On the depth of field - the F-stop value makes a difference too - I was still getting an unfocused section towards the part jutting in the camera direction, even after selecting the donut as the focus point. You'll note his value is around 30 (mine was very low for some reason - but good around that 20-30 range.)
Lower the F-stop more creamy the bokeh, more light is captured, but the focal point is narrower so you have to frame it correctly, i used 0.95 since i lke creamy bokeh, no problem.
To explain why. It is because your cameras focal length (practically the first camera setting) was larger than his, while the F-stop was the same. If you use a larger focal length you must compensate with a higher F-stop number to "maintain focus" - lower F-stop makes it, like Raul said, more creamy. Good portrait shots with blurred backgrounds are usually done using large focal lengths but low F-stop numbers (more open lenses), makes the background lights bigger and less diffused (google bokeh).
I didn't think I'd go through the Donut Tutorial again, but after watching 11 minutes of this, I've made up my mind to start it again! Great stuff, Andrew!
I think I've only ever done the original one, or may e it was the 2nd one, I dont recall now. That was because I had a small project that needed a donut. 😎
HELPPPP i have a problem when he says to set the geometry point at 14:29 and i do exactly like him. When I start my animation the base of my donut and the icing act like 2 separated corp and not more like one single, so the base start the animation but the icing doesn't follow the base an rotate weird. Somebody known the reason?
@@petecoleman3443 Of course not, but just using render layers to have different sample counts for different parts of one image vs. adaptive sampling doesn't make much of a difference. Haven't done real tests myself, but that's basically how adaptive sampling works.
@@Blemonade1 you can still get a lot of advantages through seperating fg and bg like chosing which aovs/ data passes you need for each and to have more control than what the adaptive will give you. It will generally be faster to adjust according to your needs than letting the algerythm take care of it:P but of course it can take a bit of time so more valueable in longer animations:)
i watched a 20 minutes and it didn't finished rendering, no tiles using and 500 samples, btw didn't use tiles cause it gives an error says( error writing tiles to file) any idea how to fix it
For those just starting out, I seriously recommend diving into Eevee and learning how to make it look nice. It's not as good as Cycles but that's not necessarily a bad thing, and once you accept that you can come to really appreciate what benefits it does have to offer. For people like me with a GTX 1050 on a laptop, it's the best option for animation no contest.
Ive got an AMD A10 7850k APU no gpu at all lol. The chip has 4 cpu cores and I want to say 12 gpu cores built in. When I built my pc I was mostly soung websites and light gaming so I didn't need graphics much. This was back in 2013 or 2015 though. I don't recall what the gpu market was back then. But yeh, we've does work the best. Though on occasion ive had some eevee scenes that took almost 5 minutes to render. Maybe due to bad setups, duplicate meshes etc though lol
I'm sorry to Contest you, but on my Computer Cycles is actually way faster than Eevee I recently did a Comparison between Eevee and the new Cycles Stuff on 3.0 Turned out, the more Complex the Scene is, the slower Eevee becomes + since 93 it became slower from Version to Version In the Test Eevee was actually 8 Seconds slower than Cycles, while looking considerably worse I don't know if I'm a God at optimising Cycles to Oblivion, but I have to, because my Computer is round about 8 Years old and otherwise that Stuff won'T work Maybe Eevee is just slow because of that - but yeah - Cycles faster than Eevee on my Computer so No Reason to use Eevee for me and I probably will never use it, because I go over Quality than Quantity anyway If LuxCore would be perfectly compatible with Blender, I would actually use LuxCore over Cycles while being Slower than Cycles, that's because it is much more Physically correct But yeah, in Animations you probably wanna go with the fastest one And for most People (Except you are a extreme Case like I am) this is gonna be Eevee
@@eskey6772 I feel like you've definitely got something weird going on. Cycles has gotten way faster, but I can't think of any scenario where it would actually be faster than Eevee.
@@eskey6772 You raise a very good point about scene complexity. I do plan on doing a fairly complex city harbor scene in the future, so I may see some advantage to Cycles there.
What a cliffhanger. We've finally come to a point when the donut is "finished" and eager to render it out then you drop the "But wait! Don't render it out yet, we'll do that in the next part...". Seriously though, thank you for the tutorials.
Here's a HUGE tip When rendering, only have blender open. Even having this tutorial open while rendering is a difference between 20 seconds and ~28 seconds!
The most important thing I have learned so far in this tutorial series has nothing to do with Blender, but is more about what to do and what not to do when creating a tutorial... Blender Guru has the art of creating tutorials down perfectly!
Impeccable timing, I just built a nice multi gpu system for heavier 3D tasks and I was wondering EXACTLY how to tackle rendering in 3.0. You are so loved by the community my dude.
i7-8700K 64GB RAM 1080ti Switching from CPU to GPU Compute made a cartoonish difference in render time. CPU: 30 sec/frame GPU: 5 sec/frame Thank you for keeping up this tutorial series.
If the Render Result pops up in a new Window instead of in the image editor, go to preferences > Interface > Editors > Temporary Editors and change the "Render In" field from New Window to Image Editor.
Great series Andrew! Bit more about clamping: It's actually not the higher you set it the faster it will render. It doesn't really affect render time much. Clamping affects the brightness values of the samples that are taken for each pixel. One nice pink pixel might have samples with brightness values generally around 0.4, 0.6 etc. but sometimes there's an outlier from a direct reflection ray or something that has a crazy value of like 20. With a low amount of samples, this can greatly increase the total brightness of that single pixel, because the final pixel brightness is determined by averaging over all samples. This can cause fireflies or high amounts of noise. However, if you use the clamp setting and set it to 10, you clamp off those outlier samples. So instead of a brightness of 20, it's only 10, which reduces the influence it has. The low samples (below 10) are not affected. This trick can reduce fireflies or certain types of noise at the cost of reducing accuracy. (Extreme amounts of clamping, values close to 0, will make all pixels darker) Happy holidays 😃
Same! I checked the box next to my GPU in the settings, and for all the previous parts I completely forgot to switch "CPU" under Render Properties to "GPU" xD
I have watched all of his begginner tutorials because first i got banned from using computers when i was in 10th grade and second time my computer litteraly broke the day after I finished his tutorial and was going to start doing somthing my self. I am so glad i am finaly doing it
if you're rendering on CPU as 50% of blender users are. The one setting to change is tile size, I ran a couple benchmarks on the BMW scene and with all settings tweaked for lossless quality renders I fount that the bigger the tile size the worse it gets, and you also can't go too low. My tile size sweetspot for a Ryzen 3700x 8c/16t was double the threadcount, and I also run the same benchmark on other PC systems and it was the same, you want your tile size to be double your threadcount. Since I found no tutorials on CPU rendering improvements I had to do figure it out myself. So that's my advice for CPU rendering
@@teslaromans1023 Tile size is the size of the square(-s) that render your picture. A GPU only uses one square, the CPU uses multiple squares at the same time (GPU is good at performing single tasks while the CPU is good at multitasking). You find the setting under "Render Properties" and then almost at the bottom "Performance" and then "Memory". Hope that helps.
Thanks, this really helped! But I wish I knew what I was doing instead of just blindly copying something I don't understand... 😅 Could someone here please explain what are these aperture settings and the f-stop value?
I Started using Blender a month ago, and here I am back here because consistently it's the best content on Blender I've found. He manages to cover most of the fundamentals all here. So freaking good. Thank you so freaking much brother. I never would have thought I could do this. Thank you for being my first mentor.
The difference between eevee and cycles is pretty much the (lack of) indirect lighting. Adding an Irradiance Volume and baking it brigns the two much closer together.
Glad someone mentioned the indirect lighting settings - the indirect lighting settings helps bake that into the scene. Note that it currently calculates it for one frame, so it's not going to calculate it per frame (because that's pretty much Cycles), however I did see a Blender Market addon that animated it. To add onto your differences, also Cycles relies on the whole scene for calculations while Eevee tries to use only screen space - only what's actually seen by the camera (indirect lighting gets around this)
@@SpencerMagnusson Oh it does it only for one frame? I guess that makes it pretty useless for this case (unless you have that addon). I didn't even think about that, since I never animate stuff 😂. I also found another difference, eevee doesn't have displacement.
