And let’s not forget about 2080 Supers and 2070 Supers. Those are not terribly far behind, for way less dough, obviously. I’d actually love to see some benchmarks (say, classroom scene) between 2080Ti, 2080 Super, and 2070 Super. In OctaneBench a 2080Ti gets a score of 302, 2080 Super = 233, and 2070 Super = 220 (OB scores are MegaSamples / Sec). If any of you have either, and can bang out some benchies in E-C, that would really be nice if we can confirm that E-C results are similar to OctaneBench (2070 Super is 72.8% as fast as 2080Ti, 2070 Super is 86.6% as fast as 2080 Super, etc, etc)…
I think I’m leaning towards three 2070 Super Hybrids (under $1,800 total) to replace my three 980Ti Hybrids (but might very land on 2080 Supers instead). Now, for some reason, three 2070 Supers yield an OctaneBench of 739 (not the 606 you would expect), and there are three users that did the OB on three 2070 Supers, which makes a pretty safe bet that my score would be roughly 739. I think that is often the case with OctaneBench, that more GPUs score higher than the just simply multiplying # of GPUs x bench score of one. They seem to get a boost exponentially with more GPUs. I’d love to know if this happens with E-C too actually… So please folks benchmark 2080 Supers and 2070 Supers if you have them!
I feel you on this but also feel like next gen Ampere + more RTX development might be worth waiting for…it’ll be a while, though.
But a 980Ti for display and 2x 3080s or whatever they are called seems like it could be 2x as fast as the 3x2080 Super setup you described if everything falls into place.
Maybe, but I do not have the luxury of waiting. I have more and more animation work… Also, I am not on board with the practice of wasting precious pci-e slots on lesser GPUs. Ever. When I upgrade GPUs, they are all going to be the same, and I use all three for rendering, a bunch. I’m going to go eat lunch in a couple, and all three will be crankin’ while I’m munchin’… And all three will be burning the midnight oil tonight while I slumber…
Thank you for your awesome work on a Mac version so far. For my current animation project Cycles needs 5:30 to render a frame. E-Cycles renders it in 1:30 without any tweaks. It is simply a lifesaver.
I had 2 people who said the would need a Mac version in 2020 for now. From the feedback so far, many other ones are like you. Happy with the results of E-Cycles, but forced to upgrade to a more recent MacOS without CUDA to keep other apps updated. The other ones just say bye to Mac and go to PC. I’ll give it until the end of next week to take my decision to continue the Mac version.
It would be great if Mac E-Cycles was not quite dead yet. I’ll mainly stay on macOS so I’d be happy to avoid to introduce a PS to my workflow only for Cycles/E-Cycles (I planned to use a PS with Cycles/E-Cycles remotely on a Mac). If you’ll decide to continue with a Mac version it will allow me to leave a High Sierra Mac machine in use and forget about PS.
In my opinion you could update Mac E-Cycles even every 2-3 months - just to most stable Blender version.
I’ve been a Mac user for over 10 years, and I still love Mac,
but Apple abandoned 3D professionals, and they left CUDA driver unsupported for two years. I was tired of waiting for this. now I’m happy with my new PC.
Developing with focus on priorities is a good idea.
If you scroll up to Sept 24, Mathieu tries to explain to me in layman’s terms what CUDA Boost does. But now that you report slower renders with it (might very well depend on the scene I imagine), I’ll try to test this myself when I get a chance. I’m still not clear if it is meant for GTX GPUs only, or does it support RTX?.. Which do you have Daniel?
The Mac version of E-Cycles will be updated until the end of 2019 for sure. Still only 4 peoples showed interest in a Mac version for 2020, which even for 3-4 updates a year is too low yet. I’m waiting until the end of the week.
You’re welcome
While CUDA boost will be faster in most scenes, some will indeed be slower. That’s why I made it an option. Regarding the instability it may cause, it is fixed in newer NVidia drivers as you noticed, so I recommend to updated all your rigs to the latest one. @norka it’s only for CUDA, but it works on both GTX and RTX cards.
Aaaahh, okay BB. I guess I was thinking that OptiX was maybe CUDA v2, with a new name. Gotcha. RTX are still CUDA, OptiX is a separate thing, that is still CUDA-based. ; - )
RTX is a hardware instruction set to accelerate path tracing. Only OptiX and Vulkan can use it for now as far as I know. That’s why E-Cycles RTX uses OptiX.
OptiX is a framework focused on rendering. CUDA is a more general compute API.
Both RTX cards and the OptiX framework are compatible with CUDA. Blender’s integration of OptiX communicates with the older CUDA version.
Sorry to bug about this again BB… Hopefully this will be useful info for others too…
So can we just use regular E-C if we have mixed RTX with GTX GPUs, until we are able to procure all RTX GPUs, for us multi-gpu dudes? You can’t always get all at the same time (like on evga site you can only buy two at a time)… So, for the sake of argument, let’s say I had just one RTX 2080 Super and two 980Ti… Would I use the regular E-C (which would not be able to use OptiX for the RTX, but presumably the RTX 2080 Super would still render way the hell faster than a 980Ti), or would I be using the OptiX version, and my two 980Ti would just not be rendering as powerfully/efficiently as they would in the regular E-C? (hope that makes sense)
Sorry to be a pain pal, really, but because you said “At the moment” and also “If you render mostly…”, there is wee bit of ambiguity there, and I don’t want to leave anything to chance. I need to be rendering with as little interruption as possible. And for all I know you made changes recently that fixed mixing RTX an GTX…
