Technical Update in GT7 (PS4) from GT6 (PS3)
- Thread starter FoRiZon
- 81 comments
- 14,652 views
glassjaw
Premium
- 1,472
- PA
- pioneerspine
- pioneerspine
- 69
- Portugal
- jonny
cars and licences tets
- 28,470
- Windsor, Ontario, Canada
- Johnnypenso
Something GT definitely needs to put into place IMO as well. After playing pc sims for a while and being able to completely adjust cockpit view in every direction along with the FOV, and on the fly, I can't see going back to a fixed cockpit position. Not sure how Forza does it on this front but it should be a very simple thing to put in the game if tiny sims like Assetto Corsa can do it.
- 1,239
- Italy
Late reply but I was re-watching GTS trailer now. (thanks PSX)
I said "in relation to hardware" and "not as" - obviously it still looks great but not as "wow" as GT4 at launch. (speculation)
As for blocky parts I meant this:
- 1,806
- Belgrade
- FrancSRB
I wont look cars under microscope to see that stuff... I bet that you could find same blocky part in Forza, PCARS, Aseto... They better use PS4 power for enviroment that need a lot of work and AI, physics.
- 1,239
- Italy
Actually you're right. I looked at the trailer again and in a frame before the same part looks smooth. This looks more like what we'll be seeing while driving. 👍
- 2,127
- FL, USA, Earth, Sol, Local Bubble, Milky Way
- TeamACRZARacingo
- GyroxOpex
Forza doesn't do it.
- 154
I'm going to hold judgement until the game is released, but I don't see how it can be any worse than what's already out there now. Just need to keep expectations in check 
I wonder if PD will take advantage of the extra core that's been freed for devs, can be a slight processing power advantage over what pcars and other games have access to. Maybe it's not worth it though if it means delaying the release.
I'm just happy we're finally getting a GT game on the ps4 next year.
I wonder if PD will take advantage of the extra core that's been freed for devs, can be a slight processing power advantage over what pcars and other games have access to. Maybe it's not worth it though if it means delaying the release.
I'm just happy we're finally getting a GT game on the ps4 next year.
glassjaw
Premium
- 1,472
- PA
- pioneerspine
- pioneerspine
It's probably LOD-related, so if the camera was positioned closer to the dashboard inside the car it might be more round.
- 1,548
- New Zealand
- MHFanatik
Photo Travel always uses the highest LOD though. I think the case is that they couldn't make tessellation work on PS4, which is a bit odd, but not a big stretch, tessellation always seemed like a BS deal most of the time, they made a hack workaround using CELL, but I'd imagine that code is not compatible with x86 architecture in any way.
- 1,748
- vspectra06
- vspectra06
Star Wars Battlefront and NFS have adaptive tessellation.
- 1,548
- New Zealand
- MHFanatik
- 11,788
- Marin County
Maybe somebody more well versed in real-time rendering/game engines can answer this question.
In 2015 why are we still using fixed polygon-based meshes for models? If anyone is familiar with nurb-modeling programs like Rhino, all surfaces and solids are created by mathematical formulas (so they are essentially infinitely scaleable geometry similar to vector artwork) which is then represented/drawn by the program (usually with meshes) based on its distance from the viewer.
Instead of a referenced pre-loaded triangle, every piece of surface geometry is defined by this equation within the program:
It's essentially the "pure" form of adaptive tessellation. If anyone has used autocad before, the command "regenerate" redraws all of the vector geometry to appear smooth based on the distance from the camera. Programs like illustrator and Rhino do it seamlessly. It seems GT6 has a few preloaded (ie already generated mesh) detail levels that it can scale between based on the camera angle. Why not go full monty and generate everything with mathematically-defined models that are scaled automatically based on distance? It would save TONS on disk space as well. I suppose the trade-off is that you would need a very strong graphics card to produce a smooth render.
In 2015 why are we still using fixed polygon-based meshes for models? If anyone is familiar with nurb-modeling programs like Rhino, all surfaces and solids are created by mathematical formulas (so they are essentially infinitely scaleable geometry similar to vector artwork) which is then represented/drawn by the program (usually with meshes) based on its distance from the viewer.
Instead of a referenced pre-loaded triangle, every piece of surface geometry is defined by this equation within the program:
It's essentially the "pure" form of adaptive tessellation. If anyone has used autocad before, the command "regenerate" redraws all of the vector geometry to appear smooth based on the distance from the camera. Programs like illustrator and Rhino do it seamlessly. It seems GT6 has a few preloaded (ie already generated mesh) detail levels that it can scale between based on the camera angle. Why not go full monty and generate everything with mathematically-defined models that are scaled automatically based on distance? It would save TONS on disk space as well. I suppose the trade-off is that you would need a very strong graphics card to produce a smooth render.
Last edited:
- 591
- Spain
- Freeman_Cruz
That is precisely as modeling (only for certain cars) of GT6, and presumably GT Sport, works.
Check this
And this
https://www.gtplanet.net/forum/threads/tessellation-car-list.307532/
Check this
And this
https://www.gtplanet.net/forum/threads/tessellation-car-list.307532/
- 11,788
- Marin County
That is precisely as modeling (only for certain cars) of GT6, and presumably GT Sport, works.
