• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Unreal Engine 5.0 OUT NOW

vpance

Member
is it just the AA or is TSR turned on?

Not sure what it is. Some sort of screen space effect that gets messed when it can't update as fast as the camera moves around your char I guess.

FOV 125

c.png
 

DenchDeckard

Moderated wildly
Wow. Game looks incredible in first person shaky cam mode.



Crazy to think people tried to downplay the games visuals by saying it doesnt look photorealistic.


Yeah, this looks insane. Add a few extra effects and we are more or less photo realistic.
 

PaintTinJr

Member
Flying around at max speed uses about 10mbs from disk, so your ssd is plenty fast.
You are probably correct, but throughput and latency are two different issues. You can have an internet pipe that's massive for data and still get echo on calls and poor video calling because it doesn't provide low latency isochronous data channels as needed, and the same can be true of data streaming, even with 10Mb/s it could be the latency to requests that is the issue in some cases in the future.
 

Hoddi

Member
Flying around at max speed uses about 10mbs from disk, so your ssd is plenty fast.
I did some testing when I upgraded my PS5 SSD and there's a massive difference in disk IO between the console and PC versions of this demo. Near as I can tell, the PC version just reads data once from disk and then keeps most of it in Windows' file cache. In contrast, the PS5 version read some 100GB from disk in the span of just 10 minutes.
 

PaintTinJr

Member
I did some testing when I upgraded my PS5 SSD and there's a massive difference in disk IO between the console and PC versions of this demo. Near as I can tell, the PC version just reads data once from disk and then keeps most of it in Windows' file cache. In contrast, the PS5 version read some 100GB from disk in the span of just 10 minutes.
How did you test that - or is that sarcasm? I guess taking readings of the smart info on a PC, then using in a PS5 to get setup with the demo, and then using in game, and then putting it back in a PC and checking the Smart info again would let you calculate that -even more accurate if done using two drives, one for the actual test, and one for just getting the drive in the PS5 and the demo installed.

If there is such a difference, what are you thinking is the difference? From my own testing with Valley in the pre-release of UE5 the virtual shadow mapping dropped the editor test performance down to 2-8fps from 20-30fps so hearing the VSM uses optimally bound static caching of shadows on consoles might account for a lot of I/O.
 

Xdrive05

Member
Has anyone tested the Matrix demo on a SATA SSD vs an M.2? Apparently my SATA Crucial MX500 can be eclipsed up to 6 or 7 times by a high end PCI-E 3.0 M.2 (which is as fast as my board will support), so I'm wondering if the Matrix is the first game where that would really make a difference.
 

ethomaz

Banned
Is the demo available on PC? where can I download
It is not a demo... it is the source code for UE5 project... you can bulid a executable if you wish.


Some users released a build executable but downloads for your own risk.

Google Drive: https://drive.google.com/file/d/1yQFtLGZZSnK5h4D_aiFg32TqXNIRVwRf/view
MegaUpload: https://mega.nz/folder/axN0RTZI#JUVH0QkwaGZOKCp-rMOGgQ

 
Last edited:

winjer

Gold Member
It doesn’t use TRS every time the resolution is lower than native.

TSR is enabled by default in this demo.
In fact I wasn't even able to turn it off. Don't know if it's locked out just in this packaged demo, or if it's how it's supposed to be in UE5.
 

ToTTenTranz

Banned
It is not a demo... it is the source code for UE5 project... you can bulid a executable if you wish.


Some users released a build executable but downloads for your own risk.

Google Drive: https://drive.google.com/file/d/1yQFtLGZZSnK5h4D_aiFg32TqXNIRVwRf/view
MegaUpload: https://mega.nz/folder/axN0RTZI#JUVH0QkwaGZOKCp-rMOGgQ


Nice!

Can we set the resolution, TSR level and other settings in that executable?
 

Larxia

Member
Try restarting Windows? Might have to do that after the SDK and .NET installs.

winjer winjer do you know if this is what the packaged folder should look like?

