News Horizon Zero Dawn Performance: The PC Port Struggles on Previous-Gen GPUs

artk2219

Distinguished
Jun 30, 2010
442
30
18,890
29
The R9 390 performance is definitely an optimization problem. It should be scoring around the RX 570, or at the very least the GTX 970 which should also be around the RX 570, thats a strange one.
 

JarredWaltonGPU

Senior GPU Editor
Editor
Feb 21, 2020
612
509
1,270
1
The R9 390 performance is definitely an optimization problem. It should be scoring around the RX 570, or at the very least the GTX 970 which should also be around the RX 570, thats a strange one.
I've seen quite a few newer games that appear to push enough geometry that the difference between the RX 570 (Polaris) and R9 390 (Hawaii) tilts in favor of the 570. That's my best guess as to a cause that isn't purely optimizations for the older GPU architectures: more geometry and other newer features that don't run as well on older architectures. But it's definitely not 'fully optimized' for a broad range of hardware.
 
Reactions: Makaveli

nitrium

Distinguished
Jul 27, 2009
1,472
17
19,665
195
Game runs near flawlessly on my Ryzen 3700X/RTX 2060 system at Ultra settings. Often the visuals are pretty stunning, despite this being a relatively old game. Occasionally the sound cuts out for some reason and then returns a few seconds later, but other than that it's been very stable with no crashes so far (38 hours in).
The game itself is vast, as these open world titles tend to be. Gameplay is a sort of cross between recent Far Cry and Assassin's Creed with Monster Hunter World thrown in for good measure. Lots of optional busy work, but most of the mission-based side quests have self-contained stories like The Witcher 3.
If you're not already burned out on open world games, this is a very good example of the genre and worth a look.
 

JarredWaltonGPU

Senior GPU Editor
Editor
Feb 21, 2020
612
509
1,270
1
Ouch, not even 60 FPS for my GTX 980?
Or a 980 Ti! All three 900-series cards clumped together, which is why I noted that optimizations for older GPUs seem to be severely lacking.

I actually tested the GTX 970 first, and performance seemed not too far off what I'd expect. Then I tested 980 Ti a bit later, and the performance was relatively awful. I looked at the 970 scores and the 980 Ti scores and figured my 970 results must have been wrong -- after triple-checking via additional test runs and driver reinstalls that the 980 Ti numbers were correct. But when I retested 970 it was still accurate, and then I tried the 980 for good measure and it landed right between the two.

Whatever's holding back the GTX 900 series, it seems to be a pretty hard cap on performance for Maxwell architecture. I'd test a 960/950 as well if I had them, just because now I'm curious, but I don't have those cards. :'(
 
Looks like we have found the Crysis to RDNA2/Ampere/Xe-HP...
Eh, I wouldn't really compare it to that. Going by the numbers here, you can get a mostly 60FPS experience from a sub-$150 graphics card at 1080p if you are willing to turn settings down to medium, or at high settings using a $200-$230 card, while ultra settings don't actually look much better. 1440p is of course a bit more demanding, but this article is only showing ultra settings for that resolution. Since dropping to high appears to add around 25% to frame rates without having much impact on visuals, a sub-$300 card like a 5600XT should be able to handle 1440p at 60FPS just fine.

This is a port of a game designed to run on nearly seven year old console hardware, not something designed to showcase what a game engine can do on future hardware, as Crysis was at the time of its release. There's no cutting-edge effects like raytracing to be found that would push current graphics hardware to its limits while providing some meaningful payoff to visuals. High settings appear to mostly just improve shadows somewhat over the original console version, while ultra settings improves them slightly more at a disproportionately large hit to performance. And while people typically expect higher frame rates on PC than the 30FPS that the game runs at on the PS4, being a third-person single player game it's not likely to benefit as much from super-high frame rates in excess of 60FPS.

There do seem to be some potential optimization issues here though, with many cards from prior generations performing inexplicably poorly, but we might see updates to address those in the future. And even if it is perhaps a bit of a shoddy port as far as optimization is concerned, it's good to see some of these PS4 exclusives making their way to PC.
 
