Unreal Engine 5, [UE5 Developer Availability 2022-04-05]

Monitoring VRAM usage on a 3090/10900K/3.5 GB/s Nvme - 5 second load time
Right before entering portal: 4339 mb
In portal load (white screeen): 3386 mb
Right after portal: 4604 mb

You can see it dump VRAM and reload it in real time and count up even. But of course, not anywhere using all of the VRAM possible on the 3090. It could probably load much much quicker if it had dual allocation there for the dark world cached.
What is the RAM usage though?
 
Monitoring VRAM usage on a 3090/10900K/3.5 GB/s Nvme - 5 second load time
Right before entering portal: 4339 mb
In portal load (white screeen): 3386 mb
Right after portal: 4604 mb

You can see it dump VRAM and reload it in real time and count up even. But of course, not anywhere using all of the VRAM possible on the 3090. It could probably load much much quicker if it had dual allocation there for the dark world cached.

Cheers Alex, but not the question I was asking.

Have you seen the magical laptop video this whole argument is based on?
 
System Ram Usage (per process):
Right before entering portal: 3954 mb
In portal load (white screeen): 2783 mb
Right after portal: 3132 mb

Wow, that sounds shockingly low.

You'd think there could be an engine process that caches the next areas to RAM if the disc speed is insufficient. Presumably that would require two concurrent solutions for loading on PC and local system spec dependent.

I'd hazard a guess that cross platform games likely won't make brilliant use of huge RAM pools, since the next gen consoles have shifted focus away from traditional loading techniques.

Would be interesting to see a full PC game designed around large RAM pools and fast SSD. Could probably fully eliminate visible loading altogether. Would require some smart caching and decompression between SSD, RAM, and VRAM.
 
This again..

Where is last year's demo called "Lumen in the land of Nanite" running on a laptop with a 2080 max-q?
Perhaps you're mistaking two very different demos.

Depends on what you're doing though, right? Assuming you're travelling around a lot from area to area, then you'd need either a lot of substitute RAM, or have a fast SSD.

Travel speed may need to be determined by how fast the SSD is, or you get a degradation in the mesh/texture.

It could be a laptop with e.g. 64GB RAM, but even then having a 6.4TFLOPs Turing GPU showing that same demo @ ISO settings would mean the PS5 demo wasn't properly optimized for the hardware. Which doesn't make sense if you're trying to show off what the engine can do on Epic's side, and what the PS5 can do on Sony's side.

I see these posts are mine, so the comments made during the past 2 pages were apparently targetted at me.


Nowhere do I say in those posts that the UE5 only runs on the PS5. I even mention more platforms that could run the demo.

I did (and do) say that the 2020 demo wouldn't run off a 2020 PC's storage IO, and from Tim Sweeney's statements it does not.
I have plenty of posts out there discussing how a slower IO could be used to show similar results, and having a ton of RAM with pre-decompressed assets to stream to the GPU was definitely one of the ways.

And thanks to your committed digging I see now that the Chinese guy's statement (or translation) is even more preposterous: not only is a 6.5-7 TFLOPs 2080 max-q running the PS5 demo at the same settings, it's even running it 33% faster at 40FPS.


One could always run the 2021 demo (that BTW runs with lower quality assets for higher compatibility) on a 2080 Max-Q and see if it runs at 1440p 40FPS.

The demo doesn’t need 32-64 GB ram memory. This is one of the myths that Brian karis the creator of Nanite was trying to dispel.

Important details:

"There's been a bunch of misconception that the [Lumen in the land of Nanite Demo] doesn't work anymore or only runs on a PS5" - Epic Games Brian Karis (Creator of Nanite)

1) The entire nanite data for the lumen in the land of nanite demo was 6.14 GB.

2) The entire nanite data for the Valley of the Ancient demo was a bit smaller than Land of Nanite demo.

3) However The entire texture data for Valley of the Ancient demo was a bit larger than Land of Nanite demo.

4) The compiled packaged demo of valley of the ancient is in the mid 20s GB compared to the 100 GB project file.

5) Valley of the ancient demo had more assets than Land of Nanite demo.

6) Valley of the ancient compiled version requires only 3 GB system ram and 7 GB VRAM.

