WEBVTT 00:00:00.000 --> 00:00:07.440 new gpus are expensive oh ah oh my money 00:00:04.560 --> 00:00:13.679 bone but the media says that buying a secondhand GPU is risky oh it's a bad 00:00:10.740 --> 00:00:20.699 idea well as a ruthless deal hunter I disagree and to prove my point I bought 00:00:16.500 --> 00:00:23.820 19 used gpus to run through our Gauntlet 00:00:20.699 --> 00:00:27.539 of over 1500 tests settling this 00:00:23.820 --> 00:00:30.359 argument once and for all every one of 00:00:27.539 --> 00:00:34.440 these gpus was sourced anonymously from 17 different sellers on eBay and we 00:00:32.759 --> 00:00:40.739 looked at everything from clock speeds to Performance to thermals we did run 00:00:38.340 --> 00:00:45.300 into some pitfalls but with our help there's no reason for you to be afraid 00:00:42.960 --> 00:00:50.879 so when the time comes to adopt your rescue GPU you should be able to 00:00:47.520 --> 00:00:53.460 recognize and maybe even overcome any 00:00:50.879 --> 00:00:57.960 past abuse and with the money that you save buying used maybe you can check out 00:00:55.500 --> 00:01:01.320 something from our sponsor pulse way pulseway is a mobile first remote 00:00:59.460 --> 00:01:06.600 monitoring and management tool built with msps and it teams in mind to learn 00:01:04.199 --> 00:01:11.100 more about how you can see and control your network from anywhere check out the 00:01:09.060 --> 00:01:14.939 link in the video description it's always the video description when we 00:01:13.020 --> 00:01:19.260 bought these cards last year we took a shotgun approach trying to get a mixture 00:01:17.220 --> 00:01:23.939 of different brands at a variety of price points the only requirement was 00:01:21.720 --> 00:01:28.439 that at some point they had to have been desirable for Mining and hard to find 00:01:25.920 --> 00:01:34.080 for gamers so we ended up with some Radeon 5700 series carts these were 00:01:31.680 --> 00:01:37.200 super power efficient and lower demand compared to the NVIDIA competition 00:01:35.579 --> 00:01:41.340 making them some of the most profitable cards for mining on the green side of 00:01:39.180 --> 00:01:45.540 things 30 60s were popular thanks to their High vram to price ratio and not 00:01:43.860 --> 00:01:51.000 to mention the failure of NVIDIA's foolish attempt to limit hash rates and 00:01:47.880 --> 00:01:53.159 as for 30 70s and 3080s well they were 00:01:51.000 --> 00:01:57.540 just plain powerful now whenever you're buying used step one is to talk to the 00:01:55.920 --> 00:02:02.159 seller as much as people like to paint crypto miners as 00:01:59.420 --> 00:02:06.540 monsters and some of them are the majority of used GPU sellers are just 00:02:04.740 --> 00:02:10.800 regular people who saw a way to make some extra cash and would happily tell 00:02:08.759 --> 00:02:15.660 you about the card's history step two then is to inspect the product for 00:02:12.900 --> 00:02:21.300 obvious issues physical damage wear and tear scratches or in the case of a GPU 00:02:18.300 --> 00:02:23.580 any loose components be they fans or 00:02:21.300 --> 00:02:28.620 capacitors I'd recommend at this stage giving the GPU a good shake for real 00:02:25.800 --> 00:02:33.660 then you want to investigate the source of any of the noises that you 00:02:31.200 --> 00:02:37.800 hear now for obvious reasons it would be best to do this before handing over your 00:02:36.120 --> 00:02:42.120 money like if you were to meet up in person through Facebook Marketplace but 00:02:40.020 --> 00:02:45.900 eBay buyer protection is really good these days meaning that you can do it 00:02:43.980 --> 00:02:50.940 after the fact if you use a good Marketplace just make sure to report any 00:02:48.420 --> 00:02:55.800 issues with the condition of the product before the time limit runs out now most 00:02:53.400 --> 00:02:59.940 of our cards arrived looking pretty good other than maybe needing a quick blast 00:02:57.840 --> 00:03:04.200 of compressed air which seems to support the theory that I've seen many