In the demo at 1440p it was 4 polys per pixel. 20 million polys per frame. You dont see them all in the screen.
-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
-
We have made minor adjustments to how the search bar works on ResetEra. You can read about the changes here.
Tim Sweeney on UE5: "The Nanite and Lumen tech powering it will be fully supported on both PS5 and Xbox Series X and will be awesome on both."
- Thread starter P40L0
- Start date
People said that they did not mention resolution on that laptop and if it's running 40fps at 1080p with (massivly) reduced detail, than that's not surprising at all, seems like engine is indeed very scalable.
I know I heard 20 million per frame number, but I don't think all of 20M was in camera frame. I think they were saying XX billion polys were tesselated into 20 million in the scene, not in any particular camera view. You can see in the video in the OP that some of the triangles are big enough to take up dozens or even hundreds of pixels when he zooms in on the original demo video.
I think some are forgetting that Epic's CTO Kim Libreri actually already told PC Gamer that the RTX 2070 Super (basically like a 2080 laptop GPU) could run the demo with 'pretty good' performance. So it's nothing new, really.
So it was mentioned in the PS5 demo video that they're boiling the triangle soup down to 20m drawn triangles per frame - that is, ~10 tris per pixel at 1440p.
The engineer here apparently say that in terms of I/O requirements, you could run this at 1080p/2 tris per pixel and not need PS5 style high I/O. 1080p at 2 tris per pixel would be 2m tris - 10% of the above. If, per his suggestion, I/O and triangle ingestion (and ultimately drawn tris) are linked, and scale proportionately, then if single digit GB/s are needed for PS5 fidelity, then maybe MB/s are needed for his example.
I would guess that the triangle data is laid out in a hierarchy in storage/memory, and lower data access speeds simply means terminating at coarser levels of the hierarchy. And this is why the system can turn a fidelity knob on data access speed/throughout.
It would be amazing if they could just actually demonstrate that though. Like.... show a PC version with a hard drive or SATA SSD struggling to keep up, or dropping detail in comparison.
That would prove what they are claiming much more concisely.
Of course... that's not going to happen any time soon... because of course if it doesn't scale that noticeably... then there goes the marketing angle.
Yeah, the premise of this thread is pretty dumb.
Next time someone will mention that the demo is running on some android smartphone without specifying the measures taken to do so and people will use it as a consola wars argument again.
1080p = 2.073.600 pixels x 2 polygons each = 4.147.200 polygons for the scene.
4.147.200/20.000.000=20,74%
20,74% from 9 GB/s IO (suppossing is needed for 1440p) = 1,87 GB/s is what you would need from your IO instead of 9 if all would scalate directly.
In the laptop case specifically I wouldn't bet it was reducing to the extreme he specified elsewhere. Rough guess but 1080p/2 tris might be sata ssd levels (e.g. MB/s) of throughout. The laptop mentioned should do better than that.
You can see when the commenter zooms in that some of the triangles are much bigger than a pixel. I think 20M is for the scene, not for the camera frame.
Also, even if it was 20M per camera frame, 20M tris/3.6M pixels = 5.5 tris per pixel. 5.5 vs 1 isn't enough to go from 9 GBps down to less than 1 GBps.
Also, we are not accounting for texture data in this equation, which is prolly much bigger than geometry data.
1. Having common sense
2. The engineers working on the project and not the talk head executives.
Trying to disprove the comments of the people working on these things because they're "in a different country" it's just plain stupid.
Your math is right, mine was wrong! Someone mentioned elsewhere '1 or 2 tris per pixel' and I had a calculation based on 1 tri per pixel stuck in my head. For 2 it's 20% - not MB/s per my example, but more modestly GB/s perhaps. Which I think still fits the engineer's point about running it on more modest IO requirements. I'm sure it probably also would continue to scale down to lower numbers of triangles per pixel - or higher numbers of pixels per triangle! - with decreasing data access rates and/or decreasing triangle draw rates. There may be no lower bound on how far it scales, at least above whatever fixed overheads are required.
But what are we really talking about?. We knew the engine would run in a smartphone. How is not going to run with a RTX 2070 super with a Samsung EVO!.