This is a good solution- for now. Eevee’s gonna be getting screen space global illumination in future versions so it wouldn’t be necessary to bake. But until then, bake.
@@ManuelRusch If the lights don't move, then irradiance in Eevee works fine for animations. As soon as lighting changes (sun time laps, turning off a lamp etc.) then yeah Eevee needs to recalculate. I kinda ignore Eevee though, it's still far from Unreal Engine or other game engines and optimizing lighting takes way more than using Cycles with like 40-100 samples. Intel Denoiser does an excellent job at crazily low samples, for animations you can get away with 200-400 samples depending on optimizations and lighting, so I'm fine with it even though I only use CPU for rendering.
I have a single RTX 3060 TI in my system and following everything he said I got my render (for 1 frame) down to 7 seconds. I used 4096 samples with 0.05 noise threshold. Thx for the donut btw
My PC renders a frame in 9 seconds(CUDA, 200 samples, threshold of .05, and the other adjustments shown in the video, with the Optix Denoiser) with my GTX 1650 GDDR6. Thank you so much for this tutorial series!
So happy that I did this tutorial as my first - learned so much about everything to do with Blender!!! Now working on more advanced tutorials and my own projects - great video on cutting down rendering - saved me about 20 minutes on my last render of a full scene with multiple objects.
19:55 If your Blender is crashing when you switch from Cycles to Eevee, try switching your viewport from the 'rendered' view to the 'solid' view up at the top right prior to making that switch. Had my blender crash on me a few times during this series and I think that this is the culprit. SAVE OFTEN!!! And if anyone else has suggestions please share!
Before using your recommended settings you pointed out in this episode. My Cycles full render was 3.5 minutes. And after applying all settings, it went down to a whooping 11 seconds! TYVM! Im absolutely amazed by the huge difference! (Im running on an i7 12gen, with an NVidia RTX 3060, and 64gb RAM)
Strange, I would think a 2070 Super would be mostly on par with a 3060, yet my render times (mimicking blenderguru's options) are 7 and a half seconds. In comparison I have 16 gb ram and i7 10700, so you'd think these times would be worse than your system, right?
I was really concerned that my render time was like 20+ seconds per frame on cycles, but you mentioned switching to my GPU in the render properties and it JUMPED to like 2-5 seconds per frame with some of the render time saving tricks. Thank you so much for mentioning that again this video! I have an NVIDIA Geforce RTX 3070 Laptop and I'm getting like 5-7 seconds even without denoise or threshold and my samples at 1000.
I know this is old, but how? I have a 3080ti and it's taking like 40 seconds to render a frame. They were going much faster at the beginning of the tutorial. I think it fell way down during compositing.
Holy shit, i just realized I never changed it in the properties either. I thought i had but that seems tovnot be the case. It's still like 6 seconds, so something in my settings is off lol but it's waaaay better than it was.
My current setup is: AMD Ryzen 9 5900HX, 32Mb RAM NVIDIA GeForce RTX 3070 Laptop GPU I stuck pretty closely to the tutorial with the exception of icing color, and I didn't include the bluish candy balls - just the sprinkles. Rendering a single donut frame in Cycles X at: -4096 samples -0.05 noise threshold -OpenImage denoiser -Compositor enabled -Subsurf scatter off -Motion blur and depth of field enabled =5.01 seconds. I'm pretty much thrilled with that render time. I have dabbled in Blender for the last decade or so, and the thing that has always impeded my progress the most (other than lack of natural artistic ability) is having a machine that can render fast enough to not be frustrating. All of the Blender enhancements, plus having a GPU that can acceptably keep up with a viewport set to rendered view when needed is really a game changer.
Lol, I love the "seconds" that get mentioned. I'm learning that my iMac is not really good for this program. In another render earlier in the series, I calculated a 4 second render you did took my CPU 1min 30sec. FML. so saving 1 second is huge on my render times lol. Love the series and the result so far. Learning a lot.
Since you asked about render times and hardware, I can get a frame in about 8-9 seconds after turning off the subsurface, Noise Threshold .05, Max Samples 1080, OpenImage Denoise on, and with persistent data off. With subsurface, it was more like 16-18 seconds, biggest difference in the render result was that the edges of the icing in shadow had a faint glow with the subsurface on. I did make a few changes to the models themselves, although I imagine the impact was minimal. I left out the hard candy, but modeled some block text above and below the donut to make it look like an ad and threw in some spotlights to improve the lighting on the text without affecting the donut's lighting. I'm running a single NVIDIA RTX 2060 for GPU, a Ryzen 5 2600 CPU, and have 16GB of RAM. So better than a laptop, solid for a gaming rig, and noticeably weaker than what's being used for the tutorial.
I'm a newbie beginner on Blender and I use NVIDIA GeForce GTX 760(Display). Its a pretty old card. Thank you for this lesson. Love all the tutorial contents that you've made.
What about the Denoise node in the compositor? I've been tell that it was better to do it via the compositor (with the denoise data pass) rather than using the one from the property setting. I'm curious since you hasn't mentionned it.
I haven't done any tests since my pc is old, from around 2015 and uses a hard drive from a laptop that's even older...but as far as render time I my setup goes, using the compositor and the denoise I properties goes, it's about the same. I might try noise threshold and such though, haven't really given that a go yet.
The denoise node in the compositor is OpenImage denoise, but as a compositing node. It's better since you still have the original noisy image in case you do something else later(like switch denoisers)
Makes zero difference on my test render scene, either time wise or quality wise. Some people have reported differences but I haven't been able to replicate it on my Linux RTX3090 rig.
That persistent data shaves off 1/6 of my render time!! That will be a solid 6hours less render than what my current project normally would have been. Appreciate it!!:)
19:54: Eevee Rendering Setup 20:41: Render, Shadows - Cube Size: For ALL the other lamps - Cascade Size: Sun lamps (not important, in this case) ---------------- 21:50: Light -> Object Data Properties, Shadow - Clip Start: - Bias: ---------------- 22:28: Contact Shadows - For Sprinkles - Check, in order for sprinkles to cast shadow onto the icing - Turn *Bias* down, "all the way down to zero or 0.001" 23:28 Contact Shadows - are a fakery ---------------- 25:59: Ambient Occlusion
Few years ago I still had a 5850 1gb from 2010 and I wasn't able to do anything in blender, eventhough Iwanted it so bad. Now I got a decent gpu, and finally I was able to complete your tutorial. Have a good day sir! wish you the best from Türkiye
25:30 the way i use ambient occlusion is by using the actual ambient occlusion NODE! that way you can see what its doing very well. and that node also only works if ambient occlusion is turned on.
My Laptop PC Specs: CPU: Intel i7 10th GEN GPU: NVIDIA MX330 (2 GB) + Intel Iris Plus G7 Graphics (8 GB) RAM: 16 GB DDR4 ROM: 512 GB SSD I found these settings working best for my particular render: Noise Threshold: 0.05 Samples: 1024 Rest sample settings were left to default. I could render 1 frame in near around 1 minute, with Denoising on. Render Time: 00:1:00
GTX1060 = 1min 16seconds. I did the render with the same settings after following up till when you asked. Not too concerned with the time it takes to render, yet. Especially after how long the whole thing has taken me. Noice tutorial Guru, cant wait to start playing around this now. I have a stupid amount of notes from your lessons, hoping ill be able to get through parts of it on my own object without going back to them too much though I'm glad to have them. Cheers
Before this part my renders of the donut in cycles was around 3min but now they are 13sec with GTX1650 and 4096 samples. Thank you so much for this graet series!!
For anyone struggling with really long render times on cycles. You can keep the sample count at 100 with Denoise and it still looks good. I know he said there's a flickering issue but honestly it's not incredibly noticeable or distracting Edit: With low Denoise. Like 0.01
Great series! It's my first time using Blender and with this tutorial I've learned so so much and it's been easy to understand thanks to you. I have a GTX 1050 Ti (and I'm also using the Graphics of my Ryzen 5 2400G) and I'm getting renders of around 40 seconds using the adjustments shown in the video.
