Horizon Zero Dawn PC Performance Analysis – You Love To See It Here

Keith May
Horizon Zero Dawn

EDITORS NOTE: Our performance evaluation was done entirely post-patch that was issued after we had access to the game for several days resulting in retesting most of our cards. When viewing performance issues with this game it's important to know whether the evaluation was made on the original build of the game or the patched version that matches what will be released when the game goes live.

Hot on the heels of Death Stranding we PC Gamers get access to another hot exclusive to the Playstation 4 world, Horizon Zero Dawn Complete Edition. Horizon Zero Dawn has been out since February of 2017, so it's not like it's a new game, but it's a very welcome addition to the PC universe. If you're looking for a review of the game itself may I direct you to our review of the original release by Chris Wray because the Horizon Zero Dawn PC release is essentially the same game therefore the original review of the game itself still holds up. But, this is all about the technical side of things.

Horizon Zero Dawn for PC is still powered by the Decima engine, just like Death Stranding, but it comes with fewer bells and whistles, most notably is the absence of DLSS 2.0, and unfortunately no FidelityFX as well. But, unlike early reporting, I'm happy to say that ultrawide support is present as well as a FOV slider, these were originally said that they were not going to be included in order to 'preserve the original experience of the game', but I'm very happy to see them included. As you'll see down the way, Ultrawide is a bit more stressful than usual in this open-world adventure. The biggest benefit of this game on PC (outside of increased resolution and framerate) is the ability to crank way beyond the original release in draw distance as well as the excellently enhanced ambient occlusion implementation that really makes the world pop.

Related StoryFrancesco De Meo
A Horizon Zero Dawn PlayStation 5 Remake Is in the Works, Alongside a Multiplayer Game – Rumor

Testing Methodology

First, I want to make a request to Sony and Guerrilla, GET RID OF THE 10 MINUTE OPTIMIZATION WINDOW for when you update your driver or change any hardware. Okay, got that out of my system. I want to drop a good tip on this game when you launch it, alt+enter to take it out of full-screen and then alt+enter back in, there is a weird bug that 'should' be addressed where the game is rendered at the set resolution but is downscaled then upscaled back and it just looks odd.

Horizon Zero Dawn comes to us with several arrays of presets; Favor Performance, Original, Favor Quality, and Ultimate Quality. But, that being said, there seems to be a change in presets across GPU configurations resulting in the Ambient Occlusion settings being variable, so for our testing, we used the Ultimate Quality settings but changed the Ambient Occlusion setting to 'Original' for consistency. Horizon Zero Dawn also included a very solid and representative (from my experience post-patch) built-in benchmarking utility. We used that for our testing and took the results from the 176-second run.

Once we had the results from 3 runs, after discarding an initial burner run for loading purposes, we took the average of average frame rates as well as the 1% Percentile results from the run. We report our performance metrics as average frames per second and have moved away from the 1% and .1% reporting and are now using the 1% Percentile. For those uncertain of what the 1% Percentile is, representing is easily explained as showing only 1 frame out of 100 is slower than this frame rate. Put another way, 99% of the frames will achieve at least this frame rate.

*at the time of our review a known issue with the game is that Anisotropic Filtering is not functioning, but in previous games we found this setting to be superfluous in affecting overall performance.

horizon-zero-dawn-complete-edition-screenshot-2020-08-02-06-40-07-72-2
horizon-zero-dawn-complete-edition-screenshot-2020-08-06-17-08-07-06

Test System

ComponentsZ370
CPUIntel Core i9-9900k @ 5GHz
Memory 32GB Mushkin Redline DDR4 3600
MotherboardEVGA Z370 Classified K
StorageKingston KC2000 1TB NVMe SSD
PSUCooler Master V1200 Platinum
Windows VersionLatest verion of windows at the time of testing
Hardware-Accelerated GPU SchedulingOn if supported by GPU and driver.

Graphics Cards Tested

GPUArchitectureCore Count
Clock SpeedMemory Capacity
Memory Speed
NVIDIA RTX 2080ti FETuring43521350/163511GB GDDR614Gbps
NVIDIA RTX 2080 SUPER FETuring30721650/18158GB GDDR615.5Gbps
NVIDIA RTX 2070 SUPER FETuring25601605/17708GB GDDR614Gbps
NVIDIA RTX 2060 SUPERTuring21761470/16508GB GDDR614Gbps
NVIDIA RTX 2060 FETuring19041365/1686GB GDDR614Gbps
ZOTAC Gaming GTX 1660Turing14081530/17856GB GDDR58Gbps
ZOTAC GTX 1650 SUPERTuring12801530/17254GB GDDR612Gbps
NVIDIA GTX 1080 FEPascal
25601607/17338GB GDDR5X10Gbps
NVIDIA GTX 1070 FEPascal
19201506/16838GB GDDR58Gbps
NVIDIA GTX 1060 FE 6GBPascal
1280
1506/17086GB GDDR58Gbps
AMD Radeon RX 5700XTNavi 1025601605/1755/19058GB GDDR614Gbps
AMD Radeon RX 5700Navi 1023041465/1625/17258GB GDDR614Gbps
Sapphire RX 5600 XTNavi 1023041130/1660/17506GB GDDR614Gbps
AMD RX Vega 64 Vega 1040961247/15468GB HBM2945Mbps
AMD RX Vega 56Vega 1035841156/14718GB HBM2800Mbs
Sapphire RX 5500 XT 4GBNavi 1414081737/18454GB GDDR614Gbps
MSI RX 580 Armor 8GB Polaris 20230413668GB GDDR58Gbps
Sapphire Nitro+ RX 570 4GBPolaris 20204813404GB GDDR57Gbps