Check this
And this
https://www.gtplanet.net/forum/threads/tessellation-car-list.307532/
I believe AT is mesh refinement/subdivision rather than mathematical modeling (even though it involves math)
Last edited:
- 28,470
- Windsor, Ontario, Canada
- Johnnypenso
- 4,101
- California
- Sk8er913
The lighting in that screenshot is hideous. The specularity is all kinds of terrible.
- 15,532
- Cairo, Egypt
- GTP_SEMS
The adaptive tessellation stuff is interesting and it was impressive that the PS3 was capable using such an advanced rendering feature considering the age of the hardware. That being said though, the adaptive tessellation in GT6 was not that effective, anyone who has GT6 fresh in their minds will remember that the extra detail on AI premium cars used to pop-in as you got closer to them. The reason (that may explain) why you got pop-in on the car models requires a lengthy explanation so bare with me.
The whole point of the adaptive tessellation feature in GT6 is to reduce GPU RAM usage, most games use a technique called Level of Detail, as an object gets closer, the game swaps the model with a higher poly model and vice-versa which allows a game to look as detailed as possible whilst being efficient with polygon counts. The problem is that it takes up GPU memory as you have to store multiple meshes with varying poly counts. Adaptive Tessellation addresses this issue as it works by having just one-mesh that effectively morphs into higher or lower poly counts depending on how close the object is. The downside to adaptive tessellation is that it is heavily reliant on processing power and this is where the PS3 was only half successful.
The Cell processor was a powerful CPU for its time but it is still not really good enough to fully take advantage of adaptive tessellation. The reason why you had pop-in (this is my speculation) is probably because the adaptive tessellation was not quite fast enough and so it swaps the mesh with a higher poly version just as you get close enough to the model. So it sorta half works, perhaps reduces the amount of meshes needed but the game still uses LOD as well.
As far as PS4 using adaptive tessellation? I doubt it. You can make a case for the PS3 utilizing adaptive tessellation because the PS3's RSX graphics chip was a bit weaker than the X360's Xenos GPU in terms of its triangle set-up rate which caused developers problems as it was difficult for them to match the same poly counts that the 360 was capable of. In the case of the PS4 it is a much better balanced machine, the PS3 had a very powerful CPU coupled with a mediocre GPU, whilst the PS4 is nothing special compared to PC, the triangle set up rate doesn't provide the same challenges as the PS3 did. The PS4's CPU (which is part of an APU chip) is not that much more powerful than the Cell processor and so the likely hood is that even with the PS4 there isn't enough processing power to fully realize adaptive tessellation techniques. Another thing you have to consider is that the PS4 has 8GB of RAM which was a lot more than people were expecting, it is considerably more than the 256mb of RAM for the RSX GPU. With having so much more RAM, the standard LOD method probably does not present an issue anyway.
In terms of Poly counts I would expect somewhere around 100,000-150,000 polys per car mark, driveclub has 250,000 polys per car but it is a 30fps game. GT6 probably somewhere in the region of 30,000 polys per premium car at its maximum during gameplay.
The whole point of the adaptive tessellation feature in GT6 is to reduce GPU RAM usage, most games use a technique called Level of Detail, as an object gets closer, the game swaps the model with a higher poly model and vice-versa which allows a game to look as detailed as possible whilst being efficient with polygon counts. The problem is that it takes up GPU memory as you have to store multiple meshes with varying poly counts. Adaptive Tessellation addresses this issue as it works by having just one-mesh that effectively morphs into higher or lower poly counts depending on how close the object is. The downside to adaptive tessellation is that it is heavily reliant on processing power and this is where the PS3 was only half successful.
The Cell processor was a powerful CPU for its time but it is still not really good enough to fully take advantage of adaptive tessellation. The reason why you had pop-in (this is my speculation) is probably because the adaptive tessellation was not quite fast enough and so it swaps the mesh with a higher poly version just as you get close enough to the model. So it sorta half works, perhaps reduces the amount of meshes needed but the game still uses LOD as well.
As far as PS4 using adaptive tessellation? I doubt it. You can make a case for the PS3 utilizing adaptive tessellation because the PS3's RSX graphics chip was a bit weaker than the X360's Xenos GPU in terms of its triangle set-up rate which caused developers problems as it was difficult for them to match the same poly counts that the 360 was capable of. In the case of the PS4 it is a much better balanced machine, the PS3 had a very powerful CPU coupled with a mediocre GPU, whilst the PS4 is nothing special compared to PC, the triangle set up rate doesn't provide the same challenges as the PS3 did. The PS4's CPU (which is part of an APU chip) is not that much more powerful than the Cell processor and so the likely hood is that even with the PS4 there isn't enough processing power to fully realize adaptive tessellation techniques. Another thing you have to consider is that the PS4 has 8GB of RAM which was a lot more than people were expecting, it is considerably more than the 256mb of RAM for the RSX GPU. With having so much more RAM, the standard LOD method probably does not present an issue anyway.
In terms of Poly counts I would expect somewhere around 100,000-150,000 polys per car mark, driveclub has 250,000 polys per car but it is a 30fps game. GT6 probably somewhere in the region of 30,000 polys per premium car at its maximum during gameplay.
Similar threads