times 00:03:01.379 --> 00:03:08.160 that miners actually are financially incentivized to take good care of their 00:03:06.540 --> 00:03:13.739 cards in hopes of maximizing their lifespans however we also got a couple 00:03:11.040 --> 00:03:18.180 that were in questionable condition and that how do I put this 00:03:16.080 --> 00:03:22.980 um kind of smelled it has kind of a vinegar smell to it and then there's 00:03:20.040 --> 00:03:28.680 that MSI 5700 XT with the rattle to it but that otherwise looks good of course 00:03:26.159 --> 00:03:34.260 looks is only enough if you're planning to use your new gpus as bookends we want 00:03:31.560 --> 00:03:37.140 to know how they perform the best way to determine if their performance has been 00:03:35.700 --> 00:03:42.420 affected by their time in the mines would be to measure these gpus against 00:03:39.659 --> 00:03:48.659 identical brand new ones unfortunately however these are older cards and mint 00:03:45.480 --> 00:03:50.519 condition ones don't really exist so the 00:03:48.659 --> 00:03:55.200 control lines that you see on our charts are going to be from lightly used gpus 00:03:52.920 --> 00:03:58.500 from our stock that have spent most of their lives on the Shelf it's worth 00:03:56.760 --> 00:04:03.239 noting here by the way that we expect some degree of variation between these 00:04:00.659 --> 00:04:09.780 cards and the goal here is not to find the winner but rather to see if our used 00:04:06.360 --> 00:04:11.819 gpus are performing within spec a quick 00:04:09.780 --> 00:04:15.599 and dirty way to check this is MSI combustor it takes just a couple of 00:04:14.040 --> 00:04:20.100 minutes so it's the kind of thing that you could ask a seller to run before you 00:04:17.699 --> 00:04:25.199 hand over your money if your GPU is significantly degraded its clock speeds 00:04:22.440 --> 00:04:30.660 may fail to reach a normal level and right out of the gate it's obvious that 00:04:28.259 --> 00:04:36.600 some of our samples do have some problems the maximum and average clock 00:04:33.479 --> 00:04:39.419 speeds of our 3080s and our 3070s check 00:04:36.600 --> 00:04:44.520 out nicely with some variants but nothing that would indicate an issue 00:04:41.280 --> 00:04:47.880 meanwhile our 3060 results show us our 00:04:44.520 --> 00:04:51.419 first obvious problematic GPU the eagle 00:04:47.880 --> 00:04:52.800 OC its average clocks are far are lower 00:04:51.419 --> 00:04:59.280 than they should be we're going to put a pin in that for now though and move on to the 5700 Series where we see some 00:04:56.940 --> 00:05:03.180 very odd behavior as well we want to keep a particularly close eye on the 00:05:00.960 --> 00:05:06.780 performance of this Sapphire pulse card now there are a few factors that 00:05:05.040 --> 00:05:11.880 determine the Boost clock speed of a card the main ones are the type of work 00:05:09.479 --> 00:05:15.840 the card is performing the quality of the silicon and the quality of the 00:05:14.220 --> 00:05:20.880 cooler now we've standardized our workload for now combustor and there's 00:05:18.720 --> 00:05:25.860 no way for us to affect silicon quality so the first troubleshooting step for 00:05:23.220 --> 00:05:31.139 any underperforming card is to take its temperature maybe it's got a wee little 00:05:27.479 --> 00:05:33.539 fever and there it is the eagle OC 3060 00:05:31.139 --> 00:05:38.460 that we flagged before is thermally throttling with temps that far exceed 00:05:35.759 --> 00:05:43.080 any of our other 30 60s if we wanted to fix it getting those temps under control 00:05:40.080 --> 00:05:45.720 would be a very good start but this is a 00:05:43.080 --> 00:05:50.100 second hand buyer's guide not a GPU repair guide so what you need to do is 00:05:48.300 --> 00:05:54.479 return that thing and let the seller figure it out now now let's turn our 00:05:51.660 --> 00:05:58.620 attention to the 5700 xt's that Sapphire pulse that had the lower clocks seems to 00:05:56.699 --> 00:06:03.180 have its thermals well in check so