Render engine speed comparison
ฝัง
- เผยแพร่เมื่อ 6 ก.ค. 2021
- Cycles, Vray, Redshift, Luxcore, Arnold, Corona, Octane, E-Cycles and Cycles X - in the most thorough, head to head test ever.
Join Poliigon: bit.ly/3AGfQqc
Join my 'This week in 3D' newsletter: www.blenderguru.com/thisweekin3d
View the data in a spreadsheet: docs.google.com/spreadsheets/...
Sneaky Snitch Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 3.0 License
creativecommons.org/licenses/b...
-----------------------------------------------
Twitter: / andrewpprice
Instagram: / andrewpprice
Facebook: / blenderguru
ArtStation: artstation.com/artist/andrewprice
Broooo, when that two minute papers intro rolls in it felt like watching a movie with a surprise crossover
Károly Zsolnai-Fehér? What is this, a crossover episode?!
What a time to be alive!
Ian Hubert walks out of the shadows. "Andrew Price. I'm here to talk to you about the Blender Initiative."
what time to be alive!!!
Couldnt agree more lol, i started laughing in excitement like a little kid
I thought I was teleported
Finally some positive Corona news
Wahaaay! :)
XD
This rare case when words "positive Corona" in one sentence doesn't look like something evil.
Laughs in lockdown😂
the game engine is pretty decent as well
edit: apparently the engine was renamed to solar2d
as a luxcore fan, every time we have to render something we just step into a cryogenic chamber in total stasis until the renders are finished.
Try our RebusFarm rendering services for free!
@@rebusfarm yeah i'll check you guys out sometime
@@KirbyCurbwhy as prorender fan, every time we have to render something we just can easily render fast in Blender 3.0. Really, HD format had been rendered in 22 minutes and below. If you setup the Time Limit to 1 it will be rendered much faster!
Outstanding science.
"These are the tools we used, here's what we did, here's what's good/accurate about it, here's what's not good/inaccurate about it."
The fact that the scenes looked visibly different between engines in some of the renders has me concerned. Does that mean some engines aren't doing unbiased rendering? Because with unbiased rendering, you should always end up with identical looking scenes eventually if you crank up the sample count high enough. I know Cycles is supposed to be unbiased (except that it has a bounce limit, but it's very high by default I believe). This wasn't addressed much in the video outside of some subjective remarks.
@@AaronRotenberg they all fundamentally have different ways to set up materials
@@AaronRotenberg color grading is a thing, just altering contrast/gamma in some of these would make them match
@@AaronRotenberg I'd like to point out that even unbiased render engine are biased in some manner, they are faking reality. They just do it very well, and to do so, they all use different code, somehow comparable, but different.
Therefore the fact that you end up with differente result doesn't mean an engine is "less unbiased". Because even unbiased render engines are just simulations of reality.
@@AaronRotenberg if all unbiased renderers acted the exact same we wouldn't have multiple unbiased renderers.
My scenes aren't even getting finished and here everyone's talking about rendering.
I'm still on the donut tutorial 😭
@@nickchamberlin lol i just finished that. U got it
@@nickchamberlin i never finished anything enough to render since doughnut.
My level 3 donut with coffee took 40 mins to render only to found out that my condensation isn't positioned properly. After changes was done I rendered again and I forgot to turn on the plate collection. I hate my life
@@zaeriuz hey man it took me around 15 minutes everytime I rendered. One thing about 3D is you just gotta be patient
Blender Guru x Two Minute Papers? Talk about the greatest anime crossover!
Something Marvel could only dream of!
need closed captions! Two Minute Papers always have their CC... (waiting for Blender Guru to closed caption this video.... (>.>) )
But hold your paper
'What a time to be alive' eh? 😁👍
When two minute papers showed up I was like, “it’s time to get technical”
My heart literally skipped a beat when I heard his voice.
That was the moment when I remembered I hadn't clicked the Like button yet, and then clicked it immediately.
@@Lumpiluk That is the most true thing I’ve ever heard.
This is terrific. Very curious to see more types of these hardware / tech deep-dives and how the community can help feed you more info and understanding for future content--this stuff helps the community start off with quality foundational research to help others build off of.
Restarting a poliigon sub just to show some appreciation for the hard work.
Question! Still watching, but at 4:28 it says Arnold only works on the CPU - it does have GPU support (at least in Maya) and in my experience is far more accurate, more reliable, and MUCH faster than CPU rendering. Crossing my fingers that this is a typo as I keep watching; I'm going to be so bummed if Arnold's GPU testing isn't included!!
PepeLaugh
I use HtoA, Ther's a GPU render option. And look like it works.
I also wondered, but reckon that the GPU part wasn't available when they started the test.
@@felsenstern it's been out for more than 2 years
Understandably though, they didn't test in Maya (a nice program but dear lord is it expensive!)
