WEBVTT

00:00:00.160 --> 00:00:09.120
stop no matter the price you should never buy a secondhand GPU that was used

00:00:06.080 --> 00:00:10.960
for mining it could be as much as 10

00:00:09.120 --> 00:00:17.199
slower for each year it's spent mining cryptocurrencies

00:00:13.360 --> 00:00:19.520
like ethereum or dogecoin

00:00:17.199 --> 00:00:23.359
at least that's what the rumor that has been spreading ever since pallet

00:00:21.039 --> 00:00:29.039
microsystems a large manufacturer of gpus allegedly warned benchmark.pl about

00:00:26.880 --> 00:00:34.960
the supposed dangers of buying used gpus but hold on a minute

00:00:32.160 --> 00:00:39.920
a GPU manufacturer only makes money when you buy a new GPU of course they would

00:00:37.920 --> 00:00:44.800
want to warn you off buying a cheaper secondhand one so is there any truth to

00:00:42.239 --> 00:00:49.280
this to find out we contacted an acquaintance of mine a local full-time

00:00:47.520 --> 00:00:53.760
crypto miner yeah i know

00:00:51.360 --> 00:00:56.239
i know but he's otherwise a totally cool guy

00:00:54.559 --> 00:01:03.280
so we got in touch with him and we borrowed three gpus that have been

00:00:58.960 --> 00:01:04.159
mining 24 7 365 for one

00:01:03.280 --> 00:01:09.840
two and three years and i'll be comparing

00:01:06.960 --> 00:01:15.119
their performance to our closest available equivalent gpus that have done

00:01:12.640 --> 00:01:19.439
pretty much nothing but sit on the shelf in our storeroom this is an important

00:01:17.520 --> 00:01:23.520
video because when the inevitable GPU price crash does happen it'll determine

00:01:22.240 --> 00:01:30.400
whether it's worth it to pull the trigger on a used GPU to finally turn

00:01:26.640 --> 00:01:32.960
the tables on those filthy filthy miners

00:01:30.400 --> 00:01:36.479
oh oh no offense but by the way thanks for lending me these cards

00:01:34.799 --> 00:01:40.240
and thanks to world of tanks for sponsoring this video world of tanks is

00:01:38.320 --> 00:01:44.880
the free to play epic online combat multiplayer game that features tanks use

00:01:42.880 --> 00:01:48.840
invite code tank mania at the link below to get some great bonuses

00:01:55.360 --> 00:02:01.920
we've actually talked about performance degradation before in this video where i

00:01:59.439 --> 00:02:06.560
compared a 10 year old gtx 480 that had been quite heavily used to a brand new

00:02:04.320 --> 00:02:10.640
inbox one and that time around our conclusion was that if it's still

00:02:08.399 --> 00:02:15.360
running at the same clock speed its performance will be identical to the day

00:02:12.800 --> 00:02:19.200
it was new however if the silicon has degraded to the point where it can't run

00:02:17.360 --> 00:02:22.400
at those clock speeds it's gonna artifact or black screen rather than

00:02:21.440 --> 00:02:26.720
somehow lose performance so why are we

00:02:24.959 --> 00:02:31.840
revisiting this topic i'm so glad you asked modern graphics

00:02:28.959 --> 00:02:36.080
cards ever since the gtx 680 no longer run at a fixed clock speed and it's

00:02:34.319 --> 00:02:40.400
common practice for manufacturers to advertise both a base clock speed as

00:02:38.400 --> 00:02:44.800
well as a boost clock speed for what they would expect the GPU to reach under

00:02:42.239 --> 00:02:49.280
typical conditions and this is where things get really interesting all other

00:02:47.360 --> 00:02:54.560
things being equal the performance of a processor like a GPU is directly

00:02:52.319 --> 00:03:00.640
determined by how many cycles it performs per second and with nearly a 20

00:02:58.000 --> 00:03:06.800
percent difference between the base and boost clocks of say NVIDIA's rtx 3080

00:03:04.080 --> 00:03:11.360
it's conceivable to imagine a card that still works but with dramatically worse

00:03:10.080 --> 00:03:18.480
performance so let's meet our cards shall we starting

00:03:14.560 --> 00:03:20.720
with the ASUS tough rtx 3080 10g gaming

00:03:18.480 --> 00:03:24.640
now we were really lucky here in that we happened to have

00:03:23.040 --> 00:03:29.440
in the box to give you some idea how little it's been used a tough 3080

00:03:27.200 --> 00:03:34.159
gaming card that is identical to the one that has been running for 10 months of

