The most wanted feature?

[John_Do]

As appealing as could be the ability to run Corona on GPU(s) and gain a (really) big boost of raw performance, I can understand the reluctance of the devs to port the engine to GPU :

• a CPU render engine code is almost universal. A GPU render engine is Nvidia only or you have to be ready to maintain no less than 3 other versions : OneAPI( Intel), HIP(AMD), Metal (Apple). I don't know the details but I guess that's a bit more work than maintaining just one render engine if you don't want to tie your user to one videocard brand ( and even just NVidia is two implementations, CUDA and OptiX).
• you are heavily dependent on GPU drivers, and this is not a detail. Here how it goes for Redshift users for 6 months : Super slow redshift IPR and rendering
• And finally, the small amount of VRAM available, even on prosumer models, does not fit very well the type of project that the typical Corona user has to deal with. And NVidia killed NVlink with the RTX 4XXX series without offering a true alternative, so either you're stuck with 24Gb or you bite the bullet and spend 5K+€ on a RTX6000. Even so, it's "only" 48Gb and like the gaming cards, no more NVLink too.

Time to first pixel and responsivess on Cycles and Octane are crazy compared to CPU, so I would really like to see big improvements on the interactive rendering.

[Jpjapers]

As appealing as could be the ability to run Corona on GPU(s) and gain a (really) big boost of raw performance, I can understand the reluctance of the devs to port the engine to GPU :

• a CPU render engine code is almost universal. A GPU render engine is Nvidia only or you have to be ready to maintain no less than 3 other versions : OneAPI( Intel), HIP(AMD), Metal (Apple). I don't know the details but I guess that's a bit more work than maintaining just one render engine if you don't want to tie your user to one videocard brand ( and even just NVidia is two implementations, CUDA and OptiX).
• you are heavily dependent on GPU drivers, and this is not a detail. Here how it goes for Redshift users for 6 months : Super slow redshift IPR and rendering
• And finally, the small amount of VRAM available, even on prosumer models, does not fit very well the type of project that the typical Corona user has to deal with. And NVidia killed NVlink with the RTX 4XXX series without offering a true alternative, so either you're stuck with 24Gb or you bite the bullet and spend 5K+€ on a RTX6000. Even so, it's "only" 48Gb and like the gaming cards, no more NVLink too.

Time to first pixel and responsivess on Cycles and Octane are crazy compared to CPU, so I would really like to see big improvements on the interactive rendering.

I think hybrid could the the ideal. Where you have specific features that run really well on a GPU, supplementing the power of the CPU. Volumetrics, caustics ansd displacement being three things off the top of my head that i can think of that will add significant render time to a scene at the moment.

If theres any way that any part of the most intensive processes can be moved off to the GPU, even partially, leaving more CPU free to deal with everything else. I'd absolutely welcome that. That being said i have no idea how linear the computational rendering process is or whether anything i just said is viable. But more speed is always welcome!

[Dionysios.TS]

As appealing as could be the ability to run Corona on GPU(s) and gain a (really) big boost of raw performance, I can understand the reluctance of the devs to port the engine to GPU :

• a CPU render engine code is almost universal. A GPU render engine is Nvidia only or you have to be ready to maintain no less than 3 other versions : OneAPI( Intel), HIP(AMD), Metal (Apple). I don't know the details but I guess that's a bit more work than maintaining just one render engine if you don't want to tie your user to one videocard brand ( and even just NVidia is two implementations, CUDA and OptiX).
• you are heavily dependent on GPU drivers, and this is not a detail. Here how it goes for Redshift users for 6 months : Super slow redshift IPR and rendering
• And finally, the small amount of VRAM available, even on prosumer models, does not fit very well the type of project that the typical Corona user has to deal with. And NVidia killed NVlink with the RTX 4XXX series without offering a true alternative, so either you're stuck with 24Gb or you bite the bullet and spend 5K+€ on a RTX6000. Even so, it's "only" 48Gb and like the gaming cards, no more NVLink too.

Time to first pixel and responsivess on Cycles and Octane are crazy compared to CPU, so I would really like to see big improvements on the interactive rendering.

But, the VRAM used in rendering process, is completely different than the traditional RAM.
In 24GB can fit a lot of data that in the traditional RAM will occupy 3 times more.

In 2009 I had the privilege to be one of the first artists using the iray rendering engine. At that time, the GPU cards were having only 1.5GB VRam and look how many stuff was rendered with that "limited" memory! :D

Anyway, I am curious to see how things will evolve in the years to come.