1 00:00:00,320 --> 00:00:02,460 8k red code raw footage 2 00:00:02,760 --> 00:00:10,560 How do you work with files that are multiple hundreds of megabytes per second of recording with a complex and 3 00:00:11,120 --> 00:00:14,460 Proprietary compression algorithm that allows them to retain effectively 4 00:00:15,000 --> 00:00:20,340 Nearly the full raw quality of the footage while taking up a fraction of the space 5 00:00:21,130 --> 00:00:22,550 well 6 00:00:22,550 --> 00:00:29,910 LG's monitor team seems to have an interest in this and the many technological challenges faced by creative professionals because they sponsor 7 00:00:30,030 --> 00:00:34,110 This video featuring their 49 WL 95 C 8 00:00:34,630 --> 00:00:38,850 Ultra wide monitor where I'll be figuring out just that 9 00:00:39,550 --> 00:00:47,310 So in the old days when anywhere from 4k to 6k was cutting-edge the solution was this 10 00:00:49,310 --> 00:00:51,750 $7,000 red rocket X card 11 00:00:52,030 --> 00:00:59,370 But in spite of it using either a custom ASIC or an FPGA so both highly specialized hardware 12 00:00:59,710 --> 00:01:00,010 even 13 00:01:00,010 --> 00:01:07,410 These puppies were easily overwhelmed by the epic W and the red weapon featuring reds 8k 14 00:01:07,810 --> 00:01:10,230 Helium sensor. So now this is 15 00:01:11,050 --> 00:01:13,050 basically an expensive paperweight 16 00:01:13,770 --> 00:01:14,890 ha 17 00:01:14,890 --> 00:01:16,270 That's it 18 00:01:16,430 --> 00:01:18,670 That's the joys of cutting-edge technology, right? 19 00:01:19,670 --> 00:01:27,010 Fortunately NVIDIA has been running around doing demos with their RTX series graphics cards showing GPU 20 00:01:27,390 --> 00:01:29,970 accelerated real-time full quality 21 00:01:29,970 --> 00:01:34,910 8k playback so then is this the new way now 22 00:01:35,730 --> 00:01:39,950 $1,300 is still a lot of money, but it's a lot less than 23 00:01:41,210 --> 00:01:44,030 $7,000 could this be progress? 24 00:01:44,790 --> 00:01:46,790 It's only one way to find out 25 00:01:47,130 --> 00:02:02,080 So my first encounter with this monster was back at CES and while 49 inch ultra wides like this one 26 00:02:02,180 --> 00:02:06,840 So 32 by 9 aspect ratio look at this thing have existed before 27 00:02:07,820 --> 00:02:11,160 What makes this one special is that the pixel density 28 00:02:11,800 --> 00:02:13,800 Since it's effectively 29 00:02:14,080 --> 00:02:21,860 5k by 1440 is much higher than the ones that have existed before making it actually suitable for creative work 30 00:02:22,400 --> 00:02:27,500 So it has become apparent that I'm going to need to clear some space 31 00:02:28,240 --> 00:02:33,760 So it has become apparent that I'm going to need to clear even more space 32 00:02:34,560 --> 00:02:37,240 Wow, that's a lot of timeline even with 33 00:02:37,240 --> 00:02:39,980 with my task manager open here. 34 00:02:39,980 --> 00:02:41,980 Okay, so first order of business 35 00:02:41,980 --> 00:02:44,480 is to fire up a plausible project. 36 00:02:44,480 --> 00:02:47,260 So this is an episode of Fast as Possible 37 00:02:47,260 --> 00:02:50,740 and use the normal workstations that we have downstairs. 38 00:02:50,740 --> 00:02:51,900 So my CPU is a little better. 39 00:02:51,900 --> 00:02:55,280 This is a 7980XE 18 core processor, 40 00:02:55,280 --> 00:02:58,460 and this is a GTX Titan X Pascal. 