Current Generation Games Analysis Technical Discussion [2024] [XBSX|S, PS5, PC]

Good talks folks! Lots of good questions with no real answers yet.

I have to agree with Shifty and some others, chasing film from gaming world is a moving target. They are completely limited by the architecture and bandwidth of the hardware GPUs. Film is more reliant on the CPU and it's easy of use.
 
Are there any details on render times for an animation produced on these GPU accelerated renderers?
Yeah we have a few examples.

The CineBench score of the 14900K is 1795, the GPU score of the 4080S is 27790! The 4080S is 15x faster than 14900K.
In Blender, the 14900K scores 442, the 4080S scores 9012, the 4080S is 20x faster than 14900K.


In another set of examples, we can compare accelerated ray tracing on vs off.

In Octane, the 4090 is 2.2x faster with it on.
In Blender, the 3090 is 3x faster than 6900XT due to better ray tracing acceleration.

 
It's not so much raw performance when it comes to CPU Vs GPU.

As stated already, it's about flexibility and memory access.

The 4080s may very well be 20x faster in Blender when working within in its 16GB VRAM limit, but give it something that's 64GB to work on and that 20x advantage will come crashing down.

A few pages ago it was explained that GPU's could crash when working on scenes that wouldn't fit in their VRAM where as the CPU's have no issues. So showing benchmark numbers of tests that operate within the VRAM limits of a GPU isn't showing the whole truth.

In some cases it's flexibility >>> raw performance.
 
Yeah we have a few examples.

The CineBench score of the 14900K is 1795, the GPU score of the 4080S is 27790! The 4080S is 15x faster than 14900K.
That's a synthetic benchmark but as we've just learned, that doesn't necessarily translate to real-world gains. I've seen GPU acceleration in Blender myself, which is why I imagined render times for movies had come right down, so it's a surprise it hasn't for reasons I hadn't appreciated, particularly the RAM one. But that's a certain class of rendering. Below that there'll be animated TV series and advertisements and student films and indie shorts that aren't 500 GB a frame. These are the real-world examples we can look to to see what's being rendered and how and in what timeframes on GPUs, if there's any info on them out there.

So I'm really hoping we can find a Netflix series or animated short somewhere with render times. Maybe student projects? ¯\_(ツ)_/¯
 
Who or what is threat interactive and why do they think they can fix UE?
Yea we had a couple of their posts on here earlier. Just some guys who seem to be following the misinformation format to get funding for their own fork of UE which will likely never be delivered.
 
This guy is such a clown, with a side dish of "give money to my startup to make graphics good again". The scam couldn't be clearer, but as we have seen all over the world, most people are stupid. There is no other way to say it.
I don't know about 'most'. It's not quantified, but people always give too much emphasis to the vocal minority, basing estimates on amount of evidence.

The main issue is reach. History shows anyone can make a cult as there are people who'll effectively be brainwashable. You just need to reach enough. If only 1 in 1000 people will listen, and you reach 100,000 people in a city, that's 100 followers giving you their life savings. When there's an internet of hundreds of millions, a tiny fraction can still be lucrative. That's not to say everyone looking for backing to help make the world a better place is a con artist - they might be legitimate in aspirations but just clueless as to what it'll actually take, or even if their problem is a problem (that can be solved) - but there are no checks or controls. This is pure, unrestricted freedom. Everything that happens is on the adults to make intelligent, informed choices. And we learn now in a grotesque human experiment that a sizeable number just can't. I argue brains that are perfectly suited to take in knowledge from experience moving around the trees and watching the people around them are completely out of their depth in a world of infinite information; they literally can't process the data effectively and can only accommodate a sub-set of info that's probably arbitrary in what they take on. Hence "I read/watched this on the internet. Now it is my belief and I reject anything else because there's zero capacity in the neurological constructs of my grey matter to fit and and organise anything different."
 
I don't know about 'most'. It's not quantified, but people always give too much emphasis to the vocal minority, basing estimates on amount of evidence.

The main issue is reach. History shows anyone can make a cult as there are people who'll effectively be brainwashable. You just need to reach enough. If only 1 in 1000 people will listen, and you reach 100,000 people in a city, that's 100 followers giving you their life savings. When there's an internet of hundreds of millions, a tiny fraction can still be lucrative. That's not to say everyone looking for backing to help make the world a better place is a con artist - they might be legitimate in aspirations but just clueless as to what it'll actually take, or even if their problem is a problem (that can be solved) - but there are no checks or controls. This is pure, unrestricted freedom. Everything that happens is on the adults to make intelligent, informed choices. And we learn now in a grotesque human experiment that a sizeable number just can't. I argue brains that are perfectly suited to take in knowledge from experience moving around the trees and watching the people around them are completely out of their depth in a world of infinite information; they literally can't process the data effectively and can only accommodate a sub-set of info that's probably arbitrary in what they take on. Hence "I read/watched this on the internet. Now it is my belief and I reject anything else because there's zero capacity in the neurological constructs of my grey matter to fit and and organise anything different."
Very insightful way of looking at it. Honestly, bit surprised this didn't occur to me earlier that this is what was happening. You'd be crazy not to be monetizing social media based on reach alone. You could sell anything really.
 