MxI15QSa8gJD.png


It complained of some missing dlls when I ran it so I added them, but now running the exe doesn't actually do anything. No message, no process, nothing. Think I'm gonna throw in the towel on this lol.
Okay it's fixed now... I also had to install the SDKs from the visual studio installer. I'm not really sure which one exactly was needed between all the different ones in the visual studio installer and the 2 things you linked me, it's quite confusing, but it's working now.
 

winjer

Gold Member
So I found how to disable TSR.
r.AntiAliasingMethod=
0=disables AA
2= TSR low quality (improve performance by 3-4 fps)
4= TSR high quality (seems to be the default)


I also found a command that gives a nice increase in image quality
This costs around 3 fps, but it looks much better.
r.TSR.History.ScreenPercentage=200 (default 100)
 

Hoddi

Member
How did you test that - or is that sarcasm? I guess taking readings of the smart info on a PC, then using in a PS5 to get setup with the demo, and then using in game, and then putting it back in a PC and checking the Smart info again would let you calculate that -even more accurate if done using two drives, one for the actual test, and one for just getting the drive in the PS5 and the demo installed.

If there is such a difference, what are you thinking is the difference? From my own testing with Valley in the pre-release of UE5 the virtual shadow mapping dropped the editor test performance down to 2-8fps from 20-30fps so hearing the VSM uses optimally bound static caching of shadows on consoles might account for a lot of I/O.
Yes, except that I use an NVMe to USB adapter which spares me from shutting down the PC. Getting a readout only takes a minute this way.

I don't know the cause but it's presumably just reading/evicting data quicker on the consoles in order to save memory. It's been one of the main selling points of the console SSDs and this demo is a showcase of that. The PC version doesn't evict any data as long as there is free memory so the data is read back from the file cache instead of the disk.

So I found how to disable TSR.
r.AntiAliasingMethod=
0=disables AA
2= TSR low quality (improve performance by 3-4 fps)
4= TSR high quality (seems to be the default)
That's a good find. Setting 2 removes the TSR pass from Nsight and replaces it with a TAA pass.
 

winjer

Gold Member
That's a good find. Setting 2 removes the TSR pass from Nsight and replaces it with a TAA pass.

That explains why I saw no difference, aside from a mild performance boost. I was running native, and having no TSR means nothing.

Still trying to find what these do, regarding TSR.

r.TSR.History.UpdateQuality
r.TSR.RejectionAntiAliasingQuality
r.TSR.ShadingRejection.HalfRes
r.TSR.ShadingRejection.SpatialFilter
r.TSR.Translucency.EnableResponiveAA
r.TSR.Translucency.HighlightLuminance
r.TSR.Translucency.SeparateTemporalAccumulation
r.TSR.Velocity.Extrapolation
r.TSR.Velocity.HoleFill
r.TSR.Velocity.HoleFill.MaxScatterVelocity
r.TSR.Velocity.WeightClampingPixelSpeed
 

PaintTinJr

Member
Yes, except that I use an NVMe to USB adapter which spares me from shutting down the PC. Getting a readout only takes a minute this way.

I don't know the cause but it's presumably just reading/evicting data quicker on the consoles in order to save memory. It's been one of the main selling points of the console SSDs and this demo is a showcase of that. The PC version doesn't evict any data as long as there is free memory so the data is read back from the file cache instead of the disk.


That's a good find. Setting 2 removes the TSR pass from Nsight and replaces it with a TAA pass.
In the beyond3D link a page or two back, the comment on that website said the major difference in settings between PC and consoles was VSM, and following the UE5 docs about using static caches as the poster advised - for anyone that didn't know - VSM static caching is a means of telling the VSM that certain world areas are filled with static geometry, so unless they have a dynamic shadow casting object interact in their tightly bounded region, that part of the shadow cascade doesn't need regenerated - and presumably with the I/O of the PS5/XsX, storing those large 4 or 8k, 32(?)bit shadow maps in RAM - instead of on disk and streaming in on demand - isn't needed.

