WEBVTT 00:01.030 --> 00:04.410 Welcome to lowering your cycles a render time. 00:06.090 --> 00:12.250 So looking at the cycles wonder engine we're going to have cycles cpq vs. g.p. 00:12.300 --> 00:19.470 It's important that you go through this process now so that you know which one is quicker and B how 00:19.470 --> 00:22.340 to set it up for your particular computer. 00:22.560 --> 00:26.310 So we're going to go through how to change your render settings. 00:26.310 --> 00:33.660 You will have to run multiple tests as we go through and different scenes will have different characteristics 00:33.660 --> 00:36.000 depending on their complexity. 00:36.300 --> 00:41.150 But in general all that will follow a similar pattern. 00:41.340 --> 00:44.300 So let's hop back on over into blender. 00:44.760 --> 00:49.960 So once you've saved your user preferences we can close down the user preferences window. 00:50.160 --> 00:54.240 And I'm going to expand these windows on the left hand side. 00:54.240 --> 00:59.070 Now the render tab is this one on the camera on the left on the properties panel. 00:59.460 --> 01:02.870 So once we've clicked that we have a series of options. 01:02.970 --> 01:07.600 And as we scroll down we can see there are lots of them to go through. 01:07.710 --> 01:11.780 And we're only going to go through the very basic ones at this point in time. 01:11.790 --> 01:14.640 First of all are we in cycles. 01:14.670 --> 01:15.360 No we're not. 01:15.360 --> 01:16.410 So let's change it. 01:16.410 --> 01:21.500 Because when you're using blender render the only thing you can use is you'll see you. 01:21.510 --> 01:23.660 It does not use the GP at all. 01:23.700 --> 01:27.340 So we switch that from Blender end to cycles render. 01:28.240 --> 01:30.560 We get a few other options. 01:30.760 --> 01:38.950 First of all device cpq or g.p you compute depending on which one you select will change which component 01:39.010 --> 01:41.970 blender uses to perform the rendering. 01:42.190 --> 01:45.080 So let's change that now to GP you compute. 01:45.310 --> 01:50.530 Now the two of the things we're going to pay attention to at this point in time is the dimensions of 01:50.530 --> 01:51.640 the render. 01:51.640 --> 01:59.090 Now this point we're only focussed on the resolution and by default it's set to 10 ATP or high def. 01:59.140 --> 02:03.910 Now we notice there's a slide underneath that says 50 percent so actually this is going to render 50 02:03.910 --> 02:05.970 percent of 10 ATP. 02:06.040 --> 02:10.550 So you can adjust that to get a good feel of rended times. 02:10.550 --> 02:15.610 If we rack up fall from 50 percent to 100 percent I think it's going to take about four times longer 02:15.610 --> 02:20.410 to do the render than if it's at 50 percent and it it'll be four times quicker if we knock that down 02:20.410 --> 02:22.440 to 25 percent. 02:22.480 --> 02:24.360 So let's scroll right down. 02:24.370 --> 02:28.240 There's one of the option that's very important that's under performance. 02:28.420 --> 02:33.610 If we open that up and scrolled down we have tiles. 02:33.640 --> 02:39.930 Now this is very important and will vary from card to card graphics card to graphics card or processor 02:39.930 --> 02:41.280 to processor. 02:41.290 --> 02:49.080 So x and y settings play about those and we'll find that some things take longer than others. 02:49.240 --> 02:54.400 So let's leave it at 64 64 and render up this cube here. 02:54.820 --> 02:57.400 So let's go to render and render image. 02:57.490 --> 02:58.300 And there we go. 02:58.300 --> 03:04.420 It's rendered it out and we can see the time at the top here is just over one second. 03:04.420 --> 03:12.010 Now this cube is rather basic So we cant really do any testing with it because the render times are 03:12.010 --> 03:15.280 just going to be around one second whatever we pick. 03:15.280 --> 03:23.260 So what I'd like you to do is go to the section assets and open up the BMW files. 03:23.260 --> 03:26.480 So let's go and open up our section assets now. 03:27.230 --> 03:27.980 So here we go. 03:27.980 --> 03:31.320 BMW twenty seven blend. 03:31.520 --> 03:37.250 If you open that up you should end up with something like this on your screen and it's got a convenient 03:37.280 --> 03:40.700 notepad remember to save the file as you go. 03:40.700 --> 03:47.270 Now the person that set the scene up we can see that they used an Intel seven three point eight gigahertz. 