throttling is unlikely it also doesn't 00:06:01.020 --> 00:06:06.900 have clock speeds that are substantially limited which seems to rule out the 00:06:05.100 --> 00:06:11.460 possibility that our seller was using a modified BIOS optimized for mining stuff 00:06:09.180 --> 00:06:16.020 like this is why combustor can only be considered a quick and dirty tool so for 00:06:13.800 --> 00:06:21.000 this one we are going to need to go a little bit deeper and look at the 00:06:17.880 --> 00:06:23.400 amazing new underwear on ltdstore.com oh 00:06:21.000 --> 00:06:28.020 and also at some real world games we settled on six games at three 00:06:25.319 --> 00:06:32.400 resolutions and at risk of spoiling the conclusion a little bit here the vast 00:06:30.000 --> 00:06:37.620 majority of our results were neck and neck with our control cards now the lab 00:06:34.800 --> 00:06:44.220 hasn't defined specific values yet for what counts as equal but we figure for 00:06:40.800 --> 00:06:46.199 the purposes of this video 58 60fps 00:06:44.220 --> 00:06:52.319 that's going to be an imperceptible difference so as long as our mining gpus 00:06:48.900 --> 00:06:54.419 managed to be 97 10 percent or higher 00:06:52.319 --> 00:06:59.400 than the Control Card we're calling them in Spec starting with the 3080s 00:06:56.940 --> 00:07:04.319 interestingly enough once we averaged out our results we noticed that we had a 00:07:01.259 --> 00:07:06.120 problem with our control GPU while its 00:07:04.319 --> 00:07:10.800 average frame rates were fine we found that its lows were lower than our mining 00:07:08.940 --> 00:07:15.419 gpus which is going to manifest as stuttering and choppy animations during 00:07:13.380 --> 00:07:19.020 demanding sections of the game kind of funny that the lightly used card ended 00:07:17.160 --> 00:07:22.560 up being the borked one here we'll talk a little bit more about that later for 00:07:20.460 --> 00:07:26.340 now the similarities between our mining cards are still enough for us to 00:07:24.120 --> 00:07:31.919 demonstrate that they at least are fine or in the case of our EVGA model a bit 00:07:29.099 --> 00:07:36.300 more than fine I miss you already EVGA for the 3070s we see every card 00:07:34.259 --> 00:07:40.500 performing as expected with little to no variance across our samples and we've 00:07:38.460 --> 00:07:46.020 got similar results with the 3060s except for the ventis 2x which landed a 00:07:43.860 --> 00:07:50.099 little below our expectations thankfully it looks like this one was pretty easy 00:07:47.819 --> 00:07:53.460 to explain and rectify our event is targets clock speeds that are about 00:07:51.660 --> 00:07:58.380 three percent lower than our reference card and when lightly overclocked it 00:07:55.979 --> 00:08:03.060 easily comes back in line now it is possible that the reason for this lower 00:08:00.780 --> 00:08:07.440 Target clock speed is that the Silicon is slightly degraded but it's more 00:08:05.460 --> 00:08:12.479 likely that this particular card just wasn't as good from the start I'm going 00:08:09.660 --> 00:08:16.860 to give it a pass as for the 5700 XTS our Sapphire pulse ended up performing 00:08:14.580 --> 00:08:22.020 just fine in real world applications what a relief but wait some of you are 00:08:19.919 --> 00:08:29.160 screaming we know the cores are fine what about the vram Linus you need to 00:08:25.560 --> 00:08:31.620 test the vram so we did we used Ida 64's 00:08:29.160 --> 00:08:36.539 GP GPU Benchmark to see if the modules have lost any performance or started 00:08:34.080 --> 00:08:43.919 throwing errors and every single card looks good except the eagle OC rest in 00:08:40.979 --> 00:08:51.019 peace get your money back for this one uh unless you can't for some reason in 00:08:46.980 --> 00:08:54.180 that case can we save the eagle OC sadly 00:08:51.019 --> 00:08:57.660 no there are some things that can be 00:08:54.180 --> 00:09:01.140 overcome a bad fan can be replaced a bad 00:08:57.660 --> 00:09:02.940 Tim can be replaced hey and if a mining 