The lack of static cache VSM on PC would also explain why the heavy hitting GPUs(RTX 3080, RX 6700XT and better) aren't getting a big performance boost in frame-rate over the console versions with normal tweaks IMHO.
 

CamHostage

Member
How low can you go?



I'm personally curious of the low-end experimentation as UE5 projects go on, now that the engine is past its early access phase.

We had already seen Valley of the Ancients running on hardware in the vicinity of PS4 Pro (albeit there's not an exact equivalency, and that still wouldn't mean that Nanite/Lumen would eventually be included in UE5's platform support... technically, PS4/XO can use Nanite, but Epic does not recommend it for a shippable game.) So what would a build of Matrix Awakens City Sample look like and run like in a PS4 configuration? If you fell back from Lumen and/or Nanite to traditional methods (although I don't know if procedural functions like the modular building methods of this demo work without reliance on Nanite?), what would be left of City Sample on Switch specs or an Android device?

This video is just the barest of modifications: change the final preference parameters and run the existing build. It could be interesting in the coming weeks what'll be done with this project file on both the top end and bottom end.
 

01011001

Banned
I downloaded one of the precompiled versions cuz I couldn't be arsed to do it myself quite frankly.

these are the results without changing any settings yet (I upped everything density wise to 100% tho) on a Ryzen 5600X and RTX3060ti at 1440p (TSR is on it seems... many artifacts during motion, THIS NEEDS DLSS! FAST!)

citysample64-bitdevelb1ko1.png



now lets start digging into settings :D
is it known which settings the PS5/SX versions used btw? would be interesting to compare directly

edit: uhm, is there no config file in the precompiled version people uploaded?
 
Last edited:

SlimySnake

Flashless at the Golden Globes
Saw a video of a 6600 (8.9 tflops) running the game at 1080p around 30 fps. Granted the pedestrian density and traffic density is set to 50% compared to 100% in the console versions but still a very interesting benchmark.

 

hlm666

Member
Some interesting reading in here nice work all. I havn't had a chance to play with this yet (hopefully on the weekend between motogp sessions). The UE dev that's been linked from here was asked if direct storage would help and he seems to have profiled it to see whats causing the stuttering and he didn't seem to think any of it was IO related. He pretty much blamed the shader compiles and the drivers doing conservative things and expects it to get better now they have forced dx12.
 

sertopico

Member
I'm mind blown from this project. I have forced a 1440p res and increased the graphical settings by modifying the INI file (btw, many thanks to some of the users for their findings!). Getting 25-30 fps on my 3080+10700k but it looks f*ckin amazing. GPU usage goes from 90 to 100% all the time. This is with no doubt the biggest leap in graphical fidelity so far. Crazy.

citysample_2022_04_088jkl8.png


citysample_2022_04_089ek3n.png


citysample_2022_04_08gbj8a.png


citysample_2022_04_0873knw.png
 

winjer

Gold Member
Saw a video of a 6600 (8.9 tflops) running the game at 1080p around 30 fps. Granted the pedestrian density and traffic density is set to 50% compared to 100% in the console versions but still a very interesting benchmark.

Not a bad showing for that 6600. But make me wonder how much that card is bottlenecked by memory bandwidth.
With only a 128 bit bus and just 32MB of L3 cache, it's probably hurting performance, even at 1080p.
 

Larxia

Member
So I just tried it on my GTX 1080 and it isn't so bad lol goes between 20 and 30 fps, I'd say around 24 fps average. It does stutter quite a bit when moving quickly around, but I was expecting like single digit framerate maybe :messenger_grinning_smiling:
Running at 1080p of course.

nHK8uP1.png

LTYM5l4.png
 

SlimySnake

Flashless at the Golden Globes
So I just tried it on my GTX 1080 and it isn't so bad lol goes between 20 and 30 fps, I'd say around 24 fps average. It does stutter quite a bit when moving quickly around, but I was expecting like single digit framerate maybe :messenger_grinning_smiling:
Running at 1080p of course.

nHK8uP1.png

LTYM5l4.png
It's pretty impressive how they were able to get ray traced reflections running on non RTX or RDNA 2.0 cards.
 