03:47.510 --> 03:55.220 That's They've got a g force GTA 770 and they've noted the times beneath using the GP you and s.p. you 03:56.150 --> 04:00.050 now feel free to explore around our scene. 04:00.140 --> 04:05.390 We can see how it's constructed and we'll be doing a similar thing later on in fact we zoom out. 04:05.570 --> 04:12.430 We can see that this is all these two cars are in a bowl of sorts and there's a couple of cameras all 04:12.440 --> 04:18.160 set up when you go to render your image up the top here with render render image of it. 04:18.200 --> 04:23.170 F 12 or indeed there's a button under render on the properties here. 04:23.300 --> 04:27.920 When you hit render you can see it starts building our scene. 04:27.920 --> 04:33.590 First of all because it's much more in depth and then it starts rendering. 04:33.590 --> 04:36.270 Now we're on g.p you computer. 04:36.290 --> 04:44.000 And if we scroll down the dimensions are set to 50 percent of 10 ATP which is fine you want to leave 04:44.000 --> 04:47.690 it at that for these tests on performance. 04:47.690 --> 04:52.850 We can see that the tiles size is currently set to 128 by 128. 04:52.850 --> 04:58.280 Now I'm not going to sit here and wait for the whole thing to render out as we can see on the top here 04:58.280 --> 05:01.450 it's predicting about 12 minutes in total. 05:01.460 --> 05:10.730 However if we change vs here the x and y you see that what the impact of that is so this tile size refers 05:10.730 --> 05:14.900 to how many pixels by how many pixels were actually rendering at. 05:14.900 --> 05:18.680 So if I cancel the Render with this little cross on the info head up. 05:19.660 --> 05:23.690 And change this to 12 by 5 12. 05:23.800 --> 05:29.830 You see that it renders a much larger area at once so it needs to build the scene first and do some 05:29.830 --> 05:31.390 calculations. 05:32.120 --> 05:37.180 Oh we can see it's building up this large area here and it's doing that all at once. 05:37.190 --> 05:44.800 Now depending on the settings your graphics card all s.p. you will render at different times. 05:44.870 --> 05:51.650 In general these numbers are larger with graphics cards and smaller with s.p. Use If I go up to our 05:51.660 --> 05:57.700 asses GP you compute and switch that to see you stop the current render. 05:58.650 --> 06:07.770 And then scroll back down and perhaps put this at 64 by 64 so slightly smaller than it was before and 06:07.770 --> 06:12.480 hit the render key F 12. 06:12.480 --> 06:20.750 You'll notice that I have after it's built one two three four five six seven eight squares. 06:20.880 --> 06:27.260 So it's doing 8 renders out once whereas the GP does one at once. 06:27.330 --> 06:35.160 So here that is because I have a four core processor that's got hyper threading So it acts as if it's 06:35.160 --> 06:38.720 got eight cause us able to do eight things at once. 06:38.730 --> 06:44.930 In general the faster cpq renders other ones that can utilize the whole of the processor. 06:45.090 --> 06:47.440 For the longest time. 06:47.490 --> 06:54.760 So if we set this whole scene up at 12 by five 12 for instance or even bigger than that. 06:54.930 --> 06:58.840 If we set it up to a thousand. 06:58.860 --> 07:00.980 Well let's do the Fall Reds. 07:00.990 --> 07:02.690 I've got those the wrong way round. 07:02.700 --> 07:05.300 I do apologize but I want half of those. 07:05.300 --> 07:09.270 So 540 and 9 6. 07:09.290 --> 07:11.240 The I think of my maths is right. 07:11.370 --> 07:13.500 So that'll render the whole scene at once. 07:13.620 --> 07:17.670 But instead of using all eight cores Let's stop the render instead of using all eight. 07:17.670 --> 07:24.780 Cause if I go and render this image it will just literally use one core. 07:24.870 --> 07:32.610 We got the orange borders of there and it will be rendering it only using one call which is not very 07:32.610 --> 07:33.710 efficient. 07:35.020 --> 07:36.630 OK it's Challenge time. 07:36.650 --> 07:40.810 I'd like you to get to your best render time if you haven't done already. 07:40.810 --> 07:49.150 Please open up the BMW 27 dot blend file from the outset pack and I'd like you to perform multiple renderings 07:49.150 --> 07:58.660 with the GP you and see you leave the render at 10 ATP at 50 percent and only adjust the tyre or size 07:58.900 --> 08:01.660 to see the effect on render times. 