Reactions: JarredWaltonGPU

LokkenJP

Honorable
Jun 17, 2014
18
1
10,525
4
Those numbers are waaay off, at least for Pascal cards. I don't know if you used the built in benchmark in the game (which is hardly representative of the actual performance), or a very simple wilderness run.

But I own a Asus GTX 1070Ti (which should sit between the 1070 and the 1080) updated with the latest nVidia driver, and with a mix of Original and Favor Quality (Med-High) my rig struggles to hold 60 FPS during gameplay at 1080p, with frequent drops to 30-40 even in mildly populated areas.

Meanwhile I can run any other contemporary game (The Division 2 or Ghost Recon Breakpoint for example) at much higher framerate with Ultra settings and minimum tweaks (80-90 FPS if I left them uncapped, permanent 60FPS without dips if I lock the framerate)
 
Aug 20, 2020
1
0
10
0
Super disappointed with the performance in this game, especially on the CPU side. Running a 7700HQ and a 1070 max-q gets me 30-45 fps, with frequent dips down to 20. Seems to be CPU bound, as processor usage hits 100% nearly the entire time the game is running.
 

Axell

Distinguished
Jan 14, 2007
8
0
18,510
0
I play on an i7 4770K @ 4.3GHz and a stock XFX RX 480 with 16GB of 3200GHz DDR3 and got pretty solid 50-60 FPS High on this rig, on par with the RX 580 on the graphs.
Pretty sad some people get such low FPS on much newer and beefier systems. :(
 

JarredWaltonGPU

Senior GPU Editor
Editor
Feb 21, 2020
612
509
1,270
1
Those numbers are waaay off, at least for Pascal cards. I don't know if you used the built in benchmark in the game (which is hardly representative of the actual performance), or a very simple wilderness run.

But I own a Asus GTX 1070Ti (which should sit between the 1070 and the 1080) updated with the latest nVidia driver, and with a mix of Original and Favor Quality (Med-High) my rig struggles to hold 60 FPS during gameplay at 1080p, with frequent drops to 30-40 even in mildly populated areas.

Meanwhile I can run any other contemporary game (The Division 2 or Ghost Recon Breakpoint for example) at much higher framerate with Ultra settings and minimum tweaks (80-90 FPS if I left them uncapped, permanent 60FPS without dips if I lock the framerate)
So here's the thing: you have to provide full system specs or comparisons basically become meaningless. This is why we detail the full specs of the test PCs. My test rigs had 32GB of RAM, running at DDR4-3600 speeds. I'm using a 2TB M.2 NVMe SSD. I've got Core i9-9900K (and the other CPUs, but those weren't tested with the GTX 1070 or 1080). I also am running a 'clean' OS. Do you have other apps or utilities open? Did you do a clean driver install? All of these things matter if you want to get consistent results that others could replicate.

Then you have to provide actual benchmark results at the same settings used for our tests. All you say is that your PC "struggles to hold 60 FPS during gameplay at 1080p, with frequent drops to 30-40..." According to my tests, the GTX 1070 averaged 74 fps, but with 99 percentile results of just 50 fps -- at 1080p high. If you have a slower PC on the CPU, RAM, or storage than what I used, that would be absolutely consistent with dips into the 30-40 fps range.

If you want to say the built-in benchmark results are meaningless, run the benchmark on your PC and provide a specific result. What does the benchmark give you for average and minimum fps? (You'll need to use a utility like OCAT to log results.) Then do a "wilderness run" and also log frametimes. Then send precise details on where you're testing and how you tested and someone else could corroborate the results.

From what I saw in testing, the benchmark may not fully represent all areas of the game, but it's repeatable and consistent. That's mostly why I used it. It's a baseline measurement -- there are areas that will run worse than the benchmark, others will run better, but it's a starting point. Also, the camera flyby of the city in the benchmark means lots of fast scene transitions, leading to generally lower minimum fps results.

