1 00:00:00,000 --> 00:00:07,400 Woo! Look at her go! Here in my hands is the power supply that's responsible for that 2 00:00:07,400 --> 00:00:12,840 fireworks display. And guess what, guys? We didn't try to kill it, and it isn't from 3 00:00:12,840 --> 00:00:19,920 some random Aliexpress brand. This unit has five stars on PC Part Picker and thousands 4 00:00:19,920 --> 00:00:26,520 of sales on Amazon, and yet it failed spectacularly during routine testing for Power Supply Circuit, 5 00:00:26,520 --> 00:00:30,880 or a channel that's been testing legions of power supplies and, along the way, discovering 6 00:00:30,880 --> 00:00:37,080 that many power supply manufacturers aren't so honest. Some are over-representing their 7 00:00:37,080 --> 00:00:42,920 products and, wait, some are underselling their performance? That's not what I expected. 8 00:00:42,920 --> 00:00:48,720 Well, either way, it's finally time to take a look at our power supply gauntlet. And when 9 00:00:48,720 --> 00:00:54,120 I say gauntlet, I mean it, because our power supply testing process has resulted in a staggering 10 00:00:54,120 --> 00:01:02,480 23% failure rate, really separating the good from the, uh, well, bright, at least. 11 00:01:02,480 --> 00:01:06,580 Speaking of failure, I just failed to do a segue to our sponsor, uh! 12 00:01:06,580 --> 00:01:11,280 Nine months ago now, we launched our first Labs team channel, PSU Circuit, and since 13 00:01:11,280 --> 00:01:17,160 then we've tested 39 power supplies, nine of which died at some point during our test 14 00:01:17,160 --> 00:01:20,240 suite. But we're getting ahead of ourselves a little bit. 15 00:01:20,240 --> 00:01:26,460 Next it's time to take a look at our fully armed and operational power supply station. 16 00:01:26,460 --> 00:01:30,120 Because the keen eye to mountain you might have noticed that it looks a little different 17 00:01:30,120 --> 00:01:34,400 from how it looked two and a half years ago when we first unboxed this bad boy. Starting 18 00:01:34,400 --> 00:01:39,440 at the bottom, we've got our AC source, which is able to deliver three kilovolt amps, which 19 00:01:39,440 --> 00:01:47,080 is basically 3,000 watts, but not quite because reasons. This is enough for us to test up 20 00:01:47,160 --> 00:01:53,600 to 1,600 watt power supplies because we need to be able to briefly deliver 200% of the 21 00:01:53,600 --> 00:02:00,160 rated power for the power supply for excursion testing. As for why it isn't exactly double, 22 00:02:00,160 --> 00:02:06,480 the reasons. Anyway, above that we have what Chroma calls their on-off controller, which 23 00:02:06,480 --> 00:02:12,120 is just a big relay, which is just a fancy name for switch. Then above that we have our 24 00:02:12,120 --> 00:02:17,060 ShortCircuit and overvoltage protection tester. Moving up we have our DC power source, which 25 00:02:17,060 --> 00:02:21,220 we don't really use at this time. And then above that we have our timing and noise analyzer, 26 00:02:21,220 --> 00:02:26,140 which is basically a headless oscilloscope with a host of relays that can be used to 27 00:02:26,140 --> 00:02:31,620 automate functions like turning the power supply on or off or changing the characteristics 28 00:02:31,620 --> 00:02:38,620 of our load, say for example to add capacitance. Then there's a shelf, which does shelf things. 29 00:02:38,620 --> 00:02:43,020 Above the shelf we've got the connection panel, which allows easy connections so that poor 30 00:02:43,020 --> 00:02:47,260 Lucas doesn't have to go spelunking in the back of a rack every time a wire needs to 31 00:02:47,260 --> 00:02:54,780 be plugged in. Of course the real meat and potatoes is up here. Our 13 loads, an unlucky 32 00:02:54,780 --> 00:03:00,140 number for many a power supply. Basically these guys use MOSFETs to dissipate the energy 33 00:03:00,140 --> 00:03:05,860 from our power supply as heat. We've got 12 400 watt loads that can be strung together 34 00:03:05,860 --> 00:03:12,860 to fully load any rail on any power supply up to 1600 watts and then our 13th unit has 35 00:03:13,100 --> 00:03:19,100 two 100 watt loads, which we typically use for 5 volts standby and minus 12 volts since 36 00:03:19,100 --> 00:03:25,100 we don't really need a big individual load for those. It can get pretty warm behind these 37 00:03:25,100 --> 00:03:31,660 guys. And beside that is a digital multimeter, which we use for the seven thermocouples that 38 00:03:31,660 --> 00:03:35,740 go into our test chamber to measure ambient air, intake and exhaust air for the power 39 00:03:35,740 --> 00:03:41,900 supply and measures the connectors to make sure that they don't melt. We don't publish 40 00:03:41,900 --> 00:03:45,740 our temperature results yet, but it is something that we're working toward. Finally, there's 41 00:03:45,740 --> 00:03:51,660 the crown jewel of our tester, our road in Schwartz NXO 58 oscilloscope. She's got eight 42 00:03:51,660 --> 00:03:57,020 analog channels, 16 digital channels, which is nice, but the real benefit of this scope 43 00:03:57,020 --> 00:04:02,620 is the ease of programming and the nice waveforms you get to look at. Truthfully, we know we're 44 00:04:02,620 --> 00:04:08,140 kind of using the Ferrari of oscilloscopes to go get groceries, but that can be worth it 45 00:04:08,140 --> 00:04:14,220 when you acquire this equipment and sometimes it takes years to finally start using it, which, 46 00:04:14,220 --> 00:04:20,780 yeah, yeah, what's the deal with that? Okay, I do know the deal with that. The chroma, 47 00:04:20,780 --> 00:04:26,380 it's not exactly user friendly or documented. Making matters worse, the programming is very 48 00:04:26,380 --> 00:04:31,100 similar to assembly code, which turns the automation process for our testing from a 49 00:04:31,100 --> 00:04:36,540 two month project to a 10 month project. For poor Lucas, thankfully, I didn't have to do it. 50 00:04:37,180 --> 00:04:41,660 But now that it's finally done, we can figure out what power supplies are good, 51 00:04:41,660 --> 00:04:46,380 and which ones are not so good, which brings us to an important question. 52 00:04:46,380 --> 00:04:52,620 What makes a power supply good? Well, at its most basic level, a good power supply must adhere to 53 00:04:52,620 --> 00:04:59,420 the ATX specification and deliver these voltages to your copy regardless of the kind of load that 54 00:04:59,420 --> 00:05:05,820 it's under or the input power conditions. Now, we don't follow exactly the ATX prescribed test 55 00:05:05,820 --> 00:05:13,020 plan, but we try wherever it's practical and our tests are heavily inspired by ATX. The good news 56 00:05:13,020 --> 00:05:19,260 is pretty much everything we've tested, even the ones that failed, has pretty much operated within 57 00:05:19,260 --> 00:05:25,980 the ATX specification, until it didn't. But the bad news is we have seen some big differences 58 00:05:25,980 --> 00:05:31,740 in efficiency. In simple terms, the efficiency of a power supply is the percentage of the power 59 00:05:31,740 --> 00:05:36,060 that you draw from the wall that gets turned into useful energy that your computer can use. 60 00:05:36,620 --> 00:05:41,820 Now, many manufacturers market their power supplies with a little 80 plus certification that you'll 61 00:05:41,820 --> 00:05:47,580 find on the box. But in our testing, there were two power supplies whose measured efficiency 62 00:05:47,580 --> 00:05:56,940 didn't match their ratings. The Corsair SF850L and the FSP Vita 850GM. Both of them actually 63 00:05:56,940 --> 00:06:03,660 exceeded their rated efficiency, a result that is validated actually by ClearResult's own 64 00:06:03,660 --> 00:06:11,260 80 plus testing. Why would anybody market their unit as 80 plus gold when they could have labeled 65 00:06:11,260 --> 00:06:17,660 it 80 plus platinum instead? When we talked to Corsair and FSP about this, we got similar answers 66 00:06:17,660 --> 