00:09:01.140 --> 00:09:07.380 BIOS has been flashed for example the folks over at Tech PowerUp have an 00:09:04.620 --> 00:09:11.700 amazing database of GPU bioses that you can simply use to flash a stock one back 00:09:09.480 --> 00:09:16.500 onto your card but a bing bada boom but if you encounter artifacting which looks 00:09:14.040 --> 00:09:21.000 a little something like this or outright errors like we just saw the cold hard 00:09:18.959 --> 00:09:25.680 truth is that you probably cannot do anything which means then that there is 00:09:24.000 --> 00:09:30.120 some truth to the warnings you've seen about used cards whether it's 00:09:27.720 --> 00:09:35.580 intentional or not there are people out there trying to sell broken gpus as 00:09:32.760 --> 00:09:38.700 working ones but I still disagree with a lot of the points made by my fellow 00:09:37.019 --> 00:09:42.839 members of the media let's work through them together first the idea that you 00:09:41.100 --> 00:09:47.459 will suffer some kind of performance shortfall from the heavy use of mining 00:09:45.240 --> 00:09:51.360 was always silly and betrayed a fundamental misunderstanding of how 00:09:49.080 --> 00:09:55.620 these products even work in some ways the heating and then cooling Cycles not 00:09:53.880 --> 00:10:01.500 to mention the higher clock speeds that are typical in gaming workloads are more 00:09:58.080 --> 00:10:04.680 damaging to a GPU so a heavily mined on 00:10:01.500 --> 00:10:07.500 GPU could fail but then so could a 00:10:04.680 --> 00:10:11.940 lightly gamed on one and until either of them does it will work as well as the 00:10:09.779 --> 00:10:15.000 other and now that you're armed with the tools that you need to weed out the bad 00:10:13.680 --> 00:10:19.380 from the good gpus follow a very similar failure 00:10:17.700 --> 00:10:24.360 pattern to other electronics meaning that the lemons usually die within weeks 00:10:22.260 --> 00:10:29.160 or months of manufacture within their warranty period and those non-lemons 00:10:26.580 --> 00:10:32.339 tend to live long and happy lives until they really really don't the next 00:10:30.839 --> 00:10:37.860 misconception that I've seen thrown around is that not purchasing used cards 00:10:35.279 --> 00:10:43.260 is somehow like like sticking it to crypto Miners And while I fully support 00:10:40.740 --> 00:10:48.060 withholding demand until prices fall to acceptable levels a stubborn refusal to 00:10:46.140 --> 00:10:52.680 buy one at all can only lead to two possible outcomes either someone else 00:10:50.519 --> 00:10:58.380 will buy them or they will be scrapped as E-Waste to avoid storing them I mean 00:10:55.260 --> 00:11:00.300 I get the anger but by not purchasing a 00:10:58.380 --> 00:11:03.899 used GPU you're not only depriving yourself of something that you want 00:11:01.560 --> 00:11:09.300 you're potentially encouraging more harm to the environment from my point of view 00:11:05.760 --> 00:11:12.480 every non-obsolete GPU that is not in a 00:11:09.300 --> 00:11:14.760 gaming PC is a brand new GPU that needs 00:11:12.480 --> 00:11:19.200 to be manufactured in its place and if you think think about it buying new are 00:11:17.220 --> 00:11:24.060 you really morally Superior your purchase encourages large money grubbing 00:11:21.600 --> 00:11:27.839 corporations to keep producing more gpus out of our dwindling Rare Earth minerals 00:11:25.980 --> 00:11:32.760 and these are the same big corporations that happily drop shipped literal tons 00:11:30.600 --> 00:11:38.220 of gpus to large-scale mining facilities reaping record high profits during the 00:11:34.620 --> 00:11:40.380 pandemic but like Calpers right NVIDIA 00:11:38.220 --> 00:11:43.980 and AMD were the biggest scalpers of all leaving us at the point then where the 00:11:42.120 --> 00:11:47.180 only moral thing to do is to simply not purchase a GPU at all 00:11:48.000 --> 00:11:54.660 but come on what are we gonna do not play Modern Warfare 2 so the right thing 00:11:52.620 --> 00:11:59.279 to do would be to purchase