41 00:02:58,460 --> 00:03:00,100 So first things first, 42 00:03:00,100 --> 00:03:02,280 we've got ourselves at one quarter resolution. 43 00:03:02,280 --> 00:03:04,740 You can see scrubbing around in the timeline, 44 00:03:04,740 --> 00:03:06,480 nice and responsive. 45 00:03:06,480 --> 00:03:09,950 Playback, basically instantaneous, 46 00:03:09,950 --> 00:03:12,890 but whenever I'm not paused, 47 00:03:12,890 --> 00:03:14,430 you can see it's quite fuzzy, 48 00:03:14,430 --> 00:03:17,310 not representative of the finished product. 49 00:03:17,310 --> 00:03:19,410 So let's crank it up to full. 50 00:03:19,410 --> 00:03:25,960 So this is 8K and this is working just fine. 51 00:03:25,960 --> 00:03:27,220 No problem whatsoever. 52 00:03:27,220 --> 00:03:29,940 And in fact, while I'm playing it back, 53 00:03:29,940 --> 00:03:34,940 we're sitting at a mere 10 to about 30% CPU usage 54 00:03:35,040 --> 00:03:38,920 and our GPU is sitting tight at 25 to 30%. 55 00:03:38,920 --> 00:03:39,760 So at first, 56 00:03:39,760 --> 00:03:41,640 I was kind of confused by all of this. 57 00:03:41,640 --> 00:03:45,080 It's been a while since I've looked into 8K performance 58 00:03:45,080 --> 00:03:46,460 on the timeline. 59 00:03:46,460 --> 00:03:48,080 And I was thinking, well, hold on a second. 60 00:03:48,080 --> 00:03:51,200 Is there nothing special about these RTX cards 61 00:03:51,200 --> 00:03:53,140 that makes them suitable for this? 62 00:03:53,140 --> 00:03:56,220 Why does NVIDIA keep talking about this? 63 00:03:56,220 --> 00:04:03,620 And then I remembered something. 64 00:04:03,620 --> 00:04:06,240 We shoot most of our projects 65 00:04:06,240 --> 00:04:09,200 at a much higher compression ratio 66 00:04:09,200 --> 00:04:11,280 than what they would use while shooting 67 00:04:11,280 --> 00:04:13,500 a VFX heavy Hollywood production, 68 00:04:13,500 --> 00:04:14,980 for example. 69 00:04:14,980 --> 00:04:19,400 So our data rates are much lower. 70 00:04:19,400 --> 00:04:23,620 So let's go get some seven to one footage. 71 00:04:23,620 --> 00:04:25,840 So let's start with a sample from RED here. 72 00:04:25,840 --> 00:04:28,440 This is a seven to one compression ratio, 73 00:04:28,440 --> 00:04:33,440 8K clip that was recorded at 24 frames per second. 74 00:04:33,780 --> 00:04:37,460 So let's go ahead and fire this bad boy up, 75 00:04:37,460 --> 00:04:42,370 hit that playback button and ouch. 76 00:04:42,370 --> 00:04:44,770 We are looking at spikes of up to 90. 77 00:04:44,770 --> 00:04:49,770 95% on our GPU and CPU usage that is pinned at 100% 78 00:04:52,950 --> 00:04:55,530 while trying to play back that clip. 79 00:04:55,530 --> 00:04:58,870 That is a stark contrast to the, 80 00:04:58,870 --> 00:05:00,810 I think we used 20 to one, is that right, David? 81 00:05:00,810 --> 00:05:03,970 To the 20 to one footage that we capture. 82 00:05:03,970 --> 00:05:05,570 And that's not even the worst of it. 83 00:05:05,570 --> 00:05:07,870 Here's some 8K seven to one footage 84 00:05:07,870 --> 00:05:11,250 at 30 FPS that we captured. 85 00:05:11,250 --> 00:05:16,070 Let's go ahead and drag that onto the timeline. 86 00:05:16,070 --> 00:05:16,850 Yeah, woof. 87 00:05:16,850 --> 00:05:22,300 Not quite. 