00:06:23,420 from both of them. These power supplies, it turns out, were designed to be gold efficiency power 67 00:06:23,420 --> 00:06:28,540 supplies, but then their engineers went and did a bit too good of a job and they exceeded their 68 00:06:28,540 --> 00:06:34,300 target. But here's the thing, even though their tested units exceeded the target, they're not 69 00:06:34,300 --> 00:06:39,580 confident that every single unit off the line is going to hit platinum efficiency. So they decided 70 00:06:39,580 --> 00:06:45,020 they'd rather be conservative with their ratings. Also, in the case of Corsair, the design team 71 00:06:45,020 --> 00:06:50,700 had just plumb finished the box and label designs with that gold rating on it, so they ran with that. 72 00:06:51,020 --> 00:06:57,980 We verify more than just 80 plus though, which honestly is kind of falling out of fashion, and part 73 00:06:57,980 --> 00:07:05,260 of the reason for that is 80 plus only factors in efficiency from 20% to 100% load. So instead, 74 00:07:05,260 --> 00:07:11,900 we measure all the way from 2% load to 110% so that you know how your power supply is going to 75 00:07:11,900 --> 00:07:18,620 perform when your computer is asleep, all the way up to when it is, oops, slightly overloaded. Now, 76 00:07:18,620 --> 00:07:23,740 I can't say that it'll make a big difference on your power bill if you pick up a C850 or C1000 77 00:07:23,740 --> 00:07:30,700 from NZXT, both of which have outstanding 2% load power efficiency, LTT hat pros off to them. 78 00:07:30,700 --> 00:07:37,020 What I will say is that this expanded testing scope helps us to separate the overbuilt units from 79 00:07:37,020 --> 00:07:43,580 the underbuilt ones. While testing at 110%, we have registered four power supply deaths total, 80 00:07:43,580 --> 00:07:50,700 including this one, which is not a good result. In fairness to these power supplies, 110% of 81 00:07:50,700 --> 00:07:56,540 rated power is out of spec, and we perform some of these tests with the input voltage and frequency 82 00:07:56,540 --> 00:08:03,180 being slightly off to simulate non-ideal conditions. But the thing is, with all of the protections 83 00:08:03,180 --> 00:08:08,940 that a power supply is supposed to have, an overpower or an overcurrent event really shouldn't 84 00:08:08,940 --> 00:08:14,620 result in outright failure. But sometimes it does. The GameMax power supply that we featured in the 85 00:08:14,620 --> 00:08:19,420 thumbnail, the one that Lucas is looking at here spectacularly failed during this very test. 86 00:08:20,060 --> 00:08:24,780 The AC source on the tester went into protection mode, and when Lucas popped the power supply 87 00:08:24,780 --> 00:08:30,860 out of the test chamber, he wisely fired up his phone camera, plugged the unit into the wall, and 88 00:08:31,660 --> 00:08:39,100 yeah, you don't want that going on inside your PC. But, I mean, Linus and Lucas, 89 00:08:39,100 --> 00:08:45,900 that was probably just one bad unit, right? No. See, here's the thing. Anytime a power supply dies 90 00:08:45,900 --> 00:08:52,300 during our testing, we obtain another one. And almost every time, the second unit has also died. 91 00:08:54,620 --> 00:08:59,660 Across our overcurrent, overpower, and ShortCircuit protection testing, we've found six 92 00:08:59,660 --> 00:09:06,700 power supplies that have gone to sleep permanently, securing our coveted We Cannot Recommend Award. 93 00:09:06,700 --> 00:09:12,860 See, while catastrophic failure protection is technically a recommended part of the ATX 94 00:09:12,860 --> 00:09:18,460 specification, we feel that any power supply that you spend good money on should do three things. 