Wow Two Minuet Papers!!! Amazing surprise guest! 🤩👍
Best Collab in history
Magyar. :D
Andrew brings on Károly Zsolnai-Fehér!?!
What a time to be alive!
I have been loathe to subscribe to any more channels for years now, but after just showing what you went through to design your tests, you obviously deserve it
As an avid listener to the podcast, when there weren't any uploads for quite some time, I felt like this was what was being worked on and what was going to be posted next. I applaud this. very well done. So well done in fact, that I pressed the like button 3 times using the same account
I already know this video is gonna be good.
yep
don't take it seriously, it was made for fun and likes/dislikes, these guys are not good in rendering at all, and even worse in testing
@@pavelsuprun8113 do it then
Am surprised how honest and well presented this is. Great research, much needed and appreciated by the community. KUDOS & thank You Andrew !
Great work Andrew! Thanks for taking time to put it together.....always helpful, well organized,
easy to follow. 😉👍
I'm pretty sure it's workbench, works great even with my potato Intel GPU
Yes
I have that potato thing😭😭
Agreed
Hmm, never used workbench does it look good?
@@calebsartain4267 OH yeah you gotta try it looks GREAT.😂
Finally he's back
The return of the king
Thank you for the video, Andrew! And regards to your assistant for figuring out 9 render engines :)
This is a big useful chunk of data, I will definitely be checking your website for the details.
Thank you for all of your generous contributions of knowledge, experience and expertise.
Great video. Glad to see cycles x making such big improvements.
Wasn't expecting Károly to make an appearance. What a pleasant surprise. I love his channel!
Thank you mate your contest is, was, and will be always amazing !
That's a very welcome crossover I didn't expect! Subscribed to both your channels. Keep it up!
"Nothing is free",
Except normal maps!
What?
Its something he says in lots of videos. Normal maps are free detail
@@Texplanations eh… they take up vram. When exporting game models to render with them i often have to reduce their resolution
I bet the guy who invented normal maps almost fell out of his chair when it worked.
@@blenderguru In other words, his normal direction was facing the floor 😂
This was such a fascinating video. Really amazing to see a real and thorough bench mark on all these renderers. Also, seeing TMP show up in the latter half was so fun!
Great work!
I am glad to see two channels that I really enjoy work together in a video combining art and technology.
Thank you Andrew for this huge amount of work. You are the best! Viva el Guru ❤️.
what a time to be alive!
#2 minutes paper 😂
oh yes, Two Minute Papers collabing w Blender Guru. What a time to be alive!
Thanks a lot for video and really appreciate the effort :). Many don't understand that in these types of videos, sometimes even a single sentence can be based on days or even weeks or research and experiments. Congratulations of putting it together and sharing it with everyone
Excellent video. I always love seeing actual data.
Beautifully explained video. Best part of the video was Károly's appearance. I saw a paper a few weeks back that showed promising results on caustics and cuts down render times quite a bit while still looking true to life. A few papers down the line and we might get ultra realistic renders on just about any render engine.
Also seeing the performance of Cycles X makes it a promising future for Blender, 🤞for good volumetric times.
the man with donuts is back at it again!
He done did it again !
Great Job Andrew! I'ts really great to have a reference for comparing different rendertools.
Just thanks for all you dedication.
Summary: "No engine is good at everything"
So we need to stick to what our engine can do well or suffer the longer iteration times.
Well obviously, nothing is perfect. Use what suits you, what you can afford and what you prefer.
Remember it's not the tool that's makes the artist, it's how you use the tools available to you
Summary: Chaos group won
@@anushkachathuranga8943 Only cause there wasn't enough volumetrics, Redshift 4 life
Thank you!
I think, in the end, what really counts is in which render engine you have more experience and are more able to optimize.
Well done Andrew, a helpful study and a good presentation, what a time to be alive....
Great video, I'm already using cycles-x for my projects and I'm very excited about its performance in the final version. so promising.
Second time I'm watching this video. Very explanatory. Good work!
Im most impressed with Redshift and its amazing detail of distant objects. Luxcore also looks good if you have a supercomputer on hand.
Wow, deffinitely didn't expect to see Karly! Great video and thanks to everyone involved!
Thanks, Andrew and Bill for making this enlightening video!
right now I say Renderman XPU is the next competitor in term of image quality and speed,
Renderman XPU vs Cycles X Optix please!!
Is it even out? Can't even find anything about it, only that it's in development. But from what I've heard it could be the killer of all renderers.
@@ExacoMvm Renderman 24 for Blender 2.92 is out. You need to download the renderer from Renderman website and the plug in from Github. Just be aware that is for 2.92 and it works, but not in 2.93, you can't even install the plug in. For that you have to wait for Renderman 24.1
Vray Gpu + CPU does exist since 2018. Even better with V-Ray that can out of the box render a single image using all cpus and gpus together from all machines in a farm.
in fairness, I haven't tried XPU, but can say when I tested Renderman last year it was about 2x SLOWER than another render engine I use all the time that was created and not updated since 2008. Renderman was painfully slow.