a used GPU then I mean only you can say I've long 00:11:57.600 --> 00:12:03.300 grappled with my effect on increasing consumerism and how my very existence 00:12:01.500 --> 00:12:08.339 encourages some of the worst aspects of capitalism but in my defense I've also 00:12:06.240 --> 00:12:11.880 always been an outspoken advocate of keeping older hardware and service 00:12:09.899 --> 00:12:16.200 rather than buying things that you don't need maybe it's because I believe in the 00:12:13.920 --> 00:12:21.180 importance of the first two r's you know the reviews and reduce ones or maybe 00:12:19.260 --> 00:12:27.300 it's because I'm just a cheapskate maybe it's that I'm lucky enough to be able to 00:12:22.980 --> 00:12:29.220 afford to gamble on used products but in 00:12:27.300 --> 00:12:34.380 the grand scheme of things even if they do fail every once in a while I don't 00:12:31.440 --> 00:12:38.160 see how you can afford not to buy used I mean considering that we weren't being 00:12:35.820 --> 00:12:41.700 particularly Savvy with our purchases we didn't haggle we didn't bid on auctions 00:12:40.019 --> 00:12:46.320 we just bought whatever was available for asking price and then compared it to 00:12:44.160 --> 00:12:52.740 buying new at the time and we still saved about 22 percent that is not too 00:12:50.459 --> 00:12:57.600 shabby even if you removed the price of our broken GPU from that calculation our 00:12:55.320 --> 00:13:01.560 savings are still 18 and a half percent compared to if we bought this stuff new 00:12:59.579 --> 00:13:05.040 again we could have saved more money with patients than Hagley but as for 00:13:03.480 --> 00:13:08.700 whether it makes sense for you it's going to depend on your risk tolerance 00:13:06.540 --> 00:13:12.360 but like come on everything's a risk except buying a GPU with a transferable 00:13:10.860 --> 00:13:17.399 warranty by the way ebta and anyway I'll tell you this much 00:13:15.360 --> 00:13:21.540 as someone who's been Hands-On with more computer hardware than most people will 00:13:18.959 --> 00:13:25.920 ever even lay eyes on I've seen brand new stuff die for no reason just as 00:13:23.639 --> 00:13:30.660 often as I've seen second hand pieces of crap survive a brutal drop life is a 00:13:28.200 --> 00:13:35.519 crap shoot and then you die just like I killed the original Segway 00:13:32.639 --> 00:13:40.260 we had to our sponsor zoho1 do you own or manage a business if so Zoho one 00:13:37.920 --> 00:13:45.060 could make your life easier Zoho 1 takes the Essential Elements of any business 00:13:41.940 --> 00:13:47.639 like accounting marketing HR and then 00:13:45.060 --> 00:13:51.240 combines them into one unified operating system build your own website from the 00:13:49.560 --> 00:13:54.839 ground up and maintain it with intuitive customization and personalization 00:13:52.620 --> 00:13:59.760 options send out purchase orders create marketing campaigns manage shift 00:13:56.639 --> 00:14:01.500 scheduling all in just a few clicks you 00:13:59.760 --> 00:14:05.040 can track metrics and use data to make key decisions to increase revenue and 00:14:03.600 --> 00:14:09.180 then once you've got everything set up you can start automating your processes 00:14:07.139 --> 00:14:13.019 to never worry about them again if you're working on the go that's no 00:14:10.680 --> 00:14:16.740 problem Zoho 1 includes mobile apps so that you can run your business from 00:14:14.339 --> 00:14:20.339 anywhere so don't wait take control of your business the way that you see fit 00:14:18.240 --> 00:14:25.139 with Zoho one follow the link below to try it for free for 30 days with no 00:14:22.920 --> 00:14:28.800 credit card required if you like this video why not check out our video about 00:14:26.940 --> 00:14:32.880 the benchmarking automation that allowed this project to happen it's actually 00:14:30.240 --> 00:14:36.779 come along way since then because we had to work on it a lot to actually do all 00:14:34.500 --> 00:14:39.200 these benchmarks but the basic ideas there