08:01.660 --> 08:07.780 So you have to get a pen and paper or a spreadsheet out and record the times or indeed a good record 08:07.820 --> 08:13.990 in the little notepad that's conveniently placed in the lower right hand side of this blender file. 08:14.230 --> 08:22.390 Now do that several times and post your best render time s.p. you or GP and the tiles size in the discussions. 08:22.420 --> 08:29.050 Just a word to the wise before you go off and do that if at any point you think that the time is going 08:29.050 --> 08:36.580 to be astronomically high and then I would recommend stopping the render and changing the tile size. 08:36.580 --> 08:44.080 In general the GP use will do better with larger tiles sizes and CPS will do better with smaller but 08:44.080 --> 08:47.660 that will vary depending on your particular hardware. 08:47.950 --> 08:50.730 So Paul as a video now and give that a go. 08:52.050 --> 08:53.270 Welcome back. 08:53.280 --> 08:55.600 Let's hop over into blender. 08:56.010 --> 08:58.480 So we left our scene like this. 08:58.480 --> 09:02.950 Now I'm not going to bore you all with the rendering times at the moment. 09:03.090 --> 09:08.160 I've already gone off and done that whole so I'm going to open one of the files I've been working on 09:08.550 --> 09:12.000 which is the BMW 27 render times. 09:12.260 --> 09:16.020 So how did I fare on my computer. 09:16.270 --> 09:21.210 So I'm just going to pull this out completely as you can see I've done lots and lots of tests on my 09:21.210 --> 09:22.080 computer. 09:22.140 --> 09:26.910 Now I've gone all the way from 16 by 16 all the way up to 512 by five tough. 09:27.030 --> 09:32.190 You don't have to have these particular numbers by the way they don't have the match as you can see 09:32.190 --> 09:33.600 reds matched underneath. 09:33.600 --> 09:39.680 These are direct divisions of the resolution that were rendering to. 09:40.140 --> 09:44.880 So here we have 16 by 16 you can see it takes up to 52 minutes. 09:44.880 --> 09:52.860 I actually did not wait for that one to render 32 by 32 it gets down to 30 minutes 8 minutes 7 minutes 09:53.270 --> 09:57.860 7 minutes 23 7 a half 723 So that looks like that one. 09:57.900 --> 10:01.710 There was the fastest that the GP was going to render. 10:01.830 --> 10:04.110 Now we can see that I got is even lower. 10:04.110 --> 10:10.440 When we had a rest matches just because this happens to be closer to the ideal value for both my graphics 10:10.440 --> 10:15.500 card and this particular scene I've made a couple more notes here. 10:15.510 --> 10:21.310 When it comes to the CPM skill to pull that out so we can read my drivel. 10:21.420 --> 10:22.680 Let's just. 10:23.400 --> 10:27.090 Pull that down so you guys can read what it says as well. 10:27.300 --> 10:36.510 Okay so s.p. you time it looks like as you can see here five high 12 for cause we utilize briefly and 10:36.510 --> 10:39.000 then a rapid drop down to two cause. 10:39.060 --> 10:43.410 So my max utilization on the Mac is eight hundred percent. 10:43.410 --> 10:47.600 That's for cause with four hyper threading units running at once. 10:47.790 --> 10:53.160 So 100 percent would be one core and eight would be the eight cores in total. 10:53.220 --> 10:58.840 So the longer the whole process can be utilized the faster the render. 10:58.860 --> 11:05.820 So here we have eight by eight done in four minutes and 10 seconds so the first thing to note here is 11:05.820 --> 11:13.650 that my See you on this particular scene is significantly faster than the GP you so I will probably 11:13.650 --> 11:17.560 have cycles using the s.p. all the time. 11:17.580 --> 11:26.190 Now if you imagine if this was one so one frame that you were rendering of a let's say 24 frames per 11:26.190 --> 11:32.160 second and it was just a 10 second clip or what does that work out to be. 11:32.360 --> 11:33.870 I can spell calculator. 11:33.870 --> 11:34.680 There we go. 11:34.680 --> 11:39.830 So let's just call it four and a half minutes because it will differ. 11:39.870 --> 11:42.290 It may get more complex or more simple as it goes. 11:42.300 --> 11:48.150 But four and a half minutes would be a good one times 24 so that's how many minutes it's going to take 11:48.150 --> 11:50.700 to do one second. 11:50.700 --> 11:57.480 A hundred and eight minutes to do one second of video of animation. 11:57.480 --> 11:59.220 This particular resolution. 11:59.370 --> 12:01.810 And that's not even 10 gtp Wow. 12:01.860 --> 12:03.370 That's a lot. 12:03.390 --> 12:06.520 So let's have a look at the Reds match. 