88 00:05:22,300 --> 00:05:23,700 Yeah, he moves like that. 89 00:05:23,700 --> 00:05:24,700 Sure, why not? 90 00:05:24,700 --> 00:05:25,720 So what's weird though, 91 00:05:25,720 --> 00:05:27,580 is that even with the heaviest footage 92 00:05:27,580 --> 00:05:29,400 that we could throw at our workstation, 93 00:05:29,400 --> 00:05:33,220 our GPU is actually neither sitting idle 94 00:05:33,220 --> 00:05:35,340 nor completely maxed out. 95 00:05:35,340 --> 00:05:38,340 So it's being used to handle the debayering process, 96 00:05:38,340 --> 00:05:41,640 which reconstructs the full color image that we see here 97 00:05:41,640 --> 00:05:44,140 from the samples that are collected by the image sensor. 98 00:05:44,140 --> 00:05:47,020 But our CPU is still in charge of the, 99 00:05:47,020 --> 00:05:49,020 entropy decoding, 100 00:05:49,020 --> 00:05:52,520 which is a process that allows the most common input signals 101 00:05:52,520 --> 00:05:54,980 to be substituted for much shorter ones, 102 00:05:54,980 --> 00:05:57,580 saving space and wavelet decoding, 103 00:05:57,580 --> 00:06:00,080 which frankly, I'm not familiar enough with 104 00:06:00,080 --> 00:06:02,000 to offer even a basic explanation 105 00:06:02,000 --> 00:06:04,780 and the Wikipedia article wasn't much help. 106 00:06:04,780 --> 00:06:07,980 Anyway though, so maybe our problem here 107 00:06:07,980 --> 00:06:13,140 is we just need more CPU horsepower to keep our GPU fed. 108 00:06:13,440 --> 00:06:15,880 AMD to the rescue then? 109 00:06:15,880 --> 00:06:16,880 Why don't we try going, oh, oh, oh, oh, oh, oh, oh, oh, oh, oh. 110 00:06:16,880 --> 00:06:21,880 Why don't we try going from 18 to 32 processor cores? 111 00:06:22,280 --> 00:06:31,040 Building in progress. 112 00:06:31,040 --> 00:06:38,590 Mission failed. 113 00:06:38,590 --> 00:06:41,410 Okay, it worked. 114 00:06:41,410 --> 00:06:43,620 Didn't do anything. 115 00:06:43,620 --> 00:06:48,000 And interestingly, our GPU usage is higher, near 100%. 116 00:06:48,000 --> 00:06:50,750 So this is kind of fascinating. 117 00:06:50,750 --> 00:06:52,770 I've said it many times before 118 00:06:52,770 --> 00:06:56,690 that every system is effectively bottlenecked 119 00:06:56,690 --> 00:07:00,510 because there's something that is slowing down the process 120 00:07:00,510 --> 00:07:03,050 or else it would be infinitely fast. 121 00:07:03,050 --> 00:07:07,510 But this is like double bottlenecked. 122 00:07:07,510 --> 00:07:10,810 It's like they're perfectly matched to each other, 123 00:07:10,810 --> 00:07:13,350 but still not enough for the task. 124 00:07:13,350 --> 00:07:14,570 This is great too. 125 00:07:14,570 --> 00:07:17,490 Look what happens to timeline scrolling 126 00:07:17,490 --> 00:07:20,690 when my CPU's at 100% usage. 127 00:07:20,690 --> 00:07:23,030 So while I was troubleshooting to prepare for this video, 128 00:07:23,030 --> 00:07:26,510 I actually spent some of my time with the RTX 2080 129 00:07:26,510 --> 00:07:29,590 rather than the RTX 2080 Ti, 130 00:07:29,590 --> 00:07:30,750 because what I thought, 131 00:07:30,750 --> 00:07:34,750 was that this functionality had something to do 132 00:07:34,750 --> 00:07:37,230 with their new NVENC engine. 133 00:07:37,230 --> 00:07:39,070 But actually it doesn't. 134 00:07:39,070 --> 00:07:42,430 In fact, there's nothing about the tensor cores 135 00:07:42,430 --> 00:07:45,370 or the ray tracing cores or anything like that 136 00:07:45,370 --> 00:07:47,790 that makes the RTX card so good at this. 137 00:07:47,790 --> 00:07:52,510 It's just the sheer frigging processing power 138 00:07:52,510 --> 00:07:54,410 of all those CUDA cores. 