Adaptive tessellation is real!

Adaptive tessellation isn't a new concept to video games at all, but the fact that we notice it, and see it change, isn't something we should applaud. This should be happening much faster, and we shouldn't be seeing the transition this easily. It should be happening earlier in the zoom in, so it seems like a seamless change. It should be an easy adjustment for them. Either that, or it's just so slow to load the newly tessellated geometry, that the PS3 just can't update quick enough.

This same tessellation happens during render times using the V-Ray render engine as well, where you can adjust how much a surface will subdivide based on how far it is from your render camera. The closer you get, the more it subdivides, but with the right settings, you'll never actually see the switch.
Adaptive Tessellation was unheard of in 2006 when the PS3 came out. I think it is worth an applause no matter what that PD managed 50 times more HDR over GT5, Adaptive Tessellation, and these amazing physics on 7 year old hardware. But, as you know; to each their own. This game looks incredible for what it's running on.
 
Last edited:
I think it looks pretty good considering what they were given to work with as well. Adaptive tessellation is definitely more prominent on PC games, where we are given the choice of changing the quality. On consoles, yes, it isn't as common. Given that, I would think the result would still be better considering what they test it on is what we play it on. What PC games are tested on isn't always what people will experience at home. I'd suspect this is why PC games tend to be more on the buggy side than console games.
 
it's instantaneous
I used some trick to confuse the system so it won't change as quick as it should be
Might want to edit your first post to mention this, there are many post after this one about how slow it is in the video because they are not reading the whole thread before posting.
 
There are not LODs used with adaptive tesellation. The same base model is rendered in infinite steps in real-time depending of the distance to the camera. In photomode the car is rendered with more steps along with the increased resolutión and antialiasing.

It's a great technology that release rendering resources when are not required (high detailed models in the distance) increasing the visual quality and saving resources avoiding to load all the alternate car LODs into the memory.
This is good, and will carry over directly to the PS4. And will be handy for 20+ cars on track.

GT has been doing the LOD swap since GT 1, it's what made replays so striking and B spec so beautiful, and photo mode so detailed.

Now with a real time means to adjust models on the fly we will see a lot more cars in future versions and an indication PD are thinking ahead. We can only hope they come up with a similarly elegant solution to audio, or it will just wait till GT 7... I believe GT 6 is mostly a primer for 7... But that could be argued about every version.
 
This is good, and will carry over directly to the PS4. And will be handy for 20+ cars on track.

GT has been doing the LOD swap since GT 1, it's what made replays so striking and B spec so beautiful, and photo mode so detailed.

Now with a real time means to adjust models on the fly we will see a lot more cars in future versions and an indication PD are thinking ahead. We can only hope they come up with a similarly elegant solution to audio, or it will just wait till GT 7... I believe GT 6 is mostly a primer for 7... But that could be argued about every version.
I expect the car will be made by more than 3 million polygons in GT7
in addition, the track model will be hugely improved, no more paper trees.
 
That is incredible, haven't really noticed it until you made this video!

wyof.jpg
I am glad you like it
when I discovered this, I was so exciting and could help to tell other GT fans which is why I made this video
 
I noticed it also during a race. Details are only on cars when you're close. Especcially when you are about to overtake a rally car you will notice it. All the rally gismos like mudflaps start to show up when you're close.
 
I expect the car will be made by more than 3 million polygons in GT7
in addition, the track model will be hugely improved, no more paper trees.
Cars dont need to be 3 million polys. As mentioned, there's a base mesh, and its subdivided as much as it needs to be based on the distance from the the camera. These base meshes will never need to be more than 300-500k if you ask me. I'm modeling a high detail 135i right now that has way more intricate detail than a video game model would have, and it will be under 3 million, yet hold up to extreme close up renders. I have an 07 STi that's in the same boat, fully modeled interior too, comes in at 700k. 3 million is a waste of resources.
 
Cars dont need to be 3 million polys. As mentioned, there's a base mesh, and its subdivided as much as it needs to be based on the distance from the the camera. These base meshes will never need to be more than 300-500k if you ask me. I'm modeling a high detail 135i right now that has way more intricate detail than a video game model would have, and it will be under 3 million, yet hold up to extreme close up renders. I have an 07 STi that's in the same boat, fully modeled interior too, comes in at 700k. 3 million is a waste of resources.
I mentioned 3 million poly because even with 500k polys you can still see that interior modelling isn't smooth enough.
had they made it 3 million polys for basis, this problem could be solved.
I believe PS4 has the potential to achieve it.
 
All they'd need to do is subdivide it once. Theres no need to increase the base mesh. Its like cutting a 2x4 too short.
 
All they'd need to do is subdivide it once. Theres no need to increase the base mesh. Its like cutting a 2x4 too short.
well.....it is true that you can use mesh smooth function to quickly increase the poly count
however it would be the wast of resource
I mean they can make corner part more rounder such as dashboard and buttons more smoother especially for interior part.
 
Thats an even bigger waste of resources. Does the interior need to be that detailed when its an AI car beside you?
 
Are you sure it's adaptive tessellation ? Looks like a basic algorithm saving polygons but in real time. I wonder why this technology appears only now if it's the case.
 
if it makes graphics better then why not

But we're back to the point of the whole adaptive tessellation discussion............
The idea is to keep a base mesh low so that at a distance, the vehicle looks good without any excessive details that would go unnoticed. When that base mesh is subdivided, all the details are still there, and panels don't start to take on other shapes.

You can't read a badge on a car from 500ft away, so why have a very detailed badge on a car at that distance when it can't be seen?
 
Back