Recon that I used for playing. In ACO it was a combination of “High” and “Very High” and for Ghost Recon I had to use “Medium” to “Low” settings (according to the games naming schemes). I won’t list the details, just the numbers. I benchmarked this to see how much gain there was at the visual fidelity I was used to.
Lastly: except for The Division 2, I will show the Average, Minimum and Maximum FPS. The Division 2 didn’t report them, only the Average. However, it listed the GPU and CPU usage which I will include instead.
(Please don’t comment that 1% and .1% low are the more valuable numbers. I know that. Again, not a professional tester)
So, let’s start with my custom settings in Assassin’s Creed Odyssey and Ghost Recon Wildlands. I don’t have any custom numbers for the other games because I didn’t play them when I had purchased the cards.
For the following results all numbers are Avg/Min/Max. The numbers for TD2 are the average framerate and in parenthesis the GPU usage in percent and the CPU usage in percent.
|AC Odyssey||GR Wildlands|
|RX 570 Stock||53/28/75||64/53/77|
|RX 570 OC||57/38/85||70/58/83|
|RX 5700 XT Stock||60/19/100||81/71/93|
|GTX 1660 Super Stock||71/28/113||95/82/109|
Apart from the useless minimum values in Assassin’s Creed, there is one thing immediately obvious: The RX 5700 XT only averaged 3 (!) FPS more in ACO when compared to the RX 570 OC. And in total it lost against the GTX 1660 Super.
Spoiler Alert: Although there will be a few numbers where the RX 5700 XT shines, this card lost the war. In a previous blog post I examined this phenomenon in more detail. Make sure to read it to learn what’s going on.
Let’s have a look at the game’s built-in settings next, so that you can compare the numbers with other tests, or your own benchmark runs.
Now I will also include Far Cry 5 and The Division 2. The following results also include the RX 5700 XT with an overclocked CPU and the GTX 1660 Super with faster RAM.
|AC Odyssey||GR Wildlands||Far Cry 5||The Division 2|
|RX 570 Stock||34/13/73||32/26/37||59/51/72||48 (98%, 30%)|
|RX 570 OC||37/26/55||35/29/40||62/55/76||53 (98%, 36%)|
|RX 5700 XT Stock||53/19/85||55/41/66||89/68/118||102 (97%, 73%)|
|RX 5700 XT CPU @4 GHz||55/20/84||57/45/67||94/72/124||102 (97%, 70%)|
|GTX 1660 Super Stock||49/27/91||50/42/57||84/73/95||70 (97%, 50%)|
|GTX 1660 Super RAM||53/23/109||50/41/57||89/77/107||69 (97%, 47%)|
At this level of graphical detail, it is obvious that the RX 5700 XT is the faster card compared to the RX 570. However, only in The Division 2 it can command the lead that I was expecting from this GPU. It is supposed to be roughly twice as fast as the RX 570. I have linked to my analysis of this issue in a previous paragraph, so here’s only the short version.
- Driver issues hamper the performance at 1080p.
- CPU too slow to push the GPU in all games.
- Outlier: The Division 2 with a very high CPU usage.
As you can see, in The Division 2 the CPU usage is above 70%. There were a few situations in the benchmark where the CPU hit 100% utilization. This game’s engine is very well optimized, thanks to a good DirectX 12 implementation and this helps overcome the driver issues that constantly lower the GPU’s frequency because it’s bored to hell.
The 1660 Super performs as expected, in all games. It is, however, already at its limit in a few games at this setting. In Far Cry it could probably do better if the CPU were a faster one as the faster memory did yield a few more frames. Assassin’s Creed also gained a few more fps on average and a bulk load on the max. That, however, is probably just a fluke. It’s too much for a bit faster memory.
|AC Odyssey||GR Wildlands||Far Cry 5||The Division 2|
|RX 570 Stock||45/24/67||50/41/57||63/54/78||71 (98%, 38%)|
|RX 570 OC||50/26/68||54/45/61||67/58/83||76 (98%, 38%)|
|RX 5700 XT Stock||57/19/83||70/57/81||97/77/131||137 (95%, 61%)|
|RX 5700 XT CPU @4 GHz||59/20/86||73/57/85||97/79/130||141 (95%, 60%)|
|GTX 1660 Super Stock||64/27/119||76/64/87||88/79/96||97 (97%, 67%)|
|GTX 1660 Super RAM||65/29/118||76/60/88||95/80/110||96 (97%, 62%)|
At High Details the RX 570 begins to deliver framerates that are almost playable or are already good enough. The 5700 XT is faster, and the margin is not small, but it’s not as dominant as it should be. Look at ACO and GRW. Yes, those games prefer NVIDIA GPUs, but the RX 5700 XT should be as far ahead of the GTX 1660 Super as it is ahead of the RX 570. I am aware that my CPU is not capable of delivering enough performance for the 5700 XT do be twice as fast in all games. But look at The Division 2. That’s how it is supposed to be. At this graphics setting the GPU isn’t event pushed to its limits anymore, but it’s still way faster than the rest. The 1660 Super on the other hand seems to be going as fast as it can except for Far Cry 5 where a noticeable performance increase can be had from the faster RAM.