From the ex technical art director maybe some texture streaming artifact,. If it is the case SSD are never fast enough and this is the fastest one on the market ;)
Could be. He already said UE5 is running on highend PCs, both next gen consoles, current gen consoles and Android as well, and jokingly said not on C64. It's clear that he makes some fun of the "war" and that the key thing is that it will scale really well on all kinds of hardware.
Exactly how well it will scale will obviously be shown before it releases, devs need to know what they're dealing with before choosing what engine to use.
Anyway, it'll take years until we get to play anything on UE5 so the console war makes little sense at this point, it's not like any of us will choose console based on how a UE5 game will look or run, that choice will be made within a few months if it hasn't already been made.
The debate is how much scaling would be involved, meaning whether the PS5 demo we saw maxxes out PS5's 9GBps IO or is much less than that for what was shown.
But this is early tech. I mean we can not make any conclusion now when the engine is coming out next year and the demo was running on old PS5 devkit (afaik).
This engine reveal has been something else
Console exclusive multiplatform engine reveal fuck me what a world we live in.
Console exclusive multiplatform engine reveal fuck me what a world we live in.
Sincerely being all rock textures I doubt this demo is representative of the variation and taxing it could reach in a real game. They vould be reusing the same textures without streaming them over and over. The flying sequence vs a flying sequence in Spiderman has almost no variety in assets.
Yes, there are some triangles bigger than a pixel. Most are 1px. Then there's overdraw. They said 20m drawn per frame. With overdraw etc you can be drawing many more triangles per frame than pixels, a multiple of triangles per pixel, and still have some bigger triangles.
At half the res (1080p) it would.
Side note though: yeah, at 1 tri per pixel, if that's how things scale. At 2 you'd be in higher, if still reasonably modest, ssd speed rather than MB/s perhaps. But according to the engineer it does also scale down to that, so I'm guessing nanite can keep going with larger and larger triangles, to a point anyway.
I imagine this system scales texture steaming/access with granularity of geometry being streamed/accessed. Maybe not 1:1 smoothly, so maybe there might at different times be different proportions of texture to geom streaming going on. Tex access needs to be scalable too for the system overall to be scalable.
It's been said that it doesn't max it out. Which is why it's believed to be easy to replicate this demo on XBSX. AFAIK...
Logic would dictate as much, but some here still trying to argue that demo as shown was only viable with PS5's 9GBps throughput. Seems silly to me too but whatever...
Here comes the money...
They could have used the platform to reveal the engine but be more clear about the goals of the engine. They could not even answer simple question from Geoff about XSX in the live stream interview itself.
We don't know if it does or does not. Devs on the stream said that PS5 I/O architecture is the big reason of the fidelity in the shown demo and also mentioned that it can scale across different hardware. This Chinese stream confirms that it can scale to different hardware, nothing more.
They're there to sell the engine and PS5. Of course they'll big up the architecture.
Yeah, their weird reticence when asked about the XSX made it pretty obvious that they were there as partners, to talk up each others' thing. Absolutely nothing wrong with that, but also means we shouldn't read too much into any superlatives.
Ugh, just show me games already. I can't believe we've been reduced to microanalyzing the verbiage used in a media spot to show off a multiplatform game engine that won't even be in wide use for at least two years.
Well they are using the advantages of said architecture and showcasing it in the demo.
People are assuming that the engine should scale down to lesser hardware perfectly fine, but it should scale up with better hardware as well and as devs are saying, I/O performance is never "good enough".
Someone said it earlier in the thread (based on the Chinese q&a), could be false, like most of this thread. If I see it again I'll link it in this post...
Edit:
Maybe this poster can shed light, I can't find anything concrete...
Last edited:
That's a good spot. It could alternatively be a triangle draw rate limit, but given that the scene seems to remain otherwise unchanged when the detail is resolved, I think that's less likely than it being bound by data at that point. Whatever temporal artifacts you might have seen otherwise are hidden by the shadow passing over the scene, which was a lucky coincidence.
Ergo... if this was a data bound, 'even' 5.5GB/s tied to 16GB of RAM was not enough for this to be entirely flawless ;)
He can't be serious, can he?