Drivers Used

Drivers 
Radeon Settings 20.8.1
GeForce451.67

Preset Scaling At 4K

Testing presets at 4K gives us a couple of quick metrics before diving into the game too deeply.  First off, it shows us how the game looks at various presets as well as how performance scales with those settings. Horizon Zero Dawn scales quite well in visuals with the Favoring Performance setting still looking decent enough to not feel like you're missing out on much, but the detail draw distance on the Ultimate Quality is worth the sacrifice if you can muster over the 45 FPS mark, yes 60+ is preferred but the pacing of the game, and especially if you're using a controller, is quite comfortably played.

But, clawing back performance is not that difficult, moving to Original quality can gain you nearly 50% performance increase but I would wager you might be better off using the in-game slider and moving to around the 80% resolution mark and applying either the GeForce Sharpening filter or the Radeon Image Sharpening feature to gain back some of that sharpness. This game could have certainly benefited from either DLSS 2.0 or FidelityFX, both of which could be ported into the game with FidelityFX likely being the easier implementation. We've seen both at work in the engine with Death Stranding and it would be wonderful to see it here as well.

Favor Performance
Original
Favor Quality
Ultimate Quality

UHD 4K Preset Scaling
AVG FPS
1% Percentile
0
20
40
60
80
100
120
0
20
40
60
80
100
120
Favor Perf
85
70
Original
75
61
Favor Quality
62
52
Ultimate Quality
56
46
HD 1080p Preset Scaling
AVG FPS
1% Percentile
0
20
40
60
80
100
120
0
20
40
60
80
100
120
Favor Perf
73
53
Original
69
53
Favor Quality
58
46
Ultimate Quality
57
45

Intel Core Scaling Performance

While this test won’t tell just how many cores and threads the game can and will use, it does show how the game performs as you move up in cores and threads available. These were tested at the 1080p settings that we tested the rest of the results while pairing the CPU with the RTX 2080 Ti Founders Edition.  While this does not take into account the cache difference you would see with Intel CPUs as you move through their offering stack it does give us a better idea of how the game benefits and behaves from more cores and threads.

Scaling in Horizon Zero Dawn isn't that exciting to look at when viewed at 1080p, it shows performance hitting its side at six cores, but maintaining decent at quad-core arrangements. If you're rocking a dual-core with quad threads on this one you're likely to see issues, but without multi-threading, you'll be left out in the cold on a dual-core. The game will launch on a dual-core, at least the menus as the game never loaded into a playable scene. The most interesting note is that on six and eight-core arrangements the inclusion of multi-threading doesn't seem to help and at the high end, actually hurts performance slightly. One thing is clear, this game is exceptionally GPU heavy and is relatively unaffected at even 1080p with the RTX 2080Ti when reducing core and thread count.

Core i9-9900K Core and Thread Scaling
AVG FPS
1% Percentile
0
40
80
120
160
200
240
0
40
80
120
160
200
240
8c/16t
126
85
8c/8t
130
87
6c/12t
132
87
6c/6t
132
88
4c/8t
125
78
4c/4t
122
78
2c/4t
91
57
2c/2t
0
0

Graphics Card Results

1080p

Standard 1080p Performance
AVG FPS
1% Percentile
0
40
80
120
160
200
240
0
40
80
120
160
200
240
RTX 2080ti FE
126
85
RTX 2080 SUPER
107
71
RX 5700 XT
102
74
RTX 2070 SUPER
97
66
RX 5700
93
69
RX 5600 XT
87
67
RTX 2060 SUPER
86
59
RTX 2060 FE
78
55
GTX 1080 FE
77
46
GTX 1070 FE
66
38
RX Vega 64
65
49
RX Vega 56
59
45
GTX 1660
59
44
RX 5500 XT 4GB
57
45
GTX 1650 SUPER
54
40
RX 580 8GB
52
39
GTX 1060 6GB FE
51
30
RX 570 4GB
48
36

Ultrawide 1080p

Ultrawide 1080p Performance
AVG FPS
1% Percentile
0
40
80
120
160
200
240
0
40
80
120
160
200
240
RTX 2080ti FE
106
74
RTX 2080 SUPER
90
62
RX 5700 XT
83
61
RTX 2070 SUPER
81
55
RX 5700
76
57
RX 5600 XT
72
56
RTX 2060 SUPER
71
50
RTX 2060 FE
64
45
GTX 1080 FE
62
35
GTX 1070 FE
53
34
RX Vega 64
52
40
GTX 1660
49
39
RX Vega 56
47
35
GTX 1650 SUPER
42
34
RX 580 8GB
42
32
RX 5500 XT 4GB
42
31
GTX 1060 6GB FE
41
29
RX 570 4GB
39
30

