Wednesday, May 22nd 2019

Windows 10 May 2019 Update 1903 Gaming Performance Tested in 21 Titles, with RTX 2080 Ti and Radeon VII

Microsoft earlier today released to market its latest version of Windows 10, the May 2019 Update (version 1903). There was quite some talk about Microsoft tweaking the kernel to improve CPU performance. Other gamer-relevant changes include updates to WDDM (display driver model), and an updated DirectX 12, which now supports variable-rate shading. A similar technology is available on Vulkan, and has already been implemented in games such as "Wolfenstein: The New Colossus." With much talk about the latest Windows being better for games than the previous Windows 10 Fall 2018 Update (1809), we decided to take it for a spin.

After backing-up our 1809 installation onto a disk image, we updated to 1903 using Windows Update, with the same driver- and game versions as our recently-updated setup (details here). We then put the machine through our entire selection of 21 games, and two high-end graphics cards, the AMD Radeon VII and the NVIDIA GeForce RTX 2080 Ti. We used driver versions 19.5.1 for AMD, and 430.64 for NVIDIA, both of which support Windows 1903. Our tests span 1920x1080 (Full HD), 2560x1440 (1440p), and 3840x2160 pixels (4K) resolutions. We present our data as percentage gain/loss over Windows 1809, with three data-points per game, each representing the three resolutions in the order "Full HD", "1440p", and "4K". The first graph below covers the RTX 2080 Ti, and the second one Radeon VII.

The GeForce RTX 2080 Ti remains largely flat with its performance gains and losses averaged across all games. "Assassins Creed: Odyssey" shows a 2% performance gain in 1080p, and a roughly 2% loss at 4K. The reverse of this phenomenon is noticed with "Far Cry 5," in which 1080p performance is down 2%, and 4K performance up by that much. These look like random variations between test runs. "Battlefield V" gains 1% in all three resolutions. "Darksiders 3" gains 1% at 1440p. F1 2018 loses 1-2% in all three resolutions. Most other games, including "Strange Brigade," remain largely flat. Averaged across all games and resolutions, the net gain for RTX 2080 Ti with the move to Windows 1903 is +0.05 percent, so almost exactly "no change".

The Radeon VII fares infinitesimally better than the RTX 2080 Ti in relative terms. We see gains as high as +3% in "Civilisation VI", a little over 2% with Anno 1800, and almost 2% with "Shadow of the Tomb Raider." Losses are far less pronounced compared to the RTX 2080 Ti, with only "Far Cry 5" posting close to -2% loss, while titles such as "Battlefield V," "Darksiders 3," "Rage 2," "Sekiro," and "The Witcher 3," each post a 1% loss. Averaged across all games and resolutions, the Radeon VII achieved a +0.16 percent performance improvement.

Suffice to say, the switch to Windows 1903, with drivers that support its new WDDM 2.6 driver-model, and an updated Direct3D component, does not tangibly benefit gaming at this time, at least looking at our selection of 21 games and 3 resolutions. This can be interpreted as a good thing, too, since upgrading Windows to the latest version does not come with any gaming performance cost either. Whatever gains or losses you see here can be discounted for random variation. Your decision to [freely] upgrade to Windows 1903 should hence be predicated on the UI improvements the new Windows version brings to the table, and not its under-the-hood performance improvement claims.
Add your own comment

33 Comments on Windows 10 May 2019 Update 1903 Gaming Performance Tested in 21 Titles, with RTX 2080 Ti and Radeon VII

#1
P4-630
btarunr, post: 4052338, member: 43587"
On a side-note, and we checked if Windows 1903 finally packs the magic Retpoline performance improvements for CPU side-channel vulnerability mitigations, as was reported earlier. Turns out it doesn't.
Skylake and later generations of Intel processors are not compatible with Retpoline.
Posted on Reply
#2
Divide Overflow
I wasn't expecting any significant gaming performance improvements with the latest Windows update. Good to know that there are no significant gaming performance penalties either.
Posted on Reply
#3
srsbsns
I had no idea 1903 was supposed to give better AMD performance. I installed it last night from a clean install using an ISO. I played Star Citizen and noticed double digit frame rate increases. So maybe it was just not me. I know Star Citizen is not something you can repeat testing with but something changed there for sure.
Posted on Reply
#4
Vayra86
srsbsns, post: 4052379, member: 177699"
I had no idea 1903 was supposed to give better AMD performance. I installed it last night from a clean install using an ISO. I played Star Citizen and noticed double digit frame rate increases. So maybe it was just not me. I know Star Citizen is not something you can repeat testing with but something changed there for sure.
Yeah, correct. The game changes all the time, and gets optimized as well ;)

I have to say the average performance deviation here is hardly worth mentioning IMO.
Posted on Reply
#5
Mistral
If I might suggest, if you are keen on including all 3 resolutions, use 3 separate lines for them instead of a single one. The current graph is very awkward to make sense of.
Posted on Reply
#6
W1zzard
Mistral, post: 4052404, member: 49446"
If I might suggest, if you are keen on including all 3 resolutions, use 3 separate lines for them instead of a single one. The current graph is very awkward to make sense of.


Looks super busy and scary for less experienced readers? Not sure if better or worse
Posted on Reply
#7
Ferrum Master
W1zzard, post: 4052420, member: 1"
Looks super busy and scary for less experienced readers? Not sure if better or worse
Looks better for me.

