Hi, just for your info, I tried your scene from M1 thread and got about ~1 second to finish render on a RTX 2060.
It is a very nice scene, congratulate.
I tested your scene.
I had to activate CUDA in Blender preferences. I tested CPU, RTX 3060 and Eevee. it is a 35mb mp4 file made with OGS (yes it takes a bit of performance). I don’t have Youtube or Google files account. Do you have an alternative for a place i should upload it to?
1 Like
Renzatic
(Egg Nog King of the Eastern Seaboard)
#679
There’s Dropbox and Onedrive, though it can be kind of a pain to share files with the latter.
For a video, your best bet is to just to sign up to Youtube, and upload it there. If you’re suspicious about Google’s tendency to get all up in your grill over everything, use a burner email.
If I understand things correctly, Cycles-X will abandon OpenCL to rely on some sort of AMD API for AMD GPU hardware acceleration.
My question is if there is any sort of hardware acceleration planned for Macs, be them old x86-based ones with Metal-compatible AMD GPUs, x86-integrated GPUs, and/or Apple Silicon-based Macs.
I’m the owner of an old Mac Pro 2011 with two RX 560 Metal-compatible cards (plus a just as old PC that hopefully will be replaced somewhere next year), and I’d like to know if I can look forward to some second wind upon my old beast’s sails .
I only used the Cycle-X builds to make this (posted in the finished works section recently) I hit couple bad builds, which made me lose some work/progress but that was expected.
The maggots and the flies were populated with Geometry Nodes.
I did not even need to use the LookDev (Material Preview) mode during creating it. Most of the time the LookDev mode is totally useless because of slow shader updates and viewport lags with complicated and texture heavy scenes. I find Cycles-X viewport update to be faster than the Material Preview mode, and I get better real-time feedback since the LookDev mode drops to white shading during shader compiling.
I tested todays 3.0a build including the recently merged Cycles X update with a scene that was created in 2.93 (can’t share it due to some of the models). It has a lot of foliage scattered around and some volumes.
Compared to 2.93, I experienced a significantly higher consumption of vram without speed ups in spite of using the same settings (resolution, samples, noise threshold, denoising).
2.93: 5589MB
3.0a: 10743MB (tiled rendering)
3.0a: 18293MB (progressive refine)
Are these differences normal as it’s still in alpha stage or am I missing something?
Also, viewport navigation seems to be way slower although all scatter systems are turned off (not related to Cycles).
Has anyone done any experimenting with the tile size in C-X? The default is 2048, which is too big if you are only rendering at 1920x1080, but is too small if rendering at 2560x1440 (which I often do for stills). I tested with CPU rendering at much smaller tile sizes (32 or 64), but it just bombed out after a few tiles.
Having gotten used to not worrying about tile sizes during the beta stages, I’m wondering how much difference it really makes, now. If I have to, I’ll test with a render heavy scene, but would rather not repeat someone else’s research.
If I render an acually large size the whole render disappears when it’s finished and I can’t save the final image. I have to shrink the tile size to like 256 of smaller to get a final image I can actually save (but that was a build from a few days ago)