But because people love to try and throw shade, let me provide some specific results. You have the earlier benchmark numbers for the GTX 1070, but here's a full set of data, using 1080p Favor Quality (high), using the built-in benchmark vs. me running around the wilderness (near the beginning of the game):

1070 @ 1080p HighAvgAvg Min50th55th60th65th70th75th80th85th90th95th97th98th99th99.9thMin
Built-in Run 1
75.7​
50.3​
75.8​
75.1​
74.3​
73.5​
72.6​
71.6​
70.3​
68.6​
65.8​
56.9​
54.1​
52​
49.5​
45.7​
29.5​
Built-in Run 2
75.2​
50.3​
75.3​
74.5​
73.7​
72.9​
72​
71​
69.8​
67.9​
65​
56.3​
53.7​
51.8​
49.6​
46​
27.1​
Manual Run 1
67.3​
59.5​
67.2​
66.7​
66.2​
65.8​
65.3​
64.9​
64.4​
63.8​
63.1​
62.1​
61.4​
60.8​
59.9​
55.7​
33.1​
Manual Run 2
66.5​
58.6​
66.5​
66.2​
65.8​
65.4​
65​
64.5​
64​
63.4​
62.7​
61.5​
60.7​
59.9​
58.7​
54.1​
42.1​

So, the benchmark in this case is 13% faster than where I tested. That's a measurable difference, but it's not particularly meaningful. More importantly, look at the minimum FPS results. The benchmark is 14% slower than the manual test, mostly because of the fast scene transitions. Perhaps most importantly, the benchmark cuts the amount of effort to generate these articles down by 66%. I can start the test and go to another PC and start the test (in offline mode, naturally), and then go to a third PC and start the test. Which is how I was able to run a huge suite of test cards in a relatively short time.

Anyway, these performance analysis articles are provided largely as a service to readers. You now have precise measurements on how the game runs. I ran every benchmark at least twice, discarding the first result as it's usually a bit higher thanks to GPU boost clocks. I restarted the game between each settings change. If there was a major discrepancy between runs, or if numbers looked wrong, I'd verify the result with additional tests.

With a Core i9-9900K and a GTX 1070 FE, plus 32GB RAM and a fast 2TB SSD, running on a clean OS, I got average performance above 60 fps at 1080p high, and a bit of tweaking should get minimums above 60 as well. If you're getting substantially worse performance on similar specs, it's possibly software or drivers. More likely, it's software, drivers, and/or differences in hardware.
 
Reactions: Phaaze88

JarredWaltonGPU

Senior GPU Editor
Editor
Feb 21, 2020
612
509
1,270
1
I play on an i7 4770K @ 4.3GHz and a stock XFX RX 480 with 16GB of 3200GHz DDR3 and got pretty solid 50-60 FPS High on this rig, on par with the RX 580 on the graphs.
Pretty sad some people get such low FPS on much newer and beefier systems. :(
Yeah, that's why we do these tests -- so people can see what they should get in terms of performance. I strongly suspect a lot of the variation in performance people complain about is drivers and software, mixed with slower components elsewhere.
 

Phaaze88

Glorious
Ambassador
Nice to see that the 1080Ti is still putting up a good fight!

Pretty sad some people get such low FPS on much newer and beefier systems. :(
Like, "How do I PC?"
Some people put them together, believing that's all they need to do.
Nope. Driver/bios updates are also their responsibility, as well as software - and whether they actually need them or not. The fewer software apps, the better.
 
Reactions: JarredWaltonGPU

LokkenJP

Honorable
Jun 17, 2014
18
1
10,525
4
So here's the thing: you have to provide full system specs or comparisons basically become meaningless.
With a Core i9-9900K and a GTX 1070 FE, plus 32GB RAM and a fast 2TB SSD, running on a clean OS, I got average performance above 60 fps at 1080p high, and a bit of tweaking should get minimums above 60 as well. If you're getting substantially worse performance on similar specs, it's probably software or drivers. More likely, it's software and drivers and differences in hardware.
You must admit that those system specs are fairy unbalanced. I mean, of course I understand why you use the same rig for benchmarking all the cards, but in the real world, you must agree that not many people will get a top end system with a i9-9900k with a mid-range card like the 1070 Ti...

Having said that, my full specs are Intel i7-4790k running at stock clocks, 16Gb DDR3-1600, Asus Strix Advanced Binned 1070Ti GTX running at stock clocks too. OS is Win10 v2004, on a dedicated 512 Gb SSD, plus the game running on another 1TB SSD (both on SATA6, not NVMe). Drivers are latest nVidia 452.06 installed from scratch after using DDU. No background programs running except Steam client itself. XBox W10 gamebar disabled, and no software for streaming/voip running.