1440p

Standard 1440p Performance
AVG FPS
1% Percentile
0
20
40
60
80
100
120
0
20
40
60
80
100
120
RTX 2080ti FE
95
67
RTX 2080 SUPER
79
53
RTX 2070 SUPER
73
50
RX 5700 XT
72
53
RX 5700
64
51
RX 5600 XT
62
49
RTX 2060 SUPER
61
43
RTX 2060 FE
55
38
GTX 1080 FE
55
32
GTX 1070 FE
46
27
RX Vega 64
46
36
RX Vega 56
43
32
GTX 1660
42
32
RX 5500 XT 4GB
39
33
RX 580 8GB
37
28
GTX 1650 SUPER
36
27
GTX 1060 6GB FE
35
21
RX 570 4GB
34
26

Ultrawide 1440p

Ultrawide 1440p Performance
AVG FPS
1% Percentile
0
20
40
60
80
100
120
0
20
40
60
80
100
120
RTX 2080ti FE
79
58
RTX 2080 SUPER
65
46
RX 5700 XT
62
41
RTX 2070 SUPER
57
42
RX 5700
53
43
RTX 2060 SUPER
49
37
RTX 2060 FE
45
35
GTX 1080 FE
44
28
RX Vega 64
37
28
GTX 1070 FE
37
24
RX Vega 56
35
26
GTX 1660
34
28
GTX 1650 SUPER
30
22
RX 580 8GB
29
21
GTX 1060 6GB FE
28
18
RX 5600 XT
0
0
RX 5500 XT 4GB
0
0
RX 570 4GB
0
0

UHD 4K

UHD 4K Performance
AVG FPS
1% Percentile
0
10
20
30
40
50
60
0
10
20
30
40
50
60
RTX 2080ti FE
56
46
RTX 2080 SUPER
47
37
RX 5700 XT
39
31
RTX 2070 SUPER
41
33
RX 5700
36
29
RTX 2060 SUPER
35
28
RTX 2060 FE
32
24
GTX 1080 FE
30
18
RX Vega 64
26
19
RX Vega 56
25
18
GTX 1070 FE
25
14
GTX 1660
23
18
GTX 1650 SUPER
22
16
RX 580 8GB
20
15
GTX 1060 6GB FE
18
12
RX 5600 XT
0
0
RX 5500 XT 4GB
0
0
RX 570 4GB
0
0

Conclusion

Overall the PC performance of Horizon Zero Dawn would have been pretty shoddy if we had gone off the original build of the game so we're glad they've cleaned things up in time for the launch. This gives us hope that issues like the nonworking Anisotropic Filtering adjustment will be fixed, but this isn't giving the game a pass for that being missed. The other oddity you'll find pretty apparent here is going to be the 30 fps facial animations, that really cause cutscenes to feel odd, especially when, at times, the eyes are moving at full speed.

BUT, performance is not quite as good as I hoped for and really shows that something like DLSS 2.0 or FidelityFX would make an excellent addition here to claw back some performance on some cards at higher resolutions. Ultrawide resolutions are unusually difficult to run at levels they do in other games, there seems to be a bigger hit than usual. Neither GPU vendor has any major issues with the launch of the game, so any problems to be uncovered will likely be with the game itself and could, hopefully, be patched in the future

The Navi cards did spectacular here delivering better than anticipated performance numbers when putting up against their direct competitors. Showing greater than a 10% performance advantage in with the recent GPUs. Their Vega lineup and Polaris do okay against the Pascal counterparts. One of the more interesting points is how utilizing Hardware Accelerated GPU Scheduling really pushes the GeForce GTX 1650 to parity with the RX 5500XT and kisses the GTX 1660 despite it being the more powerful card with more VRAM.

Something of note in the results in where cards register a Zero number. What I found was a few cards in the Radeon lineup seem to fall victim to assets not loading. I recorded them as Zero because it's essentially unplayable because all that will load in are characters, the skybox, and land formations. This is an unfortunate effect that only seems to affect the Radeon cards, the GeForce cards handle the game perfectly fine across the board. Something to consider is it is only happening at resolutions above and beyond what those cards are designed for.

Before the patch there were stability issues abound, sometimes the game would stutter at a strange interval no matter what was going on, and often would simply not launch properly. I'm glad I waited until after the patch dropped to start recording data and found the game to be in a much more attractive state. I can't defend the odd 30 FPS facial animations but it is a game you'll probably want to cap at around 60 FPS anyway to smooth the fluctuations out. The issue with the Anisotropic Filtering bug should be alleviated by the time the game goes live. But, for now, I'll be back in and exploring the world with Aloy.

Deal of the Day

Comments