Actually it would make more sense to more grand experiment. Compare LTSB vs PRO.
Posted on Reply
#8
BrainCruser
Why the fuck are they updating the driver model every 6 months? It is like they want every update to result in bluescreens.
Posted on Reply
#9
Ferrum Master
How you can ad low level features without doing fundamental code changes? Leave everything as is? Screw progress???
Posted on Reply
#10
Easo
Does it ACTUALLY result in bluescreens? 1903 early reports are about some other, minor stuff, not graphic driver BSOD's. I also fail to remember it happening previously on 1809, 1803, etc.
Maybe do not scream without a reason?
Posted on Reply
#11
Ferrum Master
Easo, post: 4052433, member: 70901"
Does it ACTUALLY result in bluescreens? 1903 early reports are about some other, minor stuff, not graphic driver BSOD's. I also fail to remember it happening previously on 1809, 1803, etc.
Maybe do not scream without a reason?
There were some GSODs during early beta's. Nvidia driver had a huge memory leak. And that's it. But that's testing... and fine...
Posted on Reply
#12
Metroid
The good thing is, it wont be any major performance loss, also remember that amd and nvidia needs to updates their drivers to reflect the changes on this updated windows version.
Posted on Reply
#13
MKRonin
The graphs are hard to read, maybe bar graphs would be better?
Posted on Reply
#14
arh2o
Thanks for the article and benchmarks! This type of gaming benchmark is what makes TPU shine, as no other site is doing this type of in-depth benchmarking on 1903's positives/negatives.
Posted on Reply
#15
Metroid
arh2o, post: 4052491, member: 180904"
Thanks for the article and benchmarks! This type of gaming benchmark is what makes TPU shine, as no other site is doing this type of in-depth benchmarking on 1903's positives/negatives.
So true, what it really matters to everyday use, few do. 12 years ago, I did a huge benchmark test on windows xp3 sp3 x windows xp sp2 on many benchmarks and in the end I concluded the sp3 was a huge step forward in performance. Tests like this is very good as it tell us if upgrading will be good or not.
Posted on Reply
#16
Mistral
W1zzard, post: 4052420, member: 1"


Looks super busy and scary for less experienced readers? Not sure if better or worse
This is better, much more informative. If I were to go full UX on it I'd say remove "series 2" from the label, and switch one of the lines for m-dashes and the other for points, as green and red look the same to colour-blind users when both are just a thick line. But I'm getting carried away...

Seriously though, much appreciated.
Posted on Reply
#17
Prima.Vera
Why did you guys test with a card that only 0.001% of gamers have? You should have test with more popular cards like the ones from AMD or cheaper last gen cards from nVidia, which are still extremely popular.
Posted on Reply
#18
metalfiber
All risk and really no reward. Gaming stats stay within the margin of error...I'll pass on the so called "updated upgrade".
Posted on Reply
#20
xtreemchaos
anyone with a 7700k tryed the update yet ?, i like to look before i leap.
Posted on Reply
#21
nickbaldwin86
I hope that this is run again once the new drivers are released for 1903.

One would assume that NV and AMD can build drivers that take advantage of the updates and add performance.
Posted on Reply
#22
dozenfury
You would think Microsoft would take the time to benchmark these updates before putting them out as a gaming performance upgrade. Seems like more of a case where they are declaring a victory for something they haven't achieved. Maybe these are changes that will benefit more in the future. If I were on the team that Microsoft assigned the project to improve CPU performance through kernel rewrites etc., and all of the work culminated in only an average one-twentieth of 1% improvement before being released to the public (and in many scenarios decreased performance 2-3%), I wouldn't feel too comfortable in my job.
Posted on Reply
#23
Gasaraki
The GeForce RTX 2080 Ti remains largely flat with its performance gains and losses averaged across all games. "Assassins Creed: Odyssey" shows a 2% performance gain in 1080p, and a roughly 2% loss at 4K. ... "Far Cry 5," in which 1080p performance is down 2%, and 4K performance up by that much. These look like random variations between test runs. "Battlefield V" gains 1% in all three resolutions. "Darksiders 3" gains 1% at 1440p. F1 2018 loses 1-2% in all three resolutions. Most other games, including "Strange Brigade," remain largely flat. Averaged across all games and resolutions, the net gain for RTX 2080 Ti with the move to Windows 1903 is +0.05 percent, so almost exactly "no change".

....Losses are far less pronounced compared to the RTX 2080 Ti, with only "Far Cry 5" posting close to -2% loss, while titles such as "Battlefield V," "Darksiders 3," "Rage 2," "Sekiro," and "The Witcher 3," each post a 1% loss. Averaged across all games and resolutions, the Radeon VII achieved a +0.16 percent performance improvement.



I love the bias in the wording. Even though the 2080Ti only drops 2% also in FarCry 5 and gained 1% in Darksiders 3 for example, with the Radeon 7 dropping 1%, the Radeon 7 losses are "FAR LESS PRONOUNCED compared to the 2080Ti".

People talked about the graph. Yes it's awful because the points don't even line up with the games at the bottom... wtf. What are the points in the line graph that are in between the games? Why are there more points than games?
Posted on Reply
#24
W1zzard
Gasaraki, post: 4053355, member: 168493"
What are the points in the line graph that are in between the games? Why are there more points than games?
As mentioned in the text, each game is tested at three resolutions, 1080p, 1440p, 4K. That's the three points for each game, in that order
Posted on Reply
#25
InVasMani
Seems like it's all just PLL clock skew margin of error for now at least, but that's fine it's not hindering performance at least and in the future the new variable shading stuff should be able to help performance.
Posted on Reply
Add your own comment