I don‘t know if I understand you correct but the M3 silicon chips (2023) support hardware raytracing - for example blender 4.0 (with cycles) use the hardware raytracing on a Mac (M3) to boost render times.
M3 does support hardware raytracing, but the machines one might think of as a workstation haven’t been updated yet with those chips. The notebooks and a small-screen iMac (the 27" seems to have retired) have been released with M3 chips, but there’s no M3 Ultra out (Mac Studio or Mac Pro).
Yes, any sort of performance comparison would be highly appreciated.
I am still on RH7 and an Intel Macbook with a AMD eGPU. So far this was still faster than any Apple Silicon based mac, especially the viewport. With Rhino 8 finally having full metal support i will get an upgrade License soon and it makes probably sense to also a new Rig at the same time.
I am still undetermined on which generation to get as technically there are not many differences. M1pro/max has the best power efficiency and is not much slower than M2 and M3. M2pro is as fast and cheaper than the M3pro. And while the M3max has it all, it is insanely expensive. A M1max machine at the same price as an M3pro machine would likely be faster and has almost triple the VRAM bandwidth.
But if RH8 will make use of the hardware RT of the M3 series in the future, this could have performance implications that are greater than the difference in CPU core count and bandwidth.
Can anybody answer or guesstimate these two questions?
- Will RH8 eventually get the new Cycles renderer and make use of M3s Hardware RT?
- What is the approximate performance increase of the new Cycles renderer when it makes use of M3s hardware RT (Anybody who already used the new Cycles version with blender on an M3 maybe?)
1 Like
Got an answer to my second question from the blender opendata Database: Blender - Open Data
M2Pro (19C): 892.79
M1Max (32C): 930.67
M1Ultra (48C): 1379.18
M3Pro (14C): 1430.82
M2Max (30C): 1443.32
M3Pro (18C): 1500.92
M1Ultra (64C): 1621.08
M2Max (38C): 1765.03
M3Max (30C): 2851.28
M2Ultra (76C): 3218.52
M3Max (40C): 3424.82
RTX 4070 Laptop: 3583.34
RTX 4080 Laptop: 5638.85
RTX 4090 Laptop: 7058.49
So the M3 GPU is about 2 times as fast per core compared to the non-HW RT M2 GPU and 3 times as fast per Core compared to M1 GPUs in Blender/Cycles 4.0
It seems the speedup from Hardware RT is less than what nvidia gained when they introduced Optix, which was about 3x times faster than with CUDA (which in turn is 3x times faster than CPU rendering on an Intel 8C/16T Xeon E2286M).
I hope that version of Cycles will make its way into RH8 some time in the future.
1 Like
Also surprised, that the hardware RT difference is not that big. Maybe this depends on upcoming optimizations. But I also don’t exactly know what I am talking about haha, so lets see.
But still, the M3 Max values seem very promising.
Upgrade frim Cycles 3.5 to Cycles 4.0 or whatever latest is when I get to it will happen at some point. Unknown when.
1 Like
Octane X 2024 is being optimized for new ray tracing hardware on M3 Apple GPUS. This is scene dependent, but is already showing 2x to 12x(!) speed gains in heavily instanced scenes:
Don’t know what they are comparing the 2x to 12x to, but it sounds pretty impressive.