12:06.540 --> 12:09.250 And as you can see there's no real benefits from going this high. 12:09.390 --> 12:18.540 So in this case the eight by eight s.p. tiles on the s.p. was the well known 16 by 16 was the fastest. 12:18.600 --> 12:25.890 Now these three here are actually all within the margin of error even 64 by 64 I'd say. 12:25.950 --> 12:32.370 If you run these tests again and again you'll probably find that these three are pretty much the same 12:32.430 --> 12:34.020 across the board. 12:34.020 --> 12:40.410 So eight by eight to six to 32 by 32 It seems to be the ideal position for this. 12:40.410 --> 12:45.720 Now on the blender website if I can just take you to the blender website here. 12:45.960 --> 12:48.520 Blender dot org. 12:48.570 --> 12:50.440 Leave the page there. 12:50.640 --> 12:50.960 Okay. 12:50.970 --> 12:57.240 When we go to the downloads area there is a series of demo files and that's where we've got the BMW 12:57.240 --> 12:59.270 benchmark file from. 12:59.520 --> 13:04.920 And I've opened up fishy cat as well and run it on that and the class from the class class room is incredibly 13:04.920 --> 13:05.930 intensive. 13:06.180 --> 13:10.130 So when we recommend using that one is just a quick render test. 13:10.400 --> 13:11.740 Now a fishy cat itself. 13:11.750 --> 13:13.330 I've done that one as well. 13:13.440 --> 13:17.650 And he was the little mascot of the last version two point seventy four. 13:17.850 --> 13:22.870 So if we go back into blender and I open up my fishy cat render. 13:24.590 --> 13:27.220 We can see that a similar thing happens. 13:27.230 --> 13:29.590 Let's just pull this open here. 13:29.600 --> 13:38.000 I haven't done all of these but as you can see on the GP you it runs the best time there is here in 13:38.000 --> 13:40.540 the same place it was last time I believe. 13:40.940 --> 13:44.530 And we can see that this lot here all within the margin of error. 13:44.540 --> 13:48.100 Any way it could quite as easily flicked between the two. 13:48.110 --> 13:49.460 Maybe not this one here. 13:49.490 --> 13:57.530 Not quite sure but see you times we can see that in this case 32 by 32 happens to be the best. 13:57.530 --> 14:03.920 So again these are all pretty much within margin of error eight by eight in this case was outside of 14:03.920 --> 14:10.610 that but these three here 16 by 16 to 64 by 64 my s.p. You were the fastest. 14:10.610 --> 14:14.780 Now you may not think oh well that's a few seconds difference. 14:14.780 --> 14:23.060 But again if you're doing an animation where you've got 25 renders per second I mean that's really the 14:23.060 --> 14:24.620 minimum you'll be working too. 14:24.770 --> 14:28.360 If you wanted 60 frames per session it's going to be a very long wait. 14:28.460 --> 14:32.300 Those few seconds add up to huge magnitude. 14:32.300 --> 14:37.940 So after that you've done this what I would like you to do of course if you haven't done already is 14:37.940 --> 14:44.000 post your best friend time to the discussions I really want to see what type of hardware you've got 14:44.780 --> 14:50.900 whether you've used your GP or see you the tile size you've used and how quickly you managed to render 14:50.900 --> 14:55.580 out that BMW would be great to see the differences in the technology. 14:55.610 --> 15:00.210 Now it's also very important this point again to highlight the render time. 15:00.380 --> 15:06.680 As you can see six minutes is quite a long time to wait for something to come out. 15:06.680 --> 15:14.960 However if we go back to the BMW we were working on you'll notice as you start rendering the image and 15:14.960 --> 15:17.060 was going to switch just to see you. 15:17.270 --> 15:24.760 And we know from my tests that actually around 16 by 16 is going to produce the best results. 15:24.830 --> 15:30.030 If I now go and render this whole buzz away in the background. 15:30.200 --> 15:37.940 But you can see that you can get a relatively good idea of how it's going to be rendered by just zooming 15:37.940 --> 15:38.380 in. 15:38.570 --> 15:41.440 And if that wasn't good enough that's the final thing there. 15:41.450 --> 15:44.770 Well it's not actually the final if you follow it through right to the end. 15:44.860 --> 15:51.690 You will spot there was a bit of compositing where a light appears as well on us seen but there we go. 15:51.740 --> 15:59.660 You should now be able to set up a blender to render cycles as quickly as it can on your particular 15:59.660 --> 16:00.920 computer. 16:01.130 --> 16:03.710 And I will see you in the next lecture.