I've got RTX 3060 mobile (laptop version), core i7 11800, and 16gb RAM, and with pretty much same settings my laptop needs around 8 seconds to render a frame. Huge thanks to you, Andrew, for all tips but especially for choosing the GPU as the renderer hardware!
19:26 You asked for my specs, you shall receive them (If I missed anything important, let me know and I'll add it): GPU: GTX 1660 CPU: Ryzen 5 3600 RAM: 32GB (I don't remember if it's GDDR3 or 4; I assume 4) 1440p 21:9 monitor (could impact how much the card is being used by windows) As for render times, it took 16 seconds at default settings w/ optix @ 4096 samples (model is 77,056 Tris, w/ metallic and roughness reflections; the sword on my instagram if you're wondering which model specifically)
It took me a whopping 2.36 minutes to render a single frame on cycles with the same settings as yours! btw my specs: GPU: none 😔 CPU: AMD 4500U ram: 16Gb it is a laptop(Lenovo Flex 5 AMD version)
For everyone to reference my hardware and render time: Razer Blade 14 Laptop with built in Nvidia GeForce RTX 3080 GPU Noise threshold: .05 Max Samples: 4096 Render Time: 9 seconds for Render Image with Cycles, 41 minutes (~7s per frame) for Render Animation in EXR files I also use my laptop for gaming (often heavily modded), photo and video editing, and audio recording/song production. I have it connected to a monitor. I love this beast. Its so powerful and has been working perfectly since i got it ~1.5-2 years ago. Best tech purchase Ive ever made tbh. If anyone has any questions about my setup or what it cost or anything feel free to ask :)
now with 3.0, my average render time is somewhere around 5-10 minutes. before it was about 20 minutes on a 1070 with cuda. settings are 6mp, 512 samples, intel denoise, subsurf scattering, in a room with around 3 walls, dof. bounces are default 12. i tend to forget disabling fast gi aprox,.however even if its on 2, it still does a good job imo.
I know some people have commented on the motion blue f-stop thing but I just wanted to give my two cents as a photographer: Depth of field controls the distortion between your foreground and background. A low or shallow dof will compress the composition while a high dof will create a wide lens effect. The F-stop is what determines how much of your composition is in focus. A low f-stop will allow you to focus on a small section, either foreground, mid, or background, while a high f-stop will put the entire composition in focus. In a camera, the dof you achieve is a result of the f-stop you choose paired with the distance in mm your lens is set to. So the distance of your camera to the subject will also affect both of these outcomes. A lower distance from the subject to the camera paired with a high distance from the subject to the background with a low f-stop will result in the highest depth of field, greatest compression, and strongest bokeh affect.
I was using my CPU to render THIS ENTIRE SERIES, and watching you toggle GPU Compute, then learning that I could as well (I have a GTX 1060) made me feel SO goofy. I sat and watched it render for 18 minutes! My final render was 22 seconds.
suggestion : When you explained what "passes" are, you could have used the option of the "rendered" view in the viewport to show (in real time, in your viewport) only the passes that you select.
Quick reminder for the Eevee settings: make sure to select the right lamp when you’re checking the contact shadows. I just wondered why it’s not working and I was at the wrong lamp settings.
I have some free time over Christmas as I am visiting the family and decided to learn Blender, so far have been following along and really enjoy this tutorial series! Unfortunately I decided my shitty MacBook Air would be enough to keep me entertained so my render times per frame with just 500 samples are around 2 - 3 minutes depending on my noise threshold! This may take some time to render. Can't wait to get back to my main rig to really play with this! Thank You!!!!
If you get blur on your donut while setting the depth of field, except increasing the number of F-stop, there's another thing I do: I add an empty mesh and put it a little front of the donut and then set my focus on it. It fixes the problem, too. I think compared to take a picture in physical world, it just moves the focus point a little front with the empty mesh to anchor it. You could also change the F-stop at the same time for better output.
Thanks for making this tutorial series, it's doing a great job at teaching me the basics of Blender and I'm learning a lot of info from each section. Using Cycles with all of the render time reduction methods in the video, I was able to achieve a frame in just short of 7 seconds, which was a large improvement from before. GPU: RTX 3050ti (yes, laptop) CPU: Ryzen 7 5800h RAM: 16gb
With very similar settings To Andrew, my render time was 6.93 on Cycles. My hardware is: AMD Ryzen 9 5900X (All Core OC 4.4Ghz) and Nvidia 3090 FE - and thank you so much for this awesome video series! My doughnut looks so edible!!
I'm glad you reminded me about the GPU... I was rendering on my CPU and it took 6 minutes for some of my tests vs 48 seconds on the GPU. Remember your GPU, guys!
If your render doesnt show the pink background, go to output properties - post processing - then turn on compositing, i dont know why but that worked for me
Also, if that doesn't seem to work, it may be because you haven't waited for the render to complete. I know because I just spent 10 minutes trying to figure out why it wasn't working for no reason... sigh
Increase the F-Stop if you find that your work is so blurry. In my case, it's f/30 because the Donut is really close to the camera. This is same case using the real camera.
My god I'm glad your still around man, I remember when you taught me how to make my first donut In 2.8. That led me to realized mechanically how I could utilize blender to make anything just by your tutorial of a donut.
With the final render settings as per this tutorial I'm getting a render at about 22 seconds, I'm running Linux and have a GTX 1080 and an AMD Ryzen 7. I noticed that Andrew didn't have his processor clicked in the settings and just both his graphics cards, weirdly when I unchecked using the CPU the render time went down to 19 seconds, I would have thought if using the CPU + GPU would be faster.
My guess is tha maybe some extra communication between cpu and gpu is causing this, or having to load the data to ram and Vram, but i am by no means an expert on the topic so I wouldn't know
With a Cycles GPU render using HiPS on an AMD Radeon RX 6900 XT I'm leaving subsurface on for the color depth ( looks good with chocolate icing ) With subsurface and all other settings basically the same: Noise Threshold of 0.05 renders in 8.9 secs, noise threshold of 0.04 looks a bit cleaner and renders in 11.19 secs. Thanks for another great video series!
Cycles with all the things you said turned on, subsurface scattering OFF, 8192 sampling, 0.1 noise threshold, intel denoiser took about 6-7 seconds per frame average on RTX 2060 with Ryzen 3700X I also tried to reduce the noise threshold to 0.01 and it went over a minute, but with 1080 x 1440 resolution I can't tell any difference.
I'm running the same system and I'm currently averaging 15sec per frame, the only difference is my noise threshold is set to .03. Do you think that would cause the difference in times?
I finish for the first time the donut. Now ill do it again. I just wanted to say that for sprinkles, it help me a lot the old tutorial, cause with nodes it get frustrating in the begging.
Alright, when I activate depth of field, use the donut as the focus object and turn up the F-stop value the donut looks fine. The sprinkles however are so out of focus that they are sort of half transparent and after compositing they are almost invisible... to begin with, my scene looks way more out of focus than Andrew's when I first check the dof box and haven't selected the focus object. Could it be because the distance between my camera and objects is different to Andrew's? In my project the donut is almost inside the camera for it to take that much space of the frame. I went back a little and the last time I saw his camera, it was further back. However I remember him adjusting the zoom while in camera view, which means his camera is probably just as close as mine... so I still have no idea as to why my image is so much more out of focus ._. edit: as it turnes out, the sprinkles in the back were just way to small, so that them being out of focus and having this blurred outline made them barely visible. I increased their scale in the "instance on points" node and they looked way better and I also realized that in the final render in this video the sprinkles aren't entirely visible as well, I don't know why I thought they had to be
under where u turn on depth of field in the cameras settings i turned up the F-stop up and it brought more of it into focus and i could see the sprinkles in the back a little better. you might have to find what works best for urs
I did the first donut series on my laptop a couple years back. Standard Microsoft surface book 2, 1070ti. Took me about 7-8min per frame with like the lowest of low in every setting. Redoing the donut, but on my new rig. Amd 7950x with a 4070ti. Now its 4-6 seconds per frame. Mind is blown watching this thing crank this out.