My game options are based on Favor Quality, minus Reflections medium, Clouds medium, Motion Blur disabled, and Ambient Occlusion Medium. Resolution is 1080p. Frame rate is of course uncapped and v-sync disabled.

I've run a manual path from Mother's Watch to Mother's Heart, just running and following the main road. Frame times are captured with PresentMon v1.5.2. This has been repeated 4 times in a row, all four times starting from the same save file, and the first result was discarded.

3-run Average Framerate: 59 FPS (Frame time 16.87ms)
3-run Average Lower 10% Frame time: 21.7ms (equivalent to 46 FPS)
3-run Average Lower 1% Frame time: 31.24ms (equivalent to 32 FPS)

And before blaming other software or bad drivers, let me post the benchmark run of The Division 2 captured exactly on this same setup, same methodology except using the built in benchmark option (settings are a mix of High/Ultra minus Volumetric Clouds at medium):

3-run Average Framerate: 92 FPS (Frame time 10.85ms)
3-run Average Lower 10% Frame time: 12.63ms (equivalent to 79 FPS)
3-run Average Lower 1% Frame time: 14.56ms (equivalent to 68 FPS)

HZD port have some serious performance issues, be it due to bugs or lack of optimization, but the performance is not as expected from a AAA 3 years old game. And your results, at least for a nVidia 1070 Ti, im afraid they are not very representative. Maybe it's the Core i9. Maybe the Chipset/Motherboard/Memory. Maybe the SSD. Maybe a bug in the game that impacts some configurations more than others. Or maybe something else entirely...
 
Last edited:

JarredWaltonGPU

Senior GPU Editor
Editor
Feb 21, 2020
612
509
1,270
1
You must admit that those system specs are fairy unbalanced. I mean, of course I understand why you use the same rig for benchmarking all the cards, but in the real world, you must agree that not many people will get a top end system with a i9-9900k with a mid-range card like the 1070 Ti...
Yes, it's "unbalanced," but I have to isolate the hardware variables. If I test i9-9900K only with top-end new GPUs, then i7-8700K with high-end GPUs, i7-6700K with mid-range GPUs, and i7-4770K with budget GPUs, you don't know if the GPU or the CPU is the true bottleneck. It's why I do separate CPU testing as well. The number of permutations obviously prohibits doing every GPU with every CPU.
Having said that, my full specs are Intel i7-4790k running at stock clocks, 16Gb DDR3-1600, Asus Strix Advanced Binned 1070Ti GTX running at stock clocks too. OS is Win10 v2004, on a dedicated 512 Gb SSD, plus the game running on another 1TB SSD (both on SATA6, not NVMe). Drivers are latest nVidia 452.06 installed from scratch after using DDU. No background programs running except Steam client itself. XBox W10 gamebar disabled, and no software for streaming/voip running.

My game options are based on Favor Quality, minus Reflections medium, Clouds medium, Motion Blur disabled, and Ambient Occlusion Medium. Resolution is 1080p. Frame rate is of course uncapped and v-sync disabled.

I've run a manual path from Mother's Watch to Mother's Heart, just running and following the main road. Frame times are captured with PresentMon v1.5.2. This has been repeated 4 times in a row, all four times starting from the same save file, and the first result was discarded.

3-run Average Framerate: 59 FPS (Frame time 16.87ms)
3-run Average Lower 10% Frame time: 21.7ms (equivalent to 46 FPS)
3-run Average Lower 1% Frame time: 31.24ms (equivalent to 32 FPS)

And before blaming other software or bad drivers, let me post the benchmark run of The Division 2 captured exactly on this same setup, same methodology except using the built in benchmark option (settings are a mix of High/Ultra minus Volumetric Clouds at medium):