@@3DCGdesign that's because renderman always favors quality over speed. But that has changed with xpu, like it's really fast, and I say it's on par with Octane in term of speed while retaining the image quality we all love about renderman,andwe all know Octane takes the top in term of speed.
Still though the current version of renderman bridge for blender is no where perfect, lots of missing features or unexplained like the lack of udim support, but it does work .
Considering Cycles X is in its ALPHA stage, and it being at higher levels compared to the other render engines is extremely promising. It really just makes you wonder how much faster it could be in its final stages and for it to be a free render engine too!
Being in Alpha doesn't mean it's only gonna get better in terms of performance unfortunately. The algorithms/technology that make it better than regular cycles are already there, so it's core is not gonna improve. Right now they are mainly making it stable, solving bugs and adding support to volumetrics. So yes, it could get better by solving some bugs that hold it back, but also worse in order to fix other harmful bugs.
@@facuigua12 Yea that's a fair point actually
@@facuigua12 what i dont get is why don't they take a look at e-cycles optimizations and see what could be implemented without stomping on the devs work. (
@@Evandroworks I'm pretty sure E-Cycles is using a few approximations in order to render faster, that means it's no longer an unbiased renderer. Maybe the devs want to stay away from that, little changes like those can make or break realism
@@facuigua12 maybe but i mean it doesn't even support volumetric yet so they probably are still going to make changes that will probably help the speed a little.
YES YES YES! perfect video & even more perfect collab guest 😁👏👏 _my happyness is immeasurable & my day is saved_
Thanks so much for the efforts into this video!
I was literally searching this up on google yesterday.
don't take it seriously, it was made for fun and likes/dislikes, these guys are not good in rendering at all, and even worse in testing
Speed is by far the most important indeed. I've worked on client work at home.. the difference between a week of rendering, and a day.. means i spend 6 days quality improvement, which will yield better result in the end. Of course, a render engine has its limitation. But if you get 90% quality at half the time.. i'll take that more than 100% quality.
Time = money. Nothing more true has ever been spoken about business.
Try our RebusFarm rendering services for free!
Andrew this was an awesome video!! Thx for all the effort
i was waiting in anticipation for this video ever since i heard you talk about it on the podcast :DD
People with 3090's looking at this:
"We do not count"
😂
Google "Blinn's Law and the paradox of efficiency"
@@Tallone55 Wrong conversation dude.
@@sleepdeep305 Blinn's law states that with improved performance comes unchanging render times as you can make more complex and beautiful art with less time.
The claim is that people with 3090s don't care how long a render takes, but I'll just let you know, as a 3090 owner, that you use those time-savings to improve the end product. The information in the video is still absolutely relevant, regardless of the hardware you're running.
@@Tallone55 ooooooo you showed clayton whos boss
Renderman 24 would be interesting here.
Not really its slow as hell (in blender anyway).
My impression too, I feel it’s between Arnold and Luxcore, but would love to see comparisons in dense geometry scenes where renderman is excelling at.
@@GifCoDigital Renderman is a truck, not a ferrari. Try to render a city and compare it with the other engines then you'll understand why it's so powerful.
It would. Renderman gained in popularity while we were testing, but to finish the video we made the decision not to include it.
If I do the test again in the future, I'll almost certainly include it since I'm curious of the results myself.
Isn't that the render engine that Pixar used?
Thanks for this video. I'll show this video every time someone ask why I use 3ds max and vray
Absolutely loved the collaboration with 2 minute papers!
The thing about redshift (which I use for work) is although it's fast, it's by far the worst looking render engine I've tried. I use Corona at home and the difference is remarkable.
you like beer at home? why not a good beer? at least upgrade to pbr man!
Dislikes probably from LuxCore fans.
the 7 luxcore users lol
LuxCore user (and liked the video) but there were a few simple points that didn't even meet the simplified standard this comparison used:
- If Redshift/Vray had secondary solvers used than PhotonGI should be enabled for LuxCore, it's a single check box and makes huge differences to render times.
- Would have liked to see LuxCore included in the CPU results just for comparison
- For the Corona caustics example, it's again a single check box to enable them in LuxCore as well, an image for comparison would have been nice
Also workbench fan (a psyco) and eevee fan
@@zeealpal yes, phonton GI cache should be checked. anyway, I would love to see the setting in each render engine
Why? Everybody know it’s slow as fuck, but when it comes to lights calculations, caustics etc. It looks the best
Wow you guys put in a lot of work to create this video. Helpful information for sure!