I just started blender yesterday and the tutorials have really helped. I've been following along very closely, but my renders look much worse than yours even though we are using the same resolution and number of samples. There is much more aliasing on the sprinkles and I'm not sure why. Shouldn't the results be pretty much the same based on sample count and resolution alone? I am currently rendering on a 1070 using Optix.
turn on fast GI approximation under caustics in light paths and bump that up to 500 for viewport and render! I am one of the most experienced artist and using fast GI approximation for a long time so use it! I hope this helped!
For anyone who is interested in the Why: a brief history of framerates lets start with 24 When the first cinematic camera was invented it was trial and error to find the frequency our brain needs to percieve single pictures as motion. Since film was expensive and the cameras had a hard time keeping up with that many pictures they settled with the minimum amount, the magic number of 24. That is a lot less than our brain could handle wich is evident when you look at car tires in movies. The rotation of the tire is much faster than 24 fps. From one frame to the next it is able to make almost a full rotation but not quite. Thus the car tires seems to rotate bakwards in a fast going car. This obviously doesnt happen in real life, so our brain is capable of percieving enough pictures per second to keep up with the tire. So 48 is actually closer to our real life perception. And it is very usefull for 3D. In 3D we see two pictures per frame so we have double the information to take in. 48 frames gives the brain double the information to piece the full motion together. Now why 25 or 30? The answer lies in the alternating current from the power soccets. In Europe the frequenzy is 50 Hertz, in Amerika it is 60 (no idea why). That means the lamps in the TV flicker in that frequency and the TV that needed cable used that too. Now for the first TVs it was a lot to process this much information at once so they used an interlaced signal. Meaning only half the picture information is used in one frame and the other half comes in the next frame. So the TV signal switched frames 50 or 60 times a seconds with each frame holding half the information. So 25 or 30 full pictures. This is still TV standard today even though modern displays have no problem processing pogressive signals (full picture information per frame) in different frame rates. But the TV signal always has to be recievable with the oldest Tec still in use, so everyone can recieve it. How can you see the frequency difference for yourself? LED are amercan standard and usually flicker with 60 Hertz. So if you try to film an LED with a shutter speed of 50 you will see the lights flicker. This is actually a big problem in europe because the normal lamps flicker with 50 Hertz. So If you have light from an LED and a normal lamp in the same scene, one of them will always flicker no matter how you adjust the shutter speed. Last fun fact: This is also the reason why american DVDs dont work in europe and vice versa. The laser of the player flickers in the frequency of the alternating current. If the information on the DVD is written for a different frequency the laser cant read it anymore.
for me it will take about 35 hours to render 300 frames on cycles on my very bad laptop but!!!!!!! before this tutorial it would have taken like 5 days to render everything EDIT: this time is on a 100 samples it would take 4 hours on one frame if it was 4096 samples
for my render cycles - RTX 4060 laptop GPU noise threshold - 0.05 max samples - 8192 time taken = 8.13 seconds (single frame) same settings but cycles - i7 12650h time taken = 19.16 seconds (single frame)
thanks for the help I actually did some test renders before I got to this video and the render time difference is definitely a game changer once I made the changes. render time : 12:87 seconds GPU: RTX 3080
Again, kudos to all who made it this far and many thanks to Blender Guru! I can only use CPU and, with Cycles, it currently takes 59s to render with Noise Threshold at 0.05, Max Samples 200, Denoise on, Motion Blur on and set to 0.5, DoF turned off. With the same setting but Max Samples ramped up to 300, it takes 1min 7s, so will need to keep samples quite low.
Oh holy moly. Did image rendering w/out a compositor after adding a few of the tips and I was looking at a image render time of 00:00:99! Graphics have really come a long way.
QUICK TIP: If you are on Blender 4.0+ and want to export frames in Agx, you can follow the same steps that Andrew does in Filmic, just tried it and it works the exact same :)
For anyone interested in depth of field, some camera research is really good. The F/Stop bit is, in functional terms, how long the 'in focus' section is. A few comments have mentioned it, but I recommend using real camera numbers for it. You can look up an F/Stop chart if you like, but the basics are 1.4, 2, 2.8, 4, 5.6 and so on.. Just if you want to get super accurate focal lengths for real film. (Note, this is more for technical perfection, realistically just use whatever number looks good)
![Lp. Сердце Вселенной #60 РОЖДЕНИЕ ЛОЛОЛОШКИ [Финал] • Майнкрафт](http://i.ytimg.com/vi/YoR0pAV9FVQ/mqdefault.jpg)
Personally I save all of my renders for winter and use my computer as a furnace to heat my whole house
Exactly I love getting 95 degrees (not Fahrenheit) on my cpu.
🤣🤣🤣🤣🤣🤣🤣🤣
@@5finger_forehead I love when it throttles
🤣
my dell laptop is burning hot even when I'm not rendering so i get free heating all the time
If you have a blurry Donut after applying depth of field, copy his Aperture settings at 13:52
thanks man
Thanks!
Thanks !!
Thanks man
LIFE SAVEERRRR
I'm convinced people who have used blender for as long as Andrew still watch his videos to learn more.
The team that created blender asked him for some tutorials on how to develop it :)
or just watch them cause they're bored
facts, though not to learn more, more like a potcast.
@@lilgamedev who better to ask about a software than the people who use it the most, right?
@@devoid-of-life true
On the depth of field - the F-stop value makes a difference too - I was still getting an unfocused section towards the part jutting in the camera direction, even after selecting the donut as the focus point. You'll note his value is around 30 (mine was very low for some reason - but good around that 20-30 range.)
Lower the F-stop more creamy the bokeh, more light is captured, but the focal point is narrower so you have to frame it correctly, i used 0.95 since i lke creamy bokeh, no problem.
To explain why. It is because your cameras focal length (practically the first camera setting) was larger than his, while the F-stop was the same. If you use a larger focal length you must compensate with a higher F-stop number to "maintain focus" - lower F-stop makes it, like Raul said, more creamy. Good portrait shots with blurred backgrounds are usually done using large focal lengths but low F-stop numbers (more open lenses), makes the background lights bigger and less diffused (google bokeh).
I love you
hii, do u use blender? i'm recruiting people to my team
Thanks for this! I couldn't figure out why my donut was blurry, this solved it
I didn't think I'd go through the Donut Tutorial again, but after watching 11 minutes of this, I've made up my mind to start it again! Great stuff, Andrew!
I think I've only ever done the original one, or may e it was the 2nd one, I dont recall now. That was because I had a small project that needed a donut. 😎
The last time I went through it (just 4 months ago?) I just wanted to make the donut. This time I'm going to actually take notes 😆😆😆
HELPPPP i have a problem when he says to set the geometry point at 14:29 and i do exactly like him. When I start my animation the base of my donut and the icing act like 2 separated corp and not more like one single, so the base start the animation but the icing doesn't follow the base an rotate weird.
Somebody known the reason?
@@KillWoRld you need to set the Donut as the PARENT of the icing. He discusses this at the start of the animation tutorial (part 11)
@@choenriquez thanks
Secret render tip:
For large/complex scenes use render layers to use different sample counts for the foreground and background. :)
That' what the adaptive sampling does, and in cycles x, its enabled by default. so since blender 3, that doesn't make a big difference anymore.
@@Blemonade1 render layers aren't the same as noise threshold....
@@petecoleman3443 Of course not, but just using render layers to have different sample counts for different parts of one image vs. adaptive sampling doesn't make much of a difference. Haven't done real tests myself, but that's basically how adaptive sampling works.
@@Blemonade1 you can still get a lot of advantages through seperating fg and bg like chosing which aovs/ data passes you need for each and to have more control than what the adaptive will give you. It will generally be faster to adjust according to your needs than letting the algerythm take care of it:P but of course it can take a bit of time so more valueable in longer animations:)
i watched a 20 minutes and it didn't finished rendering, no tiles using and 500 samples,
btw didn't use tiles cause it gives an error says( error writing tiles to file) any idea how to fix it
For those just starting out, I seriously recommend diving into Eevee and learning how to make it look nice. It's not as good as Cycles but that's not necessarily a bad thing, and once you accept that you can come to really appreciate what benefits it does have to offer. For people like me with a GTX 1050 on a laptop, it's the best option for animation no contest.