3-run Average Framerate: 92 FPS (Frame time 10.85ms)
3-run Average Lower 10% Frame time: 12.63ms (equivalent to 79 FPS)
3-run Average Lower 1% Frame time: 14.56ms (equivalent to 68 FPS)
The Division 2 performance doesn't matter at all -- it's not Horizon Zero Dawn. It's a different engine, different developers, different game, etc. That’s just getting sidetracked.
HZD port have some serious performance issues, be it due to bugs or lack of optimization, but the performance is not as expected from a AAA 3 years old game. And your results, at least for a nVidia 1070 Ti, im afraid they are not very representative. Maybe it's the Core i9. Maybe the Chipset/Motherboard/Memory. Maybe the SSD. Maybe a bug in the game that impacts some configurations more than others. Or maybe something else entirely...
FWIW, I ran tests with OCAT and the 1070 Ti, only this time I'm using the built-in benchmark using your "high/medium" settings (which perform just slightly better than the high preset) vs. what I hope is approximately the same path you used for your manual benchmark. This is with the second patch and 452.06 drivers, and the results are weird:

1070 Ti + 9900K @ 1080p High-MediumAvgAvg Min50th55th60th65th70th75th80th85th90th95th97th98th99th99.9thMin
Built-in Run 2
69.9​
43​
71​
69.8​
68.6​
67.5​
66.3​
65​
63.1​
60.6​
57.5​
51.8​
48.2​
45.6​
42.4​
35.8​
27.5​
Manual Run 2
62.5​
51.3​
62.5​
61.9​
61.4​
60.8​
60.2​
59.4​
58.6​
57.7​
56.7​
55.3​
54.4​
53.7​
52.3​
40.4​
28.9​

The benchmark runs about 12% faster than the manual test for average fps, and 19% slower for 99th percentile fps. Compared to your numbers, my test PC is only about 6% faster, but minimum fps is about 25% better (90th percentile) and as much as 50% better (99th percentile) -- assuming I'm doing the same calculations as you for percentiles, which I'm probably not.

The weird part is that I ran a full suite of tests with the 1070 Ti, and it performed worse than the 1070. So either the patch and/or drivers degraded performance, or I messed something up ... or the 1070 Ti specifically is underperforming in this game. I‘m going to Do some retesting tomorrow and figure out which.
 

LokkenJP

Honorable
Jun 17, 2014
18
1
10,525
4
Hi. Thanks a lot for the reply and for taking the time for testing this again on a 1070Ti card.

The Division 2 performance doesn't matter at all -- it's not Horizon Zero Dawn. It's a different engine, different developers, different game, etc. That’s just getting sidetracked.
I posted Division 2 numbers just because several people in this thread blamed firmwares, drivers or background software for the poor performance of my machine.
I'm not trying to compare directly TD2 with HZD performance, but only showing that there is nothing wrong with the current configuration of my rig by running another contemporary DX12 title and getting the expected performace.

(Of course it's moot trying to compare those two games, being on different game engines, yet it puts the Decima engine to shame, or at least, it's usage in the PC port implementation done by Virtuos, the outsourcing company used by Guerrilla for HZD).


The benchmark runs about 12% faster than the manual test for average fps, and 19% slower for 99th percentile fps. Compared to your numbers, my test PC is only about 6% faster, but minimum fps is about 25% better (90th percentile) and as much as 50% better (99th percentile) -- assuming I'm doing the same calculations as you for percentiles, which I'm probably not.
Those numbers are much closer to what I've experienced with my 1070Ti card and in line with what I expected from your setup. I understand the difference in the lower percentiles could be due to a couple of reasons:
  • The i9-9900k itself is vastly more powerful than my current i7-4790k, thus it could be avoiding some CPU bottlenecks that could be causing stuttering on my setup.
  • Seems you are calculating lower percentiles as average framerates (ie, seems you take the 2*nth slowest frames or something like that and then average them, it's the only way that the 50% percentile can be so close or the same value as the overall average FPS). I'm instead taking the nth percentile Lower Frame Time (that is, for each run I'm taking directly the single Frame Time value which holds the nth percentile position, thus n percent of the frames are rendered 'slower' than this number ).
The weird part is that I ran a full suite of tests with the 1070 Ti, and it performed worse than the 1070. So either the patch and/or drivers degraded performance, or I messed something up ... or the 1070 Ti specifically is underperforming in this game. I‘m going to Do some retesting tomorrow and figure out which.
I'm very interested in how this develops. Please, share your results in this thread.