The issue with Threat Interactive is I've seen tonnes of people who call him a clown, but none of them actually attempt to debate his points.inna grown up way.

I don't agree with everything he says, but there some of his points that I agree with. Especially some of the stuff he's said about modern games having poor IQ due to poor TAA implementations.

Even some of his points about Nanite I can somewhat relate too as I've though similar things myself.
 
The issue with Threat Interactive is I've seen tonnes of people who call him a clown, but none of them actually attempt to debate his points.inna grown up way.

I don't agree with everything he says, but there some of his points that I agree with. Especially some of the stuff he's said about modern games having poor IQ due to poor TAA implementations.

Even some of his points about Nanite I can somewhat relate too as I've though similar things myself.
I think the best way to describe it, a hammer is a hammer. It can make a good piece of furniture and a bad one. It really comes down to how you wield it.

And the latter point is really the challenge devs are having, most of them may not be as strong at actually building a house and these tools are so effective at doing their job in a short time span that quality may not be up to par.
But that’s where the issue lies. These guys are blaming the tool, which is UE in this case. They have just enough know how to understand some of the key issues but not enough know how to realize that developers are not utilizing a tool correctly is a sign that either (a) they are trying to pull a fast one and make bank (b) they really have no experience and don’t know what they’re are talking about.

They have successfully taken advantage of a situation that I do believe epic will need to manage the narrative around, but this isn’t necessarily Unreal at fault here which is what they are trying to say it is.

As a side note, I don’t know what I’m talking about either; but at the very least I can see some developers having major success with UE and some developers not. And because things are not universally poor, and rather larger studios who can afford expensive talent don’t run into these issues, is a sign to me that it’s not the engine at fault here, but there could definitely be more documentation on how to avoid these issues.
 
The issue with Threat Interactive is I've seen tonnes of people who call him a clown, but none of them actually attempt to debate his points.inna grown up way.

I don't agree with everything he says, but there some of his points that I agree with. Especially some of the stuff he's said about modern games having poor IQ due to poor TAA implementations.

Even some of his points about Nanite I can somewhat relate too as I've though similar things myself.

Almost wish Threat Interactive was just outright banned subject on this forum. My main issue is he calls it his development studio, but there's no evidence it actually has employees. Does he have a registered business? He talks about their standard development setup, which is a 3060, but no evidence it's not just him and his personal computer. He says he won't release his "art"/game because he refuses to release it with the damaging visual downsides of UE5 like TAA blur, but has shown basically nothing to prove they're doing anything at all. He offers a pie in the sky "ai" tool that will create LODs on the fly, and is soliciting money for that (and a UE fork), but shows no evidence that "they" are actually working on it. Again, where's the studio, where's the product? Any demos or signs of progress at all? He's soliciting money through youtube donations instead of through a platform that offers some accountability, like an established crowd-funding platform. He can take all of that money and just do whatever he wants with it.

His latest video is basically him whining that people are attacking his character for all this. And yah, I'm am. This guy can go fuck himself until he actually proves out any of it, and most of it is super easy to do. He comes out swinging attacking professionals in the industry and then gets upset when people want accountability from him. He's mentions he was banned from a graphics discord, tries to play it off as the industry being out to get him for telling the truth. The most likely answer is he's a shit disturber and rude, like he is in his videos.

His takes about games pale in comparison to the fact that he's outrage farming and potentially scamming a lot of people.

On the gaming takes, it's all the most basic shit.
TAA sucks. I agree. This is the least controversial sentiment in gaming.
Nanite has an overdraw problem. From what I've read and understood, it does not. The quad overdraw problem does not apply because of the way shaders are launched with a visibility buffer.
Nanite has overhead. Yep. Again, not super controversial, but in the case of rendering very small triangles it's a win over the hardware path. Whether this is good or bad is subjective.
Baked lighting is better in static environments. Yes and no. The environments look good, but the dynamic objects (player characters, physics objects etc) in those environments don't.
... That's pretty much all this guy is offering. What other "interesting" feedback does he have?
 
What was that infinite resolution tech a while back? Just voxels. It was sold on all sorts of complaints about how triangles suck and how they were redefining computer graphics to do it 'properly', but they hit the same issues everyone else hit working with voxels. But he got money, ran a business and paid himself a salary. Last I heard I think they were doing exhibition stuff. No doubt they solved some of the voxel issues simply because they worked at it. We have some amazing voxel tech these days like Teardown.

Game tech, and rendering tech, is an open book. No-one is likely to come along and do anything revolutionary. There are alternatives to Unreal Engine like Flax, including ground-up engines. Backed by talented people working long hours for years, and producing something that has strengths and weaknesses.

Every rare once in a while you get something completely revolutionary like Dreams and their SDF splatting (typically spawned from some white paper and academic research) or Megatexturing. And these spawn other creations like Sebbbi's Claybook. And these ideas are then found to have limitations, and pros and cons, and aren't the magic bullet some believed them to be.