Haha, We have the exact same GPU.
Ive got an AMD A10 7850k APU no gpu at all lol. The chip has 4 cpu cores and I want to say 12 gpu cores built in. When I built my pc I was mostly soung websites and light gaming so I didn't need graphics much. This was back in 2013 or 2015 though. I don't recall what the gpu market was back then.
But yeh, we've does work the best. Though on occasion ive had some eevee scenes that took almost 5 minutes to render. Maybe due to bad setups, duplicate meshes etc though lol
I'm sorry to Contest you, but on my Computer Cycles is actually way faster than Eevee
I recently did a Comparison between Eevee and the new Cycles Stuff on 3.0
Turned out, the more Complex the Scene is, the slower Eevee becomes + since 93 it became slower from Version to Version
In the Test Eevee was actually 8 Seconds slower than Cycles, while looking considerably worse
I don't know if I'm a God at optimising Cycles to Oblivion, but I have to, because my Computer is round about 8 Years old and otherwise that Stuff won'T work
Maybe Eevee is just slow because of that - but yeah - Cycles faster than Eevee on my Computer so
No Reason to use Eevee for me and I probably will never use it, because I go over Quality than Quantity anyway
If LuxCore would be perfectly compatible with Blender, I would actually use LuxCore over Cycles while being Slower than Cycles, that's because it is much more Physically correct
But yeah, in Animations you probably wanna go with the fastest one
And for most People (Except you are a extreme Case like I am) this is gonna be Eevee
@@eskey6772 I feel like you've definitely got something weird going on. Cycles has gotten way faster, but I can't think of any scenario where it would actually be faster than Eevee.
@@eskey6772 You raise a very good point about scene complexity. I do plan on doing a fairly complex city harbor scene in the future, so I may see some advantage to Cycles there.
What a cliffhanger. We've finally come to a point when the donut is "finished" and eager to render it out then you drop the "But wait! Don't render it out yet, we'll do that in the next part...". Seriously though, thank you for the tutorials.
Here's a HUGE tip
When rendering, only have blender open.
Even having this tutorial open while rendering is a difference between 20 seconds and ~28 seconds!
yes, especially firefox ...
@Samoerai Correct me if I'm wrong, but doesn't loading objects before rendering use CPU & RAM? So it might have an impact on the first frame
I rendered 29 with this open, then 28 with closed Firefox. So.. not much impact.
True, I usually listen to UA-cam while working in blender but when it's time to render I switch to my phone lol
The most important thing I have learned so far in this tutorial series has nothing to do with Blender, but is more about what to do and what not to do when creating a tutorial... Blender Guru has the art of creating tutorials down perfectly!
Impeccable timing, I just built a nice multi gpu system for heavier 3D tasks and I was wondering EXACTLY how to tackle rendering in 3.0. You are so loved by the community my dude.
i7-8700K
64GB RAM
1080ti
Switching from CPU to GPU Compute made a cartoonish difference in render time.
CPU: 30 sec/frame
GPU: 5 sec/frame
Thank you for keeping up this tutorial series.
RTX 3070Ti took 11secs/frame on exact your settings. THANK YOU so much for this tutorial! Best teacher!
If the Render Result pops up in a new Window instead of in the image editor, go to preferences > Interface > Editors > Temporary Editors and change the "Render In" field from New Window to Image Editor.
how would i revert this back to normal, if i saved it on the new setting?
@@Crystal-op9qb change that same settings back to "New Window"
Great series Andrew!
Bit more about clamping: It's actually not the higher you set it the faster it will render. It doesn't really affect render time much.
Clamping affects the brightness values of the samples that are taken for each pixel.
One nice pink pixel might have samples with brightness values generally around 0.4, 0.6 etc. but sometimes there's an outlier from a direct reflection ray or something that has a crazy value of like 20.
With a low amount of samples, this can greatly increase the total brightness of that single pixel, because the final pixel brightness is determined by averaging over all samples. This can cause fireflies or high amounts of noise.
However, if you use the clamp setting and set it to 10, you clamp off those outlier samples. So instead of a brightness of 20, it's only 10, which reduces the influence it has. The low samples (below 10) are not affected. This trick can reduce fireflies or certain types of noise at the cost of reducing accuracy. (Extreme amounts of clamping, values close to 0, will make all pixels darker)
Happy holidays 😃
the GPU tip saved me so much time! It was about 50 seconds for a render, now it is under 8 seconds! I am using the Aplle M1 Pro (GPU - 16 cores)
My CPU is actually better than my GPU... because I don't have one
Same! I checked the box next to my GPU in the settings, and for all the previous parts I completely forgot to switch "CPU" under Render Properties to "GPU" xD
I have watched all of his begginner tutorials because
first i got banned from using computers when i was in 10th grade
and second time my computer litteraly broke the day after I finished his tutorial and was going to start doing somthing my self.
I am so glad i am finaly doing it
if you're rendering on CPU as 50% of blender users are. The one setting to change is tile size, I ran a couple benchmarks on the BMW scene and with all settings tweaked for lossless quality renders I fount that the bigger the tile size the worse it gets, and you also can't go too low. My tile size sweetspot for a Ryzen 3700x 8c/16t was double the threadcount, and I also run the same benchmark on other PC systems and it was the same, you want your tile size to be double your threadcount. Since I found no tutorials on CPU rendering improvements I had to do figure it out myself. So that's my advice for CPU rendering
Awesome. Now what is a tile size and where do you change it lol
@@teslaromans1023 Tile size is the size of the square(-s) that render your picture. A GPU only uses one square, the CPU uses multiple squares at the same time (GPU is good at performing single tasks while the CPU is good at multitasking). You find the setting under "Render Properties" and then almost at the bottom "Performance" and then "Memory". Hope that helps.
If your donut is still blurry after applying depth of field, make sure your f-stop value underneath is higher
Great comment. Solve my problem immediately.
Thank you so much!!
u saved me
Thanks, this really helped! But I wish I knew what I was doing instead of just blindly copying something I don't understand... 😅
Could someone here please explain what are these aperture settings and the f-stop value?
as a photographer, seeing his f stop at 30 almost lead me into shock
I Started using Blender a month ago, and here I am back here because consistently it's the best content on Blender I've found. He manages to cover most of the fundamentals all here. So freaking good. Thank you so freaking much brother. I never would have thought I could do this. Thank you for being my first mentor.
The difference between eevee and cycles is pretty much the (lack of) indirect lighting. Adding an Irradiance Volume and baking it brigns the two much closer together.
I didn't know that existed, thank you!
Glad someone mentioned the indirect lighting settings - the indirect lighting settings helps bake that into the scene. Note that it currently calculates it for one frame, so it's not going to calculate it per frame (because that's pretty much Cycles), however I did see a Blender Market addon that animated it.
To add onto your differences, also Cycles relies on the whole scene for calculations while Eevee tries to use only screen space - only what's actually seen by the camera (indirect lighting gets around this)
@@SpencerMagnusson Oh it does it only for one frame? I guess that makes it pretty useless for this case (unless you have that addon). I didn't even think about that, since I never animate stuff 😂. I also found another difference, eevee doesn't have displacement.
This is a good solution- for now. Eevee’s gonna be getting screen space global illumination in future versions so it wouldn’t be necessary to bake. But until then, bake.
@@ManuelRusch If the lights don't move, then irradiance in Eevee works fine for animations. As soon as lighting changes (sun time laps, turning off a lamp etc.) then yeah Eevee needs to recalculate. I kinda ignore Eevee though, it's still far from Unreal Engine or other game engines and optimizing lighting takes way more than using Cycles with like 40-100 samples. Intel Denoiser does an excellent job at crazily low samples, for animations you can get away with 200-400 samples depending on optimizations and lighting, so I'm fine with it even though I only use CPU for rendering.