I don't believe that the 1070Ti (being basically a 1080 GPU core with the 1070 memory subsystem) should be performing worse than vanilla 1070, so maybe all the debug code added by the developers in the 1.01 and 1.02 patches to track the multiple crashing reports are to blame; or maybe the nVidia 452.06 drivers have some weird interaction with HZD, while 451.67 don't have such issue.
 
Last edited:

atomicWAR

Splendid
Ambassador
Between the last patch and the latest nvidia driver most of the teething issues my set-up was having with this game are now fixed. But prior to that the game was released in a horrid state. I would get huge GPU and CPU usage drops which lead to really bad frame rates. You could tell the initial build of the game couldn't properly feed the hardware. Now that issues are fixed I was hoping a RTX 2080Ti would be enough to play the game maxed out at 4K while never dropping below 60fps but hey maybe a that is a few driver release/game patches away. Though things are getting close. I personally see dips as low as 50fps in game but another 10fps over the next few months may indeed be possible, who knows. Regardless the game has come far in a short time frame. I hope other users have a similar experience because at release I couldn't that pile of garbage.
 

JarredWaltonGPU

Senior GPU Editor
Editor
Feb 21, 2020
612
509
1,270
1
Hi. Thanks a lot for the reply and for taking the time for testing this again on a 1070Ti card.



I posted Division 2 numbers just because several people in this thread blamed firmwares, drivers or background software for the poor performance of my machine.
I'm not trying to compare directly TD2 with HZD performance, but only showing that there is nothing wrong with the current configuration of my rig by running another contemporary DX12 title and getting the expected performace.

(Of course it's moot trying to compare those two games, being on different game engines, yet it puts the Decima engine to shame, or at least, it's usage in the PC port implementation done by Virtuos, the outsourcing company used by Guerrilla for HZD).



Those numbers are much closer to what I've experienced with my 1070Ti card and in line with what I expected from your setup. I understand the difference in the lower percentiles could be due to a couple of reasons:
  • The i9-9900k itself is vastly more powerful than my current i7-4790k, thus it could be avoiding some CPU bottlenecks that could be causing stuttering on my setup.
  • Seems you are calculating lower percentiles as average framerates (ie, seems you take the 2*nth slowest frames or something like that and then average them, it's the only way that the 50% percentile can be so close or the same value as the overall average FPS). I'm instead taking the nth percentile Lower Frame Time (that is, for each run I'm taking directly the single Frame Time value which holds the nth percentile position, thus n percent of the frames are rendered 'slower' than this number ).
I'm very interested in how this develops. Please, share your results in this thread.

I don't believe that the 1070Ti (being basically a 1080 GPU core with the 1070 memory subsystem) should be performing worse than vanilla 1070, so maybe all the debug code added by the developers in the 1.01 and 1.02 patches to track the multiple crashing reports are to blame; or maybe the nVidia 452.06 drivers have some weird interaction with HZD, while 451.67 don't have such issue.
Looks like I screwed up when I installed the latest Nvidia drivers. I didn't go into the Nvidia Control Panel and turn off G-Sync and force Vsync to off, and that dropped performance. That or I had just run some other application without rebooting that was interfering, but I suspect it was just G-Sync. Anyway, just retested the built-in bench as well as the run from Mother's Watch to Mother's Heart:

1070 Ti + 9900K @ 1080p High-MediumAvgAvg Min50th55th60th65th70th75th80th85th90th95th97th98th99th99.9thMin
Built-in Run 2
77.9​
48.7​
78.8​
77.2​
75.6​
74.1​
72.8​
71.4​
69.8​
67.7​
64.6​
59.5​
56.1​
53.5​
49.6​
40.6​
8.4​
Manual Run 2
64.3​
54​
63.7​
63.1​
62.6​
62.2​
61.7​
61.2​
60.7​
60​
59.1​
57.8​
56.9​
56.1​
54.7​
46.9​
27.3​