SlimySnake

Flashless at the Golden Globes
I'm mind blown from this project. I have forced a 1440p res and increased the graphical settings by modifying the INI file (btw, many thanks to some of the users for their findings!). Getting 25-30 fps on my 3080+10700k but it looks f*ckin amazing. GPU usage goes from 90 to 100% all the time. This is with no doubt the biggest leap in graphical fidelity so far. Crazy.
Love to see a game tax a crazy powerful next GPU like the 3080 get taxed like this. All of a sudden the PS5 and XSX running this game at 1080p 25-30 fps doesnt sound too bad.

Were you able to increase the traffic and pedestrian density to a full 100% to match the console versions? The 3080 offering 75% more performance/pixels compared to consoles makes sense, but all the benchmarks ive seen have it capped to 50%. The number of cars and pedestrians in the console version is seriously impressive.

 
Last edited:

yamaci17

Member
higher amounts of pedestrians makes little to no effect on GPU performance in this demo
 
Last edited:

ethomaz

Banned
It's pretty impressive how they were able to get ray traced reflections running on non RTX or RDNA 2.0 cards.
Is it really Ray-tracing reflections? Because it doesn't look like it.
BTW Lumen is not made for Ray-tracing exactly... it works without it.

Edit - I read the documentation... Ray-tracing is only used in Lumen in supported RT hardware.


The pic your quoted is not using hardware RT.
They call they use something called "Software Ray-tracing" but I'm not sure if it is RT indeed.
And there is no Software Ray-tracing for Reflections.

Lumen
Use Hardware Ray Tracing when availableUses Hardware Ray Tracing for Lumen features when supported by the video card, RHI, and operating system. Lumen will fall back to Software Ray Tracing otherwise. Hardware Ray Tracing has significant scene update costs for scenes with more than 100,000 instances. See Ray Tracing Performance Guide for information.
Ray Lighting ModeControls how Lumen Reflection rays are lit when Lumen is using Hardware Ray Tracing. By default, Lumen uses Surface Cache for best performance, but can be set to Hit Lighting for Reflections for higher quality.
Software Ray Tracing ModeControls which tracing method Lumen uses when ray tracing the scene. Detail Tracing traces against individual mesh's Distance Fields for the highest quality. Global Tracing traces against the lower detail Global Distance Field for fastest traces.
Hardware Ray Tracing
Support Hardware Ray TracingEnables ray tracing from supported operating systems, RHI, and video cards for higher quality results.
Software Ray Tracing
Generate Mesh Distance FieldsWhether to build distance fields of Static Meshes. This is needed for Software Ray Tracing with Lumen and Distance Field Ambient Occlusion which is used to implement Movable Sky Light shadows and ray-traced distance field shadows on Directional Lights. Enabling this increases build times, memory usage and disk size of Static Meshes.
Distance Field Voxel DensityDetermines how the default scale of a mesh converts into Distance Field Voxel dimensions. Changing this causes all distance fields to be rebuilt. Large values consume memory very quickly.
 
Last edited:

Larxia

Member
Is it really Ray-tracing reflections? Because it doesn't look like it.
BTW Lumen is not made for Ray-tracing exactly... it works without it.
I don't really know what it is, looks like screen space reflection, although a bit better because things on the edge of the screen don't dissapear as easily as usual, like buildings, but here for example it's pretty obvious:


But it's not on everything, we can see some objects still being reflected.

Then there's also low res stuff like this lol

6sDDXSW.jpg
 
Last edited:
Love to see a game tax a crazy powerful next GPU like the 3080 get taxed like this. All of a sudden the PS5 and XSX running this game at 1080p 25-30 fps doesnt sound too bad.