I have a single RTX 3060 TI in my system and following everything he said I got my render (for 1 frame) down to 7 seconds. I used 4096 samples with 0.05 noise threshold. Thx for the donut btw
Your vids are the reason I switched from Maya to Blender. Love the content and thank you for all the tutorials!
Oh now I'm curious why you gave up on Maya, it seems to be a standard software out there. By now at least
@@randomike__ Maya's got too much baggage. Used and abused by too many.
Same for me but with 3ds max
My PC renders a frame in 9 seconds(CUDA, 200 samples, threshold of .05, and the other adjustments shown in the video, with the Optix Denoiser) with my GTX 1650 GDDR6. Thank you so much for this tutorial series!
I have 1650 ti, and it takes 20 seconds! wtf
@@kalpit22 i have a core i7 and takes 8 minutes
I have a mac, it takes like 1 hour
i have a pentium 4 and it took me half an hour to open youtube 🤣🤣
It's here! Can't wait to watch it.
So happy that I did this tutorial as my first - learned so much about everything to do with Blender!!! Now working on more advanced tutorials and my own projects - great video on cutting down rendering - saved me about 20 minutes on my last render of a full scene with multiple objects.
man blender guru your vids are so helpful and your easily the best blender and 3d artist youtuber out there! keep up the good work!!💗
19:55 If your Blender is crashing when you switch from Cycles to Eevee, try switching your viewport from the 'rendered' view to the 'solid' view up at the top right prior to making that switch. Had my blender crash on me a few times during this series and I think that this is the culprit. SAVE OFTEN!!! And if anyone else has suggestions please share!
Hey my render isn't displaying the rbg we added in the composting any advice on how to fix that?
Your last donut tutorial is what got me started on blender so it’s rlly fun to see it grow
Me too
Before using your recommended settings you pointed out in this episode. My Cycles full render was 3.5 minutes. And after applying all settings, it went down to a whooping 11 seconds!
TYVM! Im absolutely amazed by the huge difference!
(Im running on an i7 12gen, with an NVidia RTX 3060, and 64gb RAM)
Strange, I would think a 2070 Super would be mostly on par with a 3060, yet my render times (mimicking blenderguru's options) are 7 and a half seconds. In comparison I have 16 gb ram and i7 10700, so you'd think these times would be worse than your system, right?
I was really concerned that my render time was like 20+ seconds per frame on cycles, but you mentioned switching to my GPU in the render properties and it JUMPED to like 2-5 seconds per frame with some of the render time saving tricks. Thank you so much for mentioning that again this video!
I have an NVIDIA Geforce RTX 3070 Laptop and I'm getting like 5-7 seconds even without denoise or threshold and my samples at 1000.
I know this is old, but how? I have a 3080ti and it's taking like 40 seconds to render a frame. They were going much faster at the beginning of the tutorial. I think it fell way down during compositing.
Holy shit, i just realized I never changed it in the properties either. I thought i had but that seems tovnot be the case. It's still like 6 seconds, so something in my settings is off lol but it's waaaay better than it was.
Dear Andrew, I am new to Blender and I am really sticking to it because you make it totally easy to understand the processes. Thank you!!!!!!
My current setup is:
AMD Ryzen 9 5900HX, 32Mb RAM
NVIDIA GeForce RTX 3070 Laptop GPU
I stuck pretty closely to the tutorial with the exception of icing color, and I didn't include the bluish candy balls - just the sprinkles.
Rendering a single donut frame in Cycles X at:
-4096 samples
-0.05 noise threshold
-OpenImage denoiser
-Compositor enabled
-Subsurf scatter off
-Motion blur and depth of field enabled
=5.01 seconds.
I'm pretty much thrilled with that render time. I have dabbled in Blender for the last decade or so, and the thing that has always impeded my progress the most (other than lack of natural artistic ability) is having a machine that can render fast enough to not be frustrating.
All of the Blender enhancements, plus having a GPU that can acceptably keep up with a viewport set to rendered view when needed is really a game changer.
Lol, I love the "seconds" that get mentioned. I'm learning that my iMac is not really good for this program. In another render earlier in the series, I calculated a 4 second render you did took my CPU 1min 30sec. FML. so saving 1 second is huge on my render times lol. Love the series and the result so far. Learning a lot.
Haha, on a mac air and his 4 sec renders are 6 and a half minutes for me
@@imagromlin LMAO
I stared using Blender by watching the original donut tutorial almost 3 year ago and still I'm learning something new by this one!
Since you asked about render times and hardware, I can get a frame in about 8-9 seconds after turning off the subsurface, Noise Threshold .05, Max Samples 1080, OpenImage Denoise on, and with persistent data off. With subsurface, it was more like 16-18 seconds, biggest difference in the render result was that the edges of the icing in shadow had a faint glow with the subsurface on. I did make a few changes to the models themselves, although I imagine the impact was minimal. I left out the hard candy, but modeled some block text above and below the donut to make it look like an ad and threw in some spotlights to improve the lighting on the text without affecting the donut's lighting.
I'm running a single NVIDIA RTX 2060 for GPU, a Ryzen 5 2600 CPU, and have 16GB of RAM. So better than a laptop, solid for a gaming rig, and noticeably weaker than what's being used for the tutorial.
Woah man. The rendering tricks are so good now i render a rocket animation from 2.2 hours down to a hour!! tysm andrew
I'm a newbie beginner on Blender and I use NVIDIA GeForce GTX 760(Display). Its a pretty old card. Thank you for this lesson. Love all the tutorial contents that you've made.
i dont have a gpu so... welp
What about the Denoise node in the compositor? I've been tell that it was better to do it via the compositor (with the denoise data pass) rather than using the one from the property setting. I'm curious since you hasn't mentionned it.
I haven't done any tests since my pc is old, from around 2015 and uses a hard drive from a laptop that's even older...but as far as render time I my setup goes, using the compositor and the denoise I properties goes, it's about the same.
I might try noise threshold and such though, haven't really given that a go yet.
Andrew have mention this in his previous video. It is one of good solution.
The denoise node in the compositor is OpenImage denoise, but as a compositing node. It's better since you still have the original noisy image in case you do something else later(like switch denoisers)
@RealTimeX in another DONUT tutorial :D
ua-cam.com/video/5lr8QnR5WWU/v-deo.html
Makes zero difference on my test render scene, either time wise or quality wise. Some people have reported differences but I haven't been able to replicate it on my Linux RTX3090 rig.
That persistent data shaves off 1/6 of my render time!! That will be a solid 6hours less render than what my current project normally would have been. Appreciate it!!:)
19:54: Eevee Rendering Setup
20:41: Render, Shadows
- Cube Size: For ALL the other lamps
- Cascade Size: Sun lamps (not important, in this case)
----------------
21:50: Light -> Object Data Properties, Shadow
- Clip Start:
- Bias:
----------------
22:28: Contact Shadows - For Sprinkles
- Check, in order for sprinkles to cast shadow onto the icing
- Turn *Bias* down, "all the way down to zero or 0.001"
23:28 Contact Shadows - are a fakery
----------------
25:59: Ambient Occlusion
Few years ago I still had a 5850 1gb from 2010 and I wasn't able to do anything in blender, eventhough Iwanted it so bad. Now I got a decent gpu, and finally I was able to complete your tutorial. Have a good day sir! wish you the best from Türkiye
25:30 the way i use ambient occlusion is by using the actual ambient occlusion NODE! that way you can see what its doing very well. and that node also only works if ambient occlusion is turned on.
My Laptop PC Specs:
CPU: Intel i7 10th GEN
GPU: NVIDIA MX330 (2 GB) + Intel Iris Plus G7 Graphics (8 GB)
RAM: 16 GB DDR4
ROM: 512 GB SSD
I found these settings working best for my particular render:
Noise Threshold: 0.05
Samples: 1024
Rest sample settings were left to default. I could render 1 frame in near around 1 minute, with Denoising on.
Render Time: 00:1:00
Stuck? Join the new Blender Guru Discord to ask for help or just hang with other Blender users: discord.gg/c3B2sZDsam
ok, thx!