A 4K/HDR MP4 video can run smoothly at 60fps on a 3 years old smartphone.
They obviously weren't talking about video performance, they weren't talking about the PS5 Vimeo video running/playing on a super high end laptop, and he perfectly knows that.
The Epic China Engine Lead himself was talking about the actual demo performance running on his laptop in editor mode, not some random anonymous 4chan bullposter.
Tim Librery (Epic Games' CTO) himself confirmed the demo could run great on a RTX 2070 Super with a regular, already on the market, NVMe SSD.
Is he actually trolling on us?
Is he blatantly doing damage control after evident previous Sony PR bullshit statements?
I don't know, but for sure these reactions are even more cringey than the Aaron Greenberg ones...
So, if im reading this correctly then the comments Tim Sweeney has been making talking up PS5 is all PR talk with no substance, but a PR comment about UE5 being awesome on all platforms is not PR and taken at face value as meaning no platform will have an advantage? Makes sense...
What is cringe is your picking and choosing what to add to the OP having it trying to portray a very clear certain narrative and ignore basically everything else contradicting the message you so obviously want to be true.
The engine is SCALABLE for the 100th time and it seems according to VERY experienced developers even the PS5 running this demo have streaming issues meaning you could benefit from even faster SSDs while playing a game like this.
You also have a breakdown of a engine programmer calling out how much work the SSD have to do to pull it off.
But let's ignore all that ones again and only take note and update the OP of the one translation you like disregarding when more of it is translated and that don't fit that narrative of yours as well.
Cringe I tell you
I love how everyone assumes the 2080 they are talking about is a Max Q, those only go into thin laptops. There are chonky bois out there with non-Max Q 2080s in them. Actually if you're doing any kind of engine testing or benchmarking on the road to show off your awesome new game engine you would likely use one of those.
All of this would have been avoided if Epic had clarified anything about the demo instead of making some vague comments about the magic of Sony's SSD and it 'running super cool, awesome dude!' on other platforms. It is as though they're not able to talk about performance on other platforms for some reason... ;)
Do you have any examples of these "Sony PR bullshit statements"?
I was already updating the first post before answering to yours...
The translation was already confirmed by native Chinese user inside this very topic (which also saw the entire interview from start to finish before doing so, and not jumping the stream here and there without context).
I'm not bitter, I'm really really surprised to read a so incredibly embarrassing reply and narrative directly from the CEO of one of the biggest and most important company of the entire gaming industry.
Yeah there will obviously be a limit to how much virtual texture you can cache at once, including the geometry displacement texture.
I haven't checked but If that's the first frame of the scene it's probably a streaming thing, I don't expect that kind of thing to completely go away even with super fast SSDs.
I'm assuming the tweet is incorrect in thinking those are normal maps, when actual it's just lower res geometry mip. The difference isn't going to be easy to spot unless you look at the horizons.
I'm not sure what you are reading that is that bad. He has already said it runs across a wide variety of devices so not sure hes doing what you think he is. I mean has he said look at this ps5 demo that can not scale to any other platform at all despite already saying it can basically.
He deliberately dodged the 1440p/40+ part right now in the lab on the gaming laptop with future 60fps target on both next-gen consoles, the CPU-bound processing for things to keep up, the actual SSD requirements and directly tried to undermine/debunk the entire dev discussion stating they were only talking about the PS5 demo (when its video was playing) instead.
They were not.
Oh, they also shoot down the video upload, just in case.
Last edited:
Tim Sweeney on Twitter
“@6h4n3m @FALS3GODS @ogbrunette @XcloudTimdog Not sure exactly what was said. The translated quote on 40 fps isn't a comparison between hardware capabilities. To hit a 100% solid 30 fps with vsync in the demo, per-frame times vary from 30 fps up to much higher.”
twitter.com
They never said it was running at 1440p, just 40fps on specific location, by multiple translations.
Yeah if it ran better than a PS5 I would not imagine it to be maxq Version, rather the larger one that has Performance more in the 2070-2070s Range.
An RTX 2060 should realistically not be beating the ps5 GPU.