We’ve come to the last batch of numbers.
|AC Odyssey||GR Wildlands||Far Cry 5||The Division 2|
|RX 570 Stock||52/22/73||58/48/69||70/60/84||94 (98%, 54%)|
|RX 570 OC||56/27/79||63/53/74||73/63/88||101 (98%, 51%)|
|RX 5700 XT Stock||61/20/110||79/61/90||97/80/129||160 (84%, 60%)|
|RX 5700 XT CPU @4 GHz||63/20/95||78/69/90||100/83/131||166 (85%, 59%)|
|GTX 1660 Super Stock||70/33/119||87/77/97||92/79/96||118 (91%, 71%)|
|GTX 1660 Super RAM||74/42/141||86/75/95||100/87/120||119 (92%, 68%)|
Now the RX 570 delivers a mostly good overall performance. In Assassin’s Creed it’s actually even playable at “High”, because you don’t need as much FPS in that game. Ghost Recon is still on the edge, hence my custom settings included quite a few low details to make it mostly smooth. There’s no issue with Far Cry 5 or The Division 2. As was the case with the other results, look at the numbers of the 5700 XT and the GTX 1660 Super. The NVIDIA GPU beats the AMD model in three out of four games and being very slightly slower counts as a win for NVIDIA because the 1660 Super shouldn’t just be “very slightly slower”. I still believe that the CPU is the reason the Far Cry 5 results don’t climb proportionally like in the other games, not the GPU. But since I didn’t do any CPU/GPU usage monitoring I guess we’ll never find out.
I hindsight, it would have been helpful. However, I didn’t expect to do so many tests. I thought I bought myself a mighty fast GPU in the RX 5700 XT and simply wanted to confirm my purchase. Then it would have just been a few tables with some numbers and that’s it. But here we are.
The Ryzen 5 2600 might not be the perfect CPU to elicit all the RX 5700 XT’s capabilities, I’ll admit to that. What I won’t accept is that it is still fast enough to drive the theoretically slower GPU to a win in most games. And it’s not just about the numbers. These numbers are merely an indicator that something is wrong. Far Cry 5 at “High” details still played worse at 70 – 80 FPS on the 5700 XT when compared to the RX 570 at somewhere in the 60-ies. There was hitching and stuttering that both other cards did not exhibit. I could have lived with these benchmark numbers, attributing it to the “slow” CPU, if the gameplay were smooth. After all, the intention was to pair this card with a much faster CPU sometime in 2020 and unleash it then. But it wasn’t smooth. It was horrible. It was worse in every way than the RX 570.
So, now I am back to NVIDIA with a worse performance per cost ratio, and all because of bad drivers. You know what? I’m happy. The 1660 Super works perfectly. The MSI model was expensive, but it’s good looking and very quiet. Unfortunately, MSI’s software for controlling the RGB is crap, but at least I found a tenable workaround for it.
I know that the RX 5700 XT performs very well at higher resolutions, so the hardware ain’t broke. But apart from the performance issues I also hated AMD’s drivers in general. The UI was confusing before the update in December 2019 and although it was prettier after that, it was still convoluted. What made it worse though, was the fact that for some reason AMD thought it was an excellent idea to play beeping sounds when Radeon Anti Lag and some other feature were enabled. And not just once when the game started. It happened every time I pressed the SHIFT or CTRL key. Imagine my surprise at where those noises came from after a simple driver update. Why, AMD? WHY? Sure, the NVIDIA control panel is slow and ugly. One might debate the last adjective because, after all, it looks like a regular Windows application rather than one of those terribly skinned shareware programs of the early internet ages. I really don’t understand the fetish of that, but that’s just me. Nevertheless, the control panel works, and I don’t get lost in it. GeForce Experience, yeah, well. It’s still a better experience than AMD’s UI in my opinion.
If there is anything to take away from this post:
- Don’t buy the RX 5700 XT for 1080p gaming! That’s the most important takeaway.
- Ryzen loves faster RAM – but that’s been known for a while.
- I didn’t just upgrade the RAM from 3000 MHz to 3600 MHz, I also doubled the capacity 😉.
- AMD, please improve your drivers.
- The 1660 Super works nicely with an R5 2600 but can also do a bit more as shown by the results with faster memory.
- If the mainboard’s BIOS wouldn’t suck at overclocking, I’d run the CPU at 4 GHz all cores all the time – which is besides the point, but I’ll mention it anyway.