GTX1060 = 1min 16seconds. I did the render with the same settings after following up till when you asked. Not too concerned with the time it takes to render, yet. Especially after how long the whole thing has taken me. Noice tutorial Guru, cant wait to start playing around this now. I have a stupid amount of notes from your lessons, hoping ill be able to get through parts of it on my own object without going back to them too much though I'm glad to have them. Cheers
Fantastic series Andrew!
Rendering on 9.44 sec using the extact same setting as you do
RTX 3080 TI - i7 7700k - 32GB RAM
Before this part my renders of the donut in cycles was around 3min but now they are 13sec with GTX1650 and 4096 samples.
Thank you so much for this graet series!!
For anyone struggling with really long render times on cycles. You can keep the sample count at 100 with Denoise and it still looks good. I know he said there's a flickering issue but honestly it's not incredibly noticeable or distracting
Edit: With low Denoise. Like 0.01
it depends if you have less noise on your sample count it will be good. Use 250 or 500 or 1000 or 25000 4096
MacBook Pro 2015 over here. No GPU for Cycles. I’m crying inside. Thank you for including Eevee
Great series! It's my first time using Blender and with this tutorial I've learned so so much and it's been easy to understand thanks to you.
I have a GTX 1050 Ti (and I'm also using the Graphics of my Ryzen 5 2400G) and I'm getting renders of around 40 seconds using the adjustments shown in the video.
Bro I don't know what I did wrong but I have an RTX 3050 on my laptop and I leave it at rendering at 11 and it was not finished even at 3 😅😢
I've got RTX 3060 mobile (laptop version), core i7 11800, and 16gb RAM, and with pretty much same settings my laptop needs around 8 seconds to render a frame.
Huge thanks to you, Andrew, for all tips but especially for choosing the GPU as the renderer hardware!
19:26 You asked for my specs, you shall receive them (If I missed anything important, let me know and I'll add it):
GPU: GTX 1660
CPU: Ryzen 5 3600
RAM: 32GB (I don't remember if it's GDDR3 or 4; I assume 4)
1440p 21:9 monitor (could impact how much the card is being used by windows)
As for render times, it took 16 seconds at default settings w/ optix @ 4096 samples (model is 77,056 Tris, w/ metallic and roughness reflections; the sword on my instagram if you're wondering which model specifically)
Lucky
It takes me 4 hours in 2000 samples
And my graphics card is air (literally)
(Differant scene BTW)
with a 3600 you've probably got DDR4
@@EnterpriseKnight huh
You've grown so much I remember watching you like 3 years ago 😅
It took me a whopping 2.36 minutes to render a single frame on cycles with the same settings as yours!
btw my specs:
GPU: none 😔
CPU: AMD 4500U
ram: 16Gb
it is a laptop(Lenovo Flex 5 AMD version)
It will be even longer if you replace the default cube with a donut!
@@Dedemonn1 already replaced
I used 68 Samples with Denoise with similar hardware, it's way faster and doesn't look bad
You are just the person that Eevee was made for!
For everyone to reference my hardware and render time:
Razer Blade 14 Laptop with built in Nvidia GeForce RTX 3080 GPU
Noise threshold: .05
Max Samples: 4096
Render Time: 9 seconds for Render Image with Cycles, 41 minutes (~7s per frame) for Render Animation in EXR files
I also use my laptop for gaming (often heavily modded), photo and video editing, and audio recording/song production. I have it connected to a monitor. I love this beast. Its so powerful and has been working perfectly since i got it ~1.5-2 years ago. Best tech purchase Ive ever made tbh.
If anyone has any questions about my setup or what it cost or anything feel free to ask :)
now with 3.0, my average render time is somewhere around 5-10 minutes.
before it was about 20 minutes on a 1070 with cuda.
settings are 6mp, 512 samples, intel denoise, subsurf scattering, in a room with around 3 walls, dof. bounces are default 12.
i tend to forget disabling fast gi aprox,.however even if its on 2, it still does a good job imo.
I know some people have commented on the motion blue f-stop thing but I just wanted to give my two cents as a photographer:
Depth of field controls the distortion between your foreground and background. A low or shallow dof will compress the composition while a high dof will create a wide lens effect. The F-stop is what determines how much of your composition is in focus. A low f-stop will allow you to focus on a small section, either foreground, mid, or background, while a high f-stop will put the entire composition in focus.
In a camera, the dof you achieve is a result of the f-stop you choose paired with the distance in mm your lens is set to. So the distance of your camera to the subject will also affect both of these outcomes. A lower distance from the subject to the camera paired with a high distance from the subject to the background with a low f-stop will result in the highest depth of field, greatest compression, and strongest bokeh affect.
Getting 00:16:51 with Cycles using the final suggested settings, on my GTX 1080.
with my i7 5820k and GTX 1070 I cut my Cycles render times down from 6 minutes to ~30 seconds per frame after this video, huge difference :)
I loved he laughed saying "is it normal for it to take 20 minutes per frame...?"
My computer takes 55 minutes per frame.
I was using my CPU to render THIS ENTIRE SERIES, and watching you toggle GPU Compute, then learning that I could as well (I have a GTX 1060) made me feel SO goofy. I sat and watched it render for 18 minutes! My final render was 22 seconds.
suggestion :
When you explained what "passes" are, you could have used the option of the "rendered" view in the viewport to show (in real time, in your viewport) only the passes that you select.
Quick reminder for the Eevee settings: make sure to select the right lamp when you’re checking the contact shadows. I just wondered why it’s not working and I was at the wrong lamp settings.
I have some free time over Christmas as I am visiting the family and decided to learn Blender, so far have been following along and really enjoy this tutorial series! Unfortunately I decided my shitty MacBook Air would be enough to keep me entertained so my render times per frame with just 500 samples are around 2 - 3 minutes depending on my noise threshold! This may take some time to render. Can't wait to get back to my main rig to really play with this! Thank You!!!!
Love your donut series!
btw, MacBook pro with M1 Pro/16 cores GPU renders faster than I thought, each frame took like 10~15 seconds.
Yeah the M1 is a great chip!
If you get blur on your donut while setting the depth of field, except increasing the number of F-stop, there's another thing I do: I add an empty mesh and put it a little front of the donut and then set my focus on it. It fixes the problem, too. I think compared to take a picture in physical world, it just moves the focus point a little front with the empty mesh to anchor it. You could also change the F-stop at the same time for better output.
12:41 This video is making me realize how bad I need a new GPU. With those exact settings it takes me more than 15 mins to render the image.
Thanks for making this tutorial series, it's doing a great job at teaching me the basics of Blender and I'm learning a lot of info from each section.
Using Cycles with all of the render time reduction methods in the video, I was able to achieve a frame in just short of 7 seconds, which was a large improvement from before.
GPU: RTX 3050ti (yes, laptop)
CPU: Ryzen 7 5800h
RAM: 16gb
With very similar settings To Andrew, my render time was 6.93 on Cycles. My hardware is: AMD Ryzen 9 5900X (All Core OC 4.4Ghz) and
Nvidia 3090 FE - and thank you so much for this awesome video series! My doughnut looks so edible!!
I'm glad you reminded me about the GPU... I was rendering on my CPU and it took 6 minutes for some of my tests vs 48 seconds on the GPU. Remember your GPU, guys!
If your render doesnt show the pink background, go to output properties - post processing - then turn on compositing, i dont know why but that worked for me
it works, because the pink background was made by compositing
I forgot to turn it on in my first render x.x
Also, if that doesn't seem to work, it may be because you haven't waited for the render to complete. I know because I just spent 10 minutes trying to figure out why it wasn't working for no reason... sigh
composting is selected still not working, any tips?
I've been using blender since 2 years ago and i can still learn a lot of things from beginner tutorials!!
Increase the F-Stop if you find that your work is so blurry. In my case, it's f/30 because the Donut is really close to the camera. This is same case using the real camera.
Thanks man, It was 2.1 or something by default and half of my donut was basically GONE!
GTX 1050 Ti -- 25 seconds GPU render
I am new to blender . it was so fun to learn thank you :)
Is it better to use multiples of 2 like 512, 1024, 2048 etc. for sample sizes or did you just choose them for quickness?