139 00:07:54,410 --> 00:07:58,470 So we are gonna jump for 8K, seven to one footage, 140 00:07:58,470 --> 00:08:00,330 straight to the RTX 2080 Ti. 141 00:08:00,750 --> 00:08:03,290 And you might think to yourself, 142 00:08:03,290 --> 00:08:08,480 well, gee Linus, that's a 1200 to 1300 US dollar graphics card. 143 00:08:12,080 --> 00:08:14,000 Of course it can handle it, 144 00:08:14,000 --> 00:08:17,020 but that's not necessarily to be taken for granted. 145 00:08:17,020 --> 00:08:21,020 I mean, we installed CPUs that cost up to $2,000 146 00:08:22,480 --> 00:08:24,100 and they couldn't do it. 147 00:08:24,100 --> 00:08:27,380 It's all about having the right tool for the job. 148 00:08:27,380 --> 00:08:29,680 Okay, everything's working normally. 149 00:08:29,680 --> 00:08:30,720 Here we go. 150 00:08:31,100 --> 00:08:34,620 This is the moment. 151 00:08:39,860 --> 00:08:44,060 I mean, now that has kinda gotten its act together, 152 00:08:44,060 --> 00:08:45,830 that's a lot better. 153 00:08:47,280 --> 00:08:50,440 We are now at 100% CPU, 95% GPU, 154 00:08:51,960 --> 00:08:54,620 but we are still dropping frames. 155 00:08:57,710 --> 00:08:59,820 So that's it then. 156 00:08:59,820 --> 00:09:04,500 No CPU on earth can handle this 30 frames per second, 157 00:09:04,500 --> 00:09:06,100 8K, seven to one, 158 00:09:06,100 --> 00:09:13,880 Red code raw footage at least not as long as it has to handle the entropy decoding and wavelet decoding 159 00:09:14,980 --> 00:09:22,400 What if we could offload those to the GPU we are going to need a beta version of Red's decoder 160 00:09:22,680 --> 00:09:26,940 That unfortunately isn't supported by Adobe Premiere at this time 161 00:09:27,880 --> 00:09:35,120 Fortunately, it's built into the beta of Red Cine X Pro and we're gonna start with the 24 FPS footage now 162 00:09:35,120 --> 00:09:41,380 It's a little flaky right now. I'm actually not sure if this is going to work, but let's go ahead and give it a shot 163 00:09:41,440 --> 00:09:43,360 so we're gonna enable 164 00:09:43,360 --> 00:09:45,360 GPU decode and 165 00:09:45,960 --> 00:09:50,240 It's flaking out on me. So it's running it like eight frames per second 166 00:09:51,040 --> 00:09:57,680 Fortunately the fix right now seems to be to just toggle between image pipelines and there it is 167 00:09:58,560 --> 00:10:02,740 Smooth 8k playback that hitch by the way is the clip 168 00:10:03,100 --> 00:10:05,040 Restarting but that's not even the most 169 00:10:05,120 --> 00:10:09,820 Impressive part. So our CPU is sitting at just 10% 170 00:10:10,980 --> 00:10:16,420 Utilization and our GPU is up at 77 80 percent utilization. So we 171 00:10:17,120 --> 00:10:19,600 Potentially have some headroom to spare 172 00:10:19,640 --> 00:10:25,900 So our overall usage for the system is way down like 10% of our 32 cores 173 00:10:26,360 --> 00:10:30,740 We could be doing this on a more human workstation 174 00:10:30,740 --> 00:10:35,060 You know one with six or eight cores by offloading this 175 00:10:35,120 --> 00:10:38,900 We could be doing this work to the GPU because the thing to remember as well 176 00:10:38,900 --> 00:10:43,720 Is that when we were pinning both our CPU and GPU at a hundred percent? 