95 00:09:19,100 --> 00:09:24,780 One, it should protect you. We can't have your house, and therefore you, lighting on fire. 96 00:09:24,780 --> 00:09:29,820 Number two, it should protect your computer. We can't have your motherboard, and therefore 97 00:09:29,820 --> 00:09:35,500 your computer, lighting on fire. And three, it should protect itself. We can't have your power 98 00:09:35,500 --> 00:09:41,340 supply, and therefore your power supply, lighting on fire, producing molten material, or making 99 00:09:41,340 --> 00:09:48,700 startling noises. Another surprisingly effective killer has been our brownout test, which has 100 00:09:48,700 --> 00:09:53,820 claimed the lives of these three power supplies. This test is pretty simple. You know when the 101 00:09:53,820 --> 00:09:59,260 lights flicker briefly? Well, we're testing to see how long that needs to last before your 102 00:09:59,260 --> 00:10:04,380 computer will shut down. We're not even looking at surges or anything like that. It just turns out 103 00:10:04,380 --> 00:10:09,420 that some power supplies really don't like power getting turned off briefly and then back on. 104 00:10:10,140 --> 00:10:15,660 A surprising non-killer, though, is our excursion testing, where we hit the power supply with a 105 00:10:15,660 --> 00:10:23,900 150% load for one millisecond and then a 200% load for 0.1 milliseconds to see how it'll handle 106 00:10:23,900 --> 00:10:31,900 something, looking at you GPUs, briefly drawing way more than it should. While this hasn't killed 107 00:10:31,900 --> 00:10:37,340 anything yet, there are differences in performance in these tests. So if you plan on overclocking, 108 00:10:37,340 --> 00:10:41,100 you're going to want to look for our power supply with all green check marks here. 109 00:10:41,820 --> 00:10:45,180 Now that we know what we're testing for, let's set up a test. 110 00:10:45,180 --> 00:10:49,180 Why don't you explain step one? Because it's not obvious to me looking at it. 111 00:10:49,180 --> 00:10:52,940 Yeah, so the first thing we do, we unbox it, find out any of the manufacturer's claims, 112 00:10:52,940 --> 00:10:56,940 and write a lot of it down for metadata. That later gets uploaded to the website. Then we go 113 00:10:56,940 --> 00:11:00,540 through input and output ratings for the power to make sure that we're loading it correctly, 114 00:11:00,540 --> 00:11:04,460 and that the loads are properly portioned to it and configured for that. So we're not 115 00:11:04,460 --> 00:11:07,900 nailing it on one rail and then leaving another one alone. And then we go through and we decide 116 00:11:07,900 --> 00:11:11,660 what tests are run on it, which is exactly the same for all of our standardized testing that we 117 00:11:11,660 --> 00:11:15,020 go up on PSU circuit in the lab's website. And then we take kind of an inventory of all the 118 00:11:15,020 --> 00:11:18,860 cables that comes with it and quick measurements of the length. So we have that data as well. 119 00:11:18,860 --> 00:11:23,020 And then we can just run this cell in Python and some code I've written on the back end 120 00:11:23,020 --> 00:11:26,940 generates a whole mess of data and files that we can run off of. 121 00:11:27,580 --> 00:11:31,340 I said step one, you gave me what sounded like at least like half a dozen steps. That's great, 122 00:11:31,340 --> 00:11:34,140 that's efficient. Step one. 80 plus platinum, this guy. 123 00:11:35,100 --> 00:11:41,260 It's not even titanium, I think titanium is better. Next, it's time for Lucas to turn on the tester and the scope. 124 00:11:45,420 --> 00:11:48,620 There's a lot of buttons. Yeah, now I can even reach most of them. 125 00:11:49,180 --> 00:11:53,340 Next up, we shove our power supply into the chamber and reposition our thermal couples 126 00:11:53,340 --> 00:11:58,540 according to the exact physical dimensions of the unit. Then it's time to plug into this guy, 127 00:11:58,540 --> 00:12:01,980 connector board, interface board. I don't know what it's called, but it doesn't matter, 128 00:12:01,980 --> 00:12:07,100 but shout out CSonic for sending this thing over to us because it saves a ton of time. 129 00:12:07,100 --> 00:12:11,820 We can plug in as many connections as you could possibly find on a modern power supply 130 00:12:11,820 --> 00:12:17,580 and the best part is this was some special work from Lucas, some tedious work from Lucas. 