My god I'm glad your still around man, I remember when you taught me how to make my first donut In 2.8. That led me to realized mechanically how I could utilize blender to make anything just by your tutorial of a donut.
With the final render settings as per this tutorial I'm getting a render at about 22 seconds, I'm running Linux and have a GTX 1080 and an AMD Ryzen 7.
I noticed that Andrew didn't have his processor clicked in the settings and just both his graphics cards, weirdly when I unchecked using the CPU the render time went down to 19 seconds, I would have thought if using the CPU + GPU would be faster.
My guess is tha maybe some extra communication between cpu and gpu is causing this, or having to load the data to ram and Vram, but i am by no means an expert on the topic so I wouldn't know
With a Cycles GPU render using HiPS on an AMD Radeon RX 6900 XT I'm leaving subsurface on for the color depth ( looks good with chocolate icing )
With subsurface and all other settings basically the same:
Noise Threshold of 0.05 renders in 8.9 secs, noise threshold of 0.04 looks a bit cleaner and renders in 11.19 secs.
Thanks for another great video series!
Cycles with all the things you said turned on, subsurface scattering OFF, 8192 sampling, 0.1 noise threshold, intel denoiser
took about 6-7 seconds per frame average on RTX 2060 with Ryzen 3700X
I also tried to reduce the noise threshold to 0.01 and it went over a minute, but with 1080 x 1440 resolution I can't tell any difference.
I'm running the same system and I'm currently averaging 15sec per frame, the only difference is my noise threshold is set to .03. Do you think that would cause the difference in times?
I finish for the first time the donut. Now ill do it again. I just wanted to say that for sprinkles, it help me a lot the old tutorial, cause with nodes it get frustrating in the begging.
Alright, when I activate depth of field, use the donut as the focus object and turn up the F-stop value the donut looks fine. The sprinkles however are so out of focus that they are sort of half transparent and after compositing they are almost invisible... to begin with, my scene looks way more out of focus than Andrew's when I first check the dof box and haven't selected the focus object. Could it be because the distance between my camera and objects is different to Andrew's? In my project the donut is almost inside the camera for it to take that much space of the frame. I went back a little and the last time I saw his camera, it was further back. However I remember him adjusting the zoom while in camera view, which means his camera is probably just as close as mine... so I still have no idea as to why my image is so much more out of focus ._.
edit: as it turnes out, the sprinkles in the back were just way to small, so that them being out of focus and having this blurred outline made them barely visible. I increased their scale in the "instance on points" node and they looked way better and I also realized that in the final render in this video the sprinkles aren't entirely visible as well, I don't know why I thought they had to be
under where u turn on depth of field in the cameras settings i turned up the F-stop up and it brought more of it into focus and i could see the sprinkles in the back a little better. you might have to find what works best for urs
I did the first donut series on my laptop a couple years back. Standard Microsoft surface book 2, 1070ti. Took me about 7-8min per frame with like the lowest of low in every setting.
Redoing the donut, but on my new rig. Amd 7950x with a 4070ti. Now its 4-6 seconds per frame. Mind is blown watching this thing crank this out.
I just started blender yesterday and the tutorials have really helped. I've been following along very closely, but my renders look much worse than yours even though we are using the same resolution and number of samples. There is much more aliasing on the sprinkles and I'm not sure why. Shouldn't the results be pretty much the same based on sample count and resolution alone? I am currently rendering on a 1070 using Optix.
You are a legend Blender Guru. You have saved me soooooo much render time. Thanks.
how do I show my background?
turn on fast GI approximation under caustics in light paths and bump that up to 500 for viewport and render!
I am one of the most experienced artist and using fast GI approximation for a long time so use it!
I hope this helped!
Andrew you need to 3d print your Donut with real baking material and eat it
For anyone who is interested in the Why: a brief history of framerates
lets start with 24
When the first cinematic camera was invented it was trial and error to find the frequency our brain needs to percieve single pictures as motion. Since film was expensive and the cameras had a hard time keeping up with that many pictures they settled with the minimum amount, the magic number of 24.
That is a lot less than our brain could handle wich is evident when you look at car tires in movies. The rotation of the tire is much faster than 24 fps. From one frame to the next it is able to make almost a full rotation but not quite. Thus the car tires seems to rotate bakwards in a fast going car.
This obviously doesnt happen in real life, so our brain is capable of percieving enough pictures per second to keep up with the tire. So 48 is actually closer to our real life perception. And it is very usefull for 3D. In 3D we see two pictures per frame so we have double the information to take in. 48 frames gives the brain double the information to piece the full motion together.
Now why 25 or 30?
The answer lies in the alternating current from the power soccets. In Europe the frequenzy is 50 Hertz, in Amerika it is 60 (no idea why).
That means the lamps in the TV flicker in that frequency and the TV that needed cable used that too.
Now for the first TVs it was a lot to process this much information at once so they used an interlaced signal. Meaning only half the picture information is used in one frame and the other half comes in the next frame. So the TV signal switched frames 50 or 60 times a seconds with each frame holding half the information. So 25 or 30 full pictures.
This is still TV standard today even though modern displays have no problem processing pogressive signals (full picture information per frame) in different frame rates. But the TV signal always has to be recievable with the oldest Tec still in use, so everyone can recieve it.
How can you see the frequency difference for yourself? LED are amercan standard and usually flicker with 60 Hertz. So if you try to film an LED with a shutter speed of 50 you will see the lights flicker. This is actually a big problem in europe because the normal lamps flicker with 50 Hertz. So If you have light from an LED and a normal lamp in the same scene, one of them will always flicker no matter how you adjust the shutter speed.
Last fun fact: This is also the reason why american DVDs dont work in europe and vice versa.
The laser of the player flickers in the frequency of the alternating current. If the information on the DVD is written for a different frequency the laser cant read it anymore.
for me it will take about 35 hours to render 300 frames on cycles on my very bad laptop
but!!!!!!! before this tutorial it would have taken like 5 days to render everything
EDIT:
this time is on a 100 samples it would take 4 hours on one frame if it was 4096 samples
Same 😂 at 4000 samples I'd get the render back after Christmas ha ha
for my render
cycles - RTX 4060 laptop GPU
noise threshold - 0.05
max samples - 8192
time taken = 8.13 seconds (single frame)
same settings but
cycles - i7 12650h
time taken = 19.16 seconds (single frame)
Why don't you show your key presses anymore :(
thanks for the help I actually did some test renders before I got to this video and the render time difference is definitely a game changer once I made the changes.
render time : 12:87 seconds
GPU: RTX 3080
Us Blender users are pure evil. We work with the Not Z axis (shift z) all the time. Hard not to. Its just so powerful.
Again, kudos to all who made it this far and many thanks to Blender Guru! I can only use CPU and, with Cycles, it currently takes 59s to render with Noise Threshold at 0.05, Max Samples 200, Denoise on, Motion Blur on and set to 0.5, DoF turned off. With the same setting but Max Samples ramped up to 300, it takes 1min 7s, so will need to keep samples quite low.
Just finished and OH MY, so happy with result, looks so good.
19:50, regarding specs, with identical settings I have a render time of 7.75 seconds. I have an RTX 3080 and AMD Ryzen 3700X CPU, and 32GB of ram.
Oh holy moly. Did image rendering w/out a compositor after adding a few of the tips and I was looking at a image render time of 00:00:99! Graphics have really come a long way.
Running on a 12GB RTX 3080-Ti with 32GB of RAM and an i7-11700K CPU, if anyone was wondering
QUICK TIP: If you are on Blender 4.0+ and want to export frames in Agx, you can follow the same steps that Andrew does in Filmic, just tried it and it works the exact same :)
For anyone interested in depth of field, some camera research is really good. The F/Stop bit is, in functional terms, how long the 'in focus' section is. A few comments have mentioned it, but I recommend using real camera numbers for it. You can look up an F/Stop chart if you like, but the basics are 1.4, 2, 2.8, 4, 5.6 and so on.. Just if you want to get super accurate focal lengths for real film.
(Note, this is more for technical perfection, realistically just use whatever number looks good)