Cycles Development Updates

IIRC, the compatibility list for each of the GPU backends (as of 2.92) is:

CUDA: Nvidia 6xx and later
OptiX: Nvidia 9xx and later
OpenCL: Intel Xe, AMD GCN2 and later

In some ideal world… developers will “unite” and make one system. Instead software developers have CUDA, OptiX, OpenCL-Vulcan, Metal… I must resist urge to post some juicy meme here :laughing:

XKCD already did it for you.


…meh… I know for this picture. This is partially true, but in fact deception, false argument.
Problem is that big dogs have his own selfish goals. Imagine that Crapple throw all this money in OpenCl-Vulcan instead to develop his walled Metal. Nvidia want to exploit his Cuda-Optix advantage. AMD will probably do same, develop stuff which favor his GPU’s.
Good luck to Blender developers which must with limited resources to work on all this… and listen nagging of dissatisfied users. Same is for game developers, etc. Users - consumers are hostage of his games.

Vulkan isn’t like CUDA OpenCL or Optix, is much more, is another world.

In this case, the various standards are caused by hardware vendors, which is not really caused by software developers! If the software developers were in charge, there would be more variations :smiley:


It worked fine with USB.
Since all phones have to use USB it is much easier to use someone elses charger if you forgot yours.
Sure, USB chaged from B to C recently but one update every couple of years is a lot better than a new standard from each and every single manufacturer every couple of years.
Also, other devices are more and more using USB for powering which is convenient.

1 Like

Until this happens:

Because USB-C is actually 15 competing standards in a trenchcoat.

1 Like

And do note that this standardization only happened because of large government intervention:

so for all you “government regulations make it harder for corporations and consumers” soapboxers out there:


Do you really believe the average politician is tech. savvy enough to actually pick the best standard (so companies are not downgrading product performance to please the state)? Are you going to trust that the standards chosen won’t be because of sweetheart deals with donors and lobby groups?

For example, what if the state outlawed CUDA and Optix in favor of OpenCL, are you in favor of Brecht potentially having to cut features for GPU users for political reasons?

It is probably not perfect but a lot better than before. I have never had any trouble at all since usb became standard. Before there was allways somebody asking for a charger with not correct one.available

I wont respond to strawman what-if arguements.

Before government interaction, all cellphone companies used different chargers. after, they (mostly) standarized. That’s a win for consumers and producers.

Do you really want the state of cellphone chargers to look like this?

That doesn’t look better to me.

Note that those EU regulations are entirely voluntary. Which is why apple can feel free to piss in everyone cheerios when they change connectors every 3 or 4 years. Sometimes companies appreciate leadership from the government and willingly go along with it because it helps.


Yes… but we are on some different page now with this thread.


Fair enough, end of sidebar!

1 Like

I managed to do some OpenCL and hybrid benchmarks on AMD gpus, Barbershop scene

Path Tracing:

2.90.1 - CPU - 10.49.28 - tile 32
2.90.1 - GPU - 09.31.32 - tile 256
2.90.1 - HYB - 06.43.39 - tile 64

2.92.0 - CPU - 10.23.51 - tile 32
2.92.0 - GPU - 09.18.00 - tile 256
2.92.0 - HYB - 05.33.78 - tile 64

Branched Path Tracing:

2.90.1 - CPU - 03.17.17 - tile 32
2.90.1 - GPU - 03.52.88 - tile 256
2.90.1 - HYB - 02.43.22 - tile 64

2.92.0 - CPU - 03.07.47 - tile 32
2.92.0 - GPU - 03.47.48 - tile 256
2.92.0 - HYB - 02.19.93 - tile 64


Branched path tracing is that much faster than path tracing? that’s surprising unless the noise is considerably worse

1 Like

Indeed the noise is much worse. I just rendered with default values set in benchmark scene.
CPU is Threadripper 3970X, and the GPUs are 2 Radeon VII if anyone wants to know.

1 Like

Anyone know why this task is not talking about normal and displacment map baking?

The differences are quite significant. Did you use NLM denoising? This was the recent issue that was fixed for 2.92.

I’ve tried Branched Path Tracing several times and never came to a satisfying result. There was always noise in some areas that I didn’t have with normal Path Tracing. I had to crank the values up so high, that there was no gain in time in the end. That’s probably why they are going to ditch it.

1 Like