In short, there's no low-hanging fruit. The smartest minds have been working these problems for decades. Some of them are employed at Epic! Every solution is a balance of pros and cons. Created a next-gen mind-numbing tech? Great. Now create the tools to actually build and deploy a game with it. That's what Unreal gives you.

The difference between Threat Interactive and the 'grown ups' is professional engineers talk shop and present evidence and proper papers, whereas this guy is just courting the general public with a his perspective and arguments, which means selling a dream, and that way, like crowdfunding, you just need to work on people's faith to win over their investment. There's no science (or, I prefer, philosophy). That's where people like Sebbbi, real engineers, present useful signal in contrast to this noise, and their perspective have real basis when considering the state of the industry, where we are, how we got here, and where we ought to be headed.
 
What was that infinite resolution tech a while back? Just voxels. It was sold on all sorts of complaints about how triangles suck and how they were redefining computer graphics to do it 'properly', but they hit the same issues everyone else hit working with voxels. But he got money, ran a business and paid himself a salary. Last I heard I think they were doing exhibition stuff. No doubt they solved some of the voxel issues simply because they worked at it. We have some amazing voxel tech these days like Teardown.
You mean Euclideon's unlimited detail point cloud rendering technology ?
 
Almost wish Threat Interactive was just outright banned subject on this forum. My main issue is he calls it his development studio, but there's no evidence it actually has employees. Does he have a registered business? He talks about their standard development setup, which is a 3060, but no evidence it's not just him and his personal computer. He says he won't release his "art"/game because he refuses to release it with the damaging visual downsides of UE5 like TAA blur, but has shown basically nothing to prove they're doing anything at all. He offers a pie in the sky "ai" tool that will create LODs on the fly, and is soliciting money for that (and a UE fork), but shows no evidence that "they" are actually working on it. Again, where's the studio, where's the product? Any demos or signs of progress at all? He's soliciting money through youtube donations instead of through a platform that offers some accountability, like an established crowd-funding platform. He can take all of that money and just do whatever he wants with it.

His latest video is basically him whining that people are attacking his character for all this. And yah, I'm am. This guy can go fuck himself until he actually proves out any of it, and most of it is super easy to do. He comes out swinging attacking professionals in the industry and then gets upset when people want accountability from him. He's mentions he was banned from a graphics discord, tries to play it off as the industry being out to get him for telling the truth. The most likely answer is he's a shit disturber and rude, like he is in his videos.

His takes about games pale in comparison to the fact that he's outrage farming and potentially scamming a lot of people.

On the gaming takes, it's all the most basic shit.
TAA sucks. I agree. This is the least controversial sentiment in gaming.
Nanite has an overdraw problem. From what I've read and understood, it does not. The quad overdraw problem does not apply because of the way shaders are launched with a visibility buffer.
Nanite has overhead. Yep. Again, not super controversial, but in the case of rendering very small triangles it's a win over the hardware path. Whether this is good or bad is subjective.
Baked lighting is better in static environments. Yes and no. The environments look good, but the dynamic objects (player characters, physics objects etc) in those environments don't.
... That's pretty much all this guy is offering. What other "interesting" feedback does he have?
The problem is that a lot of these points are specific to UE5. The Finals runs at 100 FPS in nativ 4K with RTXGI (UE 5.1). Marvel Rivals is in the 6x FPS range on a 4090. UE5 lacks the middle ground - efficient geometry and lightning processing. You either get everything with low frames or "nothing" with >100 FPS. But you can not create a game like Metro Exodus EE with UE5...
 
Very insightful way of looking at it. Honestly, bit surprised this didn't occur to me earlier that this is what was happening. You'd be crazy not to be monetizing social media based on reach alone. You could sell anything really.
Read a recent article about rage-baiting. Deliberately creating content designed to rile viewers into posting rage, because that gets engagement which pushes adverts and make money. So now you have people creating personas just to be obnoxious and generate lucrative hate clicks. This almost certainly has feedback loops that shift the gestalt. It must be difficult as a tech site like DF to resist the cultural norms and stick to older standards of real journalism, but thankfully they do and have found their audience as a result. I dare say other content aspiring to be informed and objective is being skewed into, at the least, baiting titles to appease The Algorithm, if not ending up with an agenda to attract a constant emotional fanbase.
 
The problem is that a lot of these points are specific to UE5. The Finals runs at 100 FPS in nativ 4K with RTXGI (UE 5.1). Marvel Rivals is in the 6x FPS range on a 4090. UE5 lacks the middle ground - efficient geometry and lightning processing. You either get everything with low frames or "nothing" with >100 FPS. But you can not create a game like Metro Exodus EE with UE5...
Well UE5 is the problem isn't it so all the points would be centered around UE5. While I don't agree with their methods and their over simplification of many issues, what I can agree on is that poor tradeoffs are being made as it relates to games. I can also agree that optimization has taken a back seat in this generation. There have been way too many instances of this seen even outside of UE5 for example, the horribly optimized GTA online loading fixed by t0st. Of course, that happens when you have people using tools(UE5) that they don't understand.
 
Back
Top