131 00:12:17,580 --> 00:12:22,060 We take our measurements right at the bottom of where everything plugs in, 132 00:12:22,060 --> 00:12:27,420 which is important because otherwise the length of these wires could affect our readings and we 133 00:12:27,420 --> 00:12:32,380 don't want that. Oh, there's one more thing. We have a thermocouple that we plug right into the 134 00:12:32,380 --> 00:12:37,020 12 volt high power. You know, the one that has a reputation for being kind of melty. 135 00:12:38,620 --> 00:12:42,620 Once everything's hooked up, we start our tests at zero degrees Celsius, 136 00:12:42,620 --> 00:12:48,860 then we move to 20 degrees Celsius and finally 40 degrees. Everything is automated except the 137 00:12:48,860 --> 00:12:54,300 temperature of this chamber. We could automate that, but it's actually nice to have a few check-in 138 00:12:54,300 --> 00:12:58,860 times to make sure that everything is looking good with the results. Our hardest tests are only 139 00:12:58,860 --> 00:13:05,660 done at 20 degrees Celsius. 16 hours later we get this, which as you can tell needs some processing, 140 00:13:05,660 --> 00:13:11,020 but fortunately that's automated by some code that Lucas wrote. Then from there we get a script 141 00:13:11,020 --> 00:13:16,140 template for PSU circuit videos that gets filled in with the results that are sent to Emily. 142 00:13:16,140 --> 00:13:20,700 For the voice in the videos, we use 11 Labs, an AI text-to-speech dubbing service, 143 00:13:20,700 --> 00:13:25,580 and specifically we use Brian, because we like his voice the most. It can take a couple of tries 144 00:13:25,580 --> 00:13:29,820 to get a good read from Brian, and I need to balance how natural he sounds with how accurate 145 00:13:29,820 --> 00:13:33,820 his reading is. Some more niche phrases like ATX tend to trip him up. 146 00:13:38,540 --> 00:13:43,420 Each video has a template with parts of the voiceover that don't change in blue down here, 147 00:13:43,420 --> 00:13:48,540 and then I insert the generated voice above that in green. Then the footage and graphics get added 148 00:13:48,620 --> 00:13:52,700 in, and after two to three hours of work we have a video. Now if you've ever watched PSU 149 00:13:52,700 --> 00:13:57,100 circuit, you're probably wondering, why don't you guys just have a human voice the videos, 150 00:13:57,100 --> 00:14:02,460 and then have a human fully edit it? And I get it, I would genuinely love to do that. 151 00:14:02,460 --> 00:14:09,180 But let me put it this way, so far on this channel we have made an eye-watering $138 in AdSense, 152 00:14:09,900 --> 00:14:16,060 $138, plus a few bucks here and there from private internet access VPN affiliates. 153 00:14:16,060 --> 00:14:20,700 And that's across all of the videos, so the economics of humans doing that stuff, 154 00:14:21,180 --> 00:14:25,580 they just don't make any sense. But that doesn't mean that the channel's going anywhere. 155 00:14:25,580 --> 00:14:30,300 We are committed to continuing to test power supplies and continuing to upload these videos, 156 00:14:30,300 --> 00:14:35,900 because the data isn't just valuable to viewers of the videos, it's also incredibly valuable to 157 00:14:35,900 --> 00:14:41,900 LTT's ability to do build videos that feature power supplies that we can truly stand behind. 158 00:14:41,900 --> 00:14:46,700 Also, if you don't like the AI voice, you're not alone, but you can just read the articles at 159 00:14:46,700 --> 00:14:52,860 LTTLabs.com. You'll even find some extra data there. You might even find this segue to our sponsor. 160 00:14:52,860 --> 00:14:57,180 If you guys enjoyed this video, why not check out the one we did on a 3,000 watt power supply 161 00:14:57,180 --> 00:15:02,380 where you can see what our testing capabilities used to be.