You're great mate, great effort, great video
Cycles X is the future!
Ayeeee! "two minutes papers" 🎉✨
I'm currently working on Pixar's Renderman and it would have been cool to see the results of your tests on it too, but with all the informations you provided on the others I can already compare a bit with what I use now. Great video, thanks a lot !
Great Video! Thanks Andrew
Oh, TwoManyPapers on Blender Guru's channel! What a time to be alive!
someone finally did this
yes just leave it to the donut man
Amazing job to all involved, especially your new research assistant! :)
Finally. The video i have always wanted so i do not need to do tons of research and wasting tons of time testing things out.
Thanks a lot andrew, i love you forever 💕😘
14 Years, man.. I was there for like over 9 of those years! Also, you look like you haven't aged a day 😂 Good looks, man. Love your content even though I don't use Blender professionally.
With Cycles (and maybe Vray and Luxcore as well?) you can render with CPU+GPU simultaneously. This is a huge advantage when comparing speeds. I'm kinda surprised you didn't mention it.
Apart from that really great video and superb work. Thank you!
Wait i didnt know that thanks for info
Nice experiments Guru 🙏🏻
no even comparable, awesome work
An audible gasp escaped from my mouth in public when I saw the two minute papers logo appear.
And there wasn't a damn thing I could do about it.
Been thinking about render engines a lot especially with Cycles horrible volumetrics, tysm
Although you did test various engines for 2 years, you spent most of the video explaining why it is debatable, what should be taken into account and the choices you made. This complexity, humility and the inherent self-criticism is what makes this video brilliant.
Truly amazing work here!
One of the best renderer comparison videos! From my experience, I'd say Vray is the clear winner since it's been getting better in the latest versions and has improved it's weaknesses like volumetrics.
Corona's visual quality results are amazing out of the box but the IPR in Vray with direct manipulation of the scene in the viewport is a game changer for me.
Really intersting, cheers Andrew. I find Octane handles volumetrics way faster then Cycles out the box, bit surprised by those numbers.
Wonderful video. Thanks for all the work ♥
Wow, lot of work! Appreciate your efforts!
I would love to compare offline renders to something like Unigine or unreal, realtime engine come a long way.
Great video! Question: are the different render results available anywhere at high resolution to compare?
I was really interested in how different the results were visually and would love to see if my colleagues arrive at the same preferred result in terms of aesthetic. I personally thought the Corona results really shone, as you pointed out, but I'm curious if others arrive at the same conclusion. Would be great if you could make those available!
Thanks for doing this OBJECTIVELY!! Nice work
I am here to appreciate the two years of work. Thank you 😊👍
2 minute papers! Sick cameo! If you guys don't know, he does videos on the frontier of ai research including physics simulation, Ray tracing and robotics. Massively inspiring channel if you're interested in the computer science behind our favourite technology or just the technology itself. I rate this man so highly!
But hard to listen to his strange intonation
Viewing all the preview images you shown for SSS, Volumetrics and All-Rounder, I like Luxcore and Octane the best due to its detail in their rendering as well as their
overall coloring, shadows and details they show in their renderings. 😎👍
But oddly some rendering engines are good for some things and other engines are good for others. its a bit weird. lol 😁
Great video man!
superb, we need more of this videos, thanks for you effort
Would be nice if Andrew provided the scenes for artists of each engine to try optmizing it. (Only optimizations allowed)
We're you including the extra time taken for optix to start on the first render? Because once you've done one render it won't take as long
It's fine if you did/didn't include it, but I think it's an important thing to disclose
Well done and thank you for the job done
Fantastic video, incredibly comprehensive!
Notepad is the god of rendering
Terminal is much better. They even rendered shaded donut in it
@@omkarkhandekar6225 yeah 👍
@@gwynbleidd9000 you are right 👍
The Library of Babel probably includes the perfect code for a renderer, written in C, C++, assembly, etc., the challenge is finding the one without any bugs.
I think it’s worth mentioning as well that the E-cycles guy has a version called e-cycles 7. It’s considerably faster than cycles x apparently
Personally I want corona for blender though :( it’s the best looking render engine in my opinion
Excellent video! Great work. Thank you, thank you!
Absolutely fantastic video. Thanks very much!
Question: does the software matter? Like does vray work faster in Maya or Blender? Were those vray numbers averages from each software, or all rendered with just one?
No, in general from the DCC the scene gets exported to a native format (be it in HD or RAM) to be used by the engine, at that moment the scene is virtually the same no matter the DCC it came from.
@@Bone-studio Thanks!
Yes because if you try to render in maya it wil crash at least five times
This is something completely worth testing.
@@kobi399 exactly
At the moment, it might not be the best thing to say "Corona is great, I love Corona".
You should get it, will change your life.
Awesome! I've been wanting a video like this.
Very instructive and quite impressive congrats