There's a much bigger gap between the built-in and manual results now, which felt odd. Then I retested at the "favor performance" (low) settings and 1080p, with both the 451.67 and 452.06 drivers. I started with the latter, but the results were 'wrong' so I did a full driver clean and installed the earlier drivers. Then, I put the GTX 1070 FE back into the system as well. Check this out:

9900K @ 1080p LowAvgAvg Min50th55th60th65th70th75th80th85th90th95th97th98th99th99.9thMin
1070 Ti 452.06
94.5​
67.9​
95​
93.4​
91.8​
89.8​
87.7​
85.4​
83.5​
81.6​
79.4​
75.5​
72.6​
70.4​
66.9​
60.5​
36.7​
1070 Ti 451.67
93.6​
67.4​
94​
92.3​
90.6​
88.6​
86.5​
84.4​
82.6​
80.7​
78.5​
74.6​
71.7​
69.5​
66.2​
61.6​
50.1​
1070 452.06
104.6​
70.1​
105.5​
104.2​
102.8​
101.3​
99.8​
98.2​
96.4​
94.1​
90.4​
80.8​
75.3​
72.3​
69.1​
63.4​
51.9​

Yeah, that's not good. Either my 1070 Ti is somehow fubar (pretty sure it's not), or there's a specific issue with this GPU on Horizon Zero Dawn. Given the poor performance of a few other GPUs, I'm leaning toward that. But hey, let's try the high-medium settings you listed on the 1070 as well.

9900K @ 1080p High-MedAvgAvg Min50th55th60th65th70th75th80th85th90th95th97th98th99th99.9thMin
GTX 1070 452.06 High-Med Built-In
89.2​
60.1​
89.5​
88.4​
87.4​
86.3​
85.2​
84​
82.5​
80.4​
76.5​
69.2​
66.3​
64.1​
60.4​
49.4​
19.1​
GTX 1070 452.06 High-Med Manual
71.6​
62.4​
71.6​
71.1​
70.7​
70.2​
69.7​
69.1​
68.5​
67.8​
67​
65.7​
64.8​
64.1​
62.7​
55.1​
36.3​
GTX 1070 Ti High-Med Built-In
77.9​
48.7​
78.8​
77.2​
75.6​
74.1​
72.8​
71.4​
69.8​
67.7​
64.6​
59.5​
56.1​
53.5​
49.6​
40.6​
8.4​
GTX 1070 Ti High-Med Manual
64.3​
54​
63.7​
63.1​
62.6​
62.2​
61.7​
61.2​
60.7​
60​
59.1​
57.8​
56.9​
56.1​
54.7​
46.9​
27.3​

So, there you have it: GTX 1070 Ti as far as I can tell is not running 'properly'. The built-in benchmark gives the 1070 FE a 14.5% lead over the 1070 Ti FE, and the manual benchmark (which is slightly more variable) gives it an 11.4% lead. Minimum fps (second column) favors the 1070 FE by 23.4% in the built-in bench, and 15.6% in the manual run.
 

LokkenJP

Honorable
Jun 17, 2014
18
1
10,525
4
Thank you for sharing those results.

I'm a bit disappointed tbh, because issues with drivers or game patches might be eventually fixed, but I'm pretty sure noone at either Virtuos/Guerrilla or nVidia will bother looking for performance issues on one single particular card model :(
 
Reactions: JarredWaltonGPU

JarredWaltonGPU

Senior GPU Editor
Editor
Feb 21, 2020
612
509
1,270
1
Thank you for sharing those results.

I'm a bit disappointed tbh, because issues with drivers or game patches might be eventually fixed, but I'm pretty sure noone at either Virtuos/Guerrilla or nVidia will bother looking for performance issues on one single particular card model :(
I forwarded the information (this forum thread) to my Nvidia drivers contact, so at least they'll know about it. Not sure if the fix is on Nvidia's end or internal to the game. I sort of suspect it's the game -- like, it might have different DX12 code paths for Maxwell, Pascal, Turing, GCN, and Navi as an example, and then a generic mode if a GPU doesn't match any of the known GPUs. That's not a great way of coding IMO, but who knows what's going on exactly? Anyway, let's hope someone fixes it (in the next patch or driver).
 
Reactions: LokkenJP

ASK THE COMMUNITY