177 00:10:44,360 --> 00:10:49,440 That was just playback. We weren't even applying any real-time effects like 178 00:10:50,140 --> 00:10:56,340 Denoising the footage or or anything like that. We were at the limit now. We have headroom to play with 179 00:10:56,340 --> 00:10:59,920 Let's go ahead and try our 30 FPS footage though 180 00:11:00,120 --> 00:11:04,980 That's actually working a little bit better than I expected when I the first time around 181 00:11:04,980 --> 00:11:11,580 I actually had to overclock the GPU a little bit to get this quite so smooth, but you know what I'm taking it 182 00:11:12,070 --> 00:11:16,170 So there it is our GPU CUDA usage jumps up near 183 00:11:17,150 --> 00:11:18,130 90% 184 00:11:18,130 --> 00:11:23,630 But our CPU stays at 10% and we are smoothly playing back 185 00:11:24,590 --> 00:11:26,590 7 to 1 186 00:11:26,590 --> 00:11:32,070 8k red footage at 29.97 frames per second 187 00:11:33,390 --> 00:11:34,990 Can we get some applause sound effect? 188 00:11:35,010 --> 00:11:37,590 In the video or something? I don't know. This is crazy 189 00:11:37,910 --> 00:11:45,810 So all that's left now then is to thank LG for sponsoring this video featuring their 49 WL 95 CW 190 00:11:46,270 --> 00:11:47,770 Ultra wide monitor 191 00:11:47,770 --> 00:11:48,390 I mean 192 00:11:48,390 --> 00:11:54,990 I think it's pretty obvious why they wanted us to feature this in this video because it's designed for content creators like photographers 193 00:11:55,610 --> 00:12:00,810 Filmmakers music producers and more who can really benefit from this 194 00:12:01,130 --> 00:12:04,950 Kind of a canvas being able to work on more than one thing 195 00:12:05,010 --> 00:12:08,890 simultaneously or just be able to manage and see 196 00:12:09,670 --> 00:12:16,010 extremely long timelines without zooming out giving them a finer degree of control and it's not even just for creators 197 00:12:16,010 --> 00:12:19,990 But office work would also be a great option someone like a programmer 198 00:12:20,470 --> 00:12:26,330 Stock trader or really anyone who needs a ton of side-by-side screen real estate because it's basically 199 00:12:27,250 --> 00:12:29,510 Exactly the same as having two 200 00:12:30,610 --> 00:12:34,070 2560 by 1440 monitors side by side 201 00:12:34,070 --> 00:12:40,930 Except that you have no bezel in between LG's also been hard at work on software that allows you to take two inputs 202 00:12:40,930 --> 00:12:46,830 So like your desktop and your laptop and seamlessly share your mouse and keyboard between them just like moving between them 203 00:12:46,830 --> 00:12:52,890 So you can control two systems at once. It's got full support for USB type-c including power delivery 204 00:12:52,890 --> 00:12:56,430 So you can use it as a single cable docking solution for your laptop 205 00:12:56,430 --> 00:13:03,930 And it's got support for HDR 10 with two 10 watt speakers built-in. In fact, it was didn't realize they were gonna be that loud 206 00:13:03,930 --> 00:13:06,630 That happened that happened a couple times during the video 207 00:13:07,190 --> 00:13:11,130 So check it out at the link in the video description, and I guess that's pretty much it 208 00:13:11,130 --> 00:13:13,410 Thanks for watching you guys if you disliked this video 209 00:13:13,410 --> 00:13:14,610 Well, you can hit that button 210 00:13:14,610 --> 00:13:18,870 But if you liked it hit like get subscribed or maybe consider checking out where to buy the stuff 211 00:13:18,870 --> 00:13:21,450 We featured at the link below also down 212 00:13:21,450 --> 00:13:26,130 There's our merch store which has cool shirts like this one and our community forum, which you should totally join