This is basic math, logic and reason.
Land of the Nanite demo has similar requirements to Valley of the Ancient.

infact it will be less demanding because if you watched the entire stream Brian karis spent most of the time showing how unoptimized valley of the ancient was compared to Lumen in the land of Nanite.
Because of this Nanite will cost more.
 
Why is it 32Gb minimum, 64Gb recommended RAM then?

In one of the videos brian karis said the assets in new demo were sized so that a machine with 64GB ram can store whole level in ram. This will bypass any DirectStorate/Streaming issues. We can't stream optimally in PC until microsoft releases windows version with directstorage and epic releases UE5 version with directstorage support. To fit in 64GB ram is likely half of the reason for asset downgrade from ue5 reveal(valley of death). Other part is storage space required.

We are looking at technology that is going to take maybe 1-2more years before it's going to ship in any game.
 
lol I have not - but I have a Razer laptop with a Core i7 10875H CPU and RTX 2070 Super (MaxQ?) with 2560 Cuda Cores and a pretty low boost clock (1155 mhz) and 352.06 GB/s Memory Bandwidth right next to me I could try and run this on.

Go ahead, thanks beforehand :D (nice laptop btw), im thinking of getting a zephyrus M16 this summer.

In one of the videos brian karis said the assets in new demo were sized so that a machine with 64GB ram can store whole level in ram. This will bypass any DirectStorate/Streaming issues. We can't stream optimally in PC until microsoft releases windows version with directstorage and epic releases UE5 version with directstorage support. To fit in 64GB ram is likely half of the reason for asset downgrade from ue5 reveal. Other part is storage space required.

We are looking at technology that is going to take maybe 1-2more years before it's going to ship in any game.

As explained above (you missed that?), thats due to the editor like it was for UE4 engine, not to actually play the compiled demos, hence the low ram usages.
 
Here 1080p TSR up to 4K in motion vs. Native 4K (Motion Blur disabled):
movementpyjgk.jpg

Unfortunately as a large JPG because a PNG is just too big for most hosts.
 
If the demo is so small in size, then it's safe to assume there can't be much loading going on - only that one phase shift into the "dark" area.

I was initially wondering if the mesh sizes were changing in complexity along with the loading.


This video shows the demo running on a HDD and fast travelling around. Looks like the same detail throughout the flyby, so I'm thinking the entire area is present in VRAM. So it's fine if you've a HDD paired with a high spec GPU that has lots of VRAM. The screenshots someone was showing with low details must have been a slow HDD paired with a small VRAM pool.

So really we've no clear idea how this will function in an open world game just yet.
 
Last edited by a moderator:
Here 1080p TSR up to 4K in motion vs. Native 4K (Motion Blur disabled):
movementpyjgk.jpg

Unfortunately as a large JPG because a PNG is just too big for most hosts.

Really visible difference on the character, particularly on her arm. Other areas look a pretty close match, like the flag in the middle distance - it looks higher than 1080p to my eyes.
 
Expected break up. Will probably be quite noticeable on high refresh low pixel response displays.

Will be fun to compare to DLSS once this will become available in UE5.

And thats a still image, look at the details on the structure ahead of the character, the char itself and the textures/ground. Its a rather ugly on the left as compared to the native 4k taa one. When things start moving differences should be even larger (i havent tried it yet TSR vs native, ue5 is on my main system and i dont want to swap gpus now).

Edit: thanks Alex for chiming in here and sharing your findings, its quite time consuming, i hope you can use these tests for a future video on this ;)
 
Does anyone else get stuttering and lower framerate while using the drone on turbo mode?

I'm a bit curious about this. It doesn't happen in the editing mode. RAM/VRAM/CPU is not nearly being maxed out and it doesn't use my NVMe either.
 
lol I have not - but I have a Razer laptop with a Core i7 10875H CPU and RTX 2070 Super (MaxQ?) with 2560 Cuda Cores and a pretty low boost clock (1155 mhz) and 352.06 GB/s Memory Bandwidth right next to me I could try and run this on.

Well, we all look forward to seeing the original demo reproduced in its entirety as claimed.

Cheers
 
Back
Top