Were you able to increase the traffic and pedestrian density to a full 100% to match the console versions? The 3080 offering 75% more performance/pixels compared to consoles makes sense, but all the benchmarks ive seen have it capped to 50%. The number of cars and pedestrians in the console version is seriously impressive.


My gpu doesn't seem to work any harder with cars and people on or off. Something is working harder though. Just not my gpu.
 

SlimySnake

Flashless at the Golden Globes
higher amounts of pedestrians makes little to no effect on GPU performance in this demo
It made a huge effect on consoles. Reducing traffic and pedestrian density by 50% pretty much made it a locked 30 fps experience.

Even in this thread, people pointed out that reducing it to 0% improved framerate on PC as well.
 

winjer

Gold Member
It made a huge effect on consoles. Reducing traffic and pedestrian density by 50% pretty much made it a locked 30 fps experience.

Even in this thread, people pointed out that reducing it to 0% improved framerate on PC as well.

Probably it's CPU heavy on consoles as well as on PC.
 

ethomaz

Banned
I don't really know what it is, looks like screen space reflection, although a bit better because things on the edge of the screen don't dissapear as easily as usual, like buildings, but here for example it's pretty obvious:


But it's not on everything, we can see some objects still being reflected.

Then there's also low res stuff like this lol

6sDDXSW.jpg

The Epic guy said that.

"It's stock UE5 Lumen in the sample - I believe the details here are relatively up to date: https://docs.unrealengine.com/5.0/en-US/lumen-technical-details-in-unreal-engine/. The sample is set to use hardware RT if available by default."

It is the same doc I found before... so in GPUs without RT it is not doing RT or using that Software RT but it is not clear what that Software RT does.
 

01011001

Banned
jesus, this city sample is terribly optimised it seems. I finally had some time to play around with it, and god damn...

I tried the lowest possible settings at 1440p, with what still looks like TSR permanently turned on (if that's not TSR then it has to be the worst TAA ever made), so who knows what the actual internal resolution is... and here are the results

PC:
Ryzen 5600X
GeForce RTX 3060Ti
16GB DDR4 3200mhz

Monitor: Dell, 1440p, 144hz

gameusersettings.ini-4ujkx.png

citysample64-bitdevelijkfn.png

citysample64-bitdevel7ojnw.png

citysample64-bitdevel7gjru.png
citysample64-bitdeveldrkcl.png



as you can see in the image, with these settings the RT reflections are completely gone, in fact there isn't even Screen Space Reflections, and only what seems to be a very low fidelity reflection map present for a reflective sheen effect.
what seems to be the issue here is clearly either Nanite or their NPC system... Nanite seems to absolutely fuck up your GPU, what makes this clear IMO is the final screenshot that shows the city from above, having lots of buildings in view that all use Nanite.

the shadows, that I expect to still use Lumen, could also be an issue, but I feel like Nanite is the main culprit here as shadow fidelity, when raytraced, should scale way more evenly from scene to scene.
Also, scaling down lumen settings like RT reflecctions doesn't increase performance nearly enough to make any sense. in the position you spawn in I go from ~31fps to ~42fps after setting everything from 3 to 1, that's basically nothing if you consider that all reflections are gone with that and AO is completely off with that too.

with the settings above this demo looks worse than basically any modern Open World game, while running roughly 40% as fast on this machine compared to those.

what seems to be the case is that this demo was clearly 100% designed around the consoles and designed around running at 30fps and no more than that.




btw. if you wanna toast your PC, set all the settings up to 4 ;) as the standard seems to be 3, and 4 gives me 5FPS lol, mainly because my VRAM seems to be filled to the brim at that setting


just out of curiosity, I set all the settings back to the standard 3 and put the resolution quality to 25%, which means 360p, and, yeah... wtf?

citysample64-bitdevelfjjna.png


yes, your eyes aren't fooling you! 33 FPS with 48% GPU usage and somehow 81% CPU usage... the fuck is happening?
 
Last edited:
Top Bottom