00:03:31.200 --> 00:03:37.040
continuous use mining cryptocurrency now that is less than a year

00:03:36.239 --> 00:03:42.080
but if the rumored claims from palette are to be believed it should already have

00:03:40.560 --> 00:03:47.280
dropped a measurable amount of performance compared to this otherwise

00:03:44.720 --> 00:03:50.239
identical card let's get it on the test bench

00:03:48.400 --> 00:03:53.840
because we're comparing like for like that is to say that both graphics cards

00:03:52.239 --> 00:03:58.879
in each round of our testing are using the same GPU our setup can be very

00:03:56.640 --> 00:04:04.000
straightforward all we're going to do is run furmark until we reach a plateau in

00:04:01.519 --> 00:04:08.000
temperatures then we're going to log the GPU clock speed for a period of one

00:04:05.760 --> 00:04:13.360
minute and take the average that'll tell us about what the card is boosting to

00:04:10.720 --> 00:04:17.120
one down and while our results seem fine we won't know for sure until we compare

00:04:15.280 --> 00:04:20.959
to a card that doesn't have a layer of dust all over the fan

00:04:18.639 --> 00:04:25.280
here comes GPU number two divided by 1950 equals

00:04:23.360 --> 00:04:32.080
oh well our mining 3080 did end up slower but it

00:04:29.280 --> 00:04:35.280
turns out it was only by point eight percent so

00:04:33.440 --> 00:04:38.800
that is well within the margin of difference that you might expect just

00:04:36.720 --> 00:04:43.199
picking up two brand new thirty eighties off the shelf and running them against

00:04:40.639 --> 00:04:47.280
each other but that doesn't prove degradation isn't real let's move on to

00:04:45.440 --> 00:04:54.720
something that's been running for quite a bit longer representing our

00:04:50.639 --> 00:04:58.000
approximately two years is this rtx 2070

00:04:54.720 --> 00:05:01.199
gaming z8 gig from MSI it's actually

00:04:58.000 --> 00:05:03.759
only been mining for about 18 months but

00:05:01.199 --> 00:05:08.560
in theory if it's supposed to lose 10 performance per year this one should be

00:05:06.000 --> 00:05:12.240
quite a bit slower than an unused card okay i had planned to show you guys the

00:05:10.080 --> 00:05:16.880
results for the 2070s after i had run both cards but this first one is so

00:05:14.720 --> 00:05:21.759
interesting that i couldn't wait in spite of it being rated only for a boost

00:05:19.199 --> 00:05:26.240
clock of 1830 megahertz this MSI card is still running at well

00:05:24.240 --> 00:05:30.639
over 1900 in fermark definitely interesting even

00:05:28.639 --> 00:05:34.560
if i'm not sure how valid this data point will be because while we did

00:05:32.560 --> 00:05:41.360
manage to get perfectly matching cards for our 3080 test for our 2070 we were

00:05:38.639 --> 00:05:45.280
not quite as lucky i mean we'd have loved to just buy a perfect match for

00:05:43.840 --> 00:05:51.919
the one that was given to us by the miner but the problem with that is that most used

00:05:48.639 --> 00:05:54.560
gpus on the market are so called because

00:05:51.919 --> 00:05:57.919
well they've been used and the whole point of our test is that we want to

00:05:56.080 --> 00:06:02.639
compare to something that is in like new conditions so even if we found a GPU

00:06:00.000 --> 00:06:08.319
that someone said hadn't been touched we'd be just relying on their word as

00:06:04.960 --> 00:06:09.039
opposed to any kind of empirical data

00:06:08.319 --> 00:06:15.840
so what we decided to do was go with a non-matching GPU that we know has spent

00:06:13.759 --> 00:06:20.479
most of its life on a shelf even though it is rated for a lower boost clock if

00:06:18.960 --> 00:06:24.639
anything then this should make it even easier for our mining card to maintain

00:06:22.639 --> 00:06:31.120
its performance edge and this card also ran well above its

00:06:28.720 --> 00:06:35.199
rated boost clock so i guess what that means is that our test

00:06:33.280 --> 00:06:39.600
doesn't tell us really anything about performance degradation due to mining

00:06:37.120 --> 00:06:44.080
but rather that NVIDIA's sophisticated GPU boost technology has never really

00:06:41.680 --> 00:06:48.240
cared much about what the rated clock speed from the board partner is so

00:06:46.160 --> 00:06:52.240
both cards are behaving as they should and even their relative performance

00:06:50.639 --> 00:06:57.600
given that they're both rated for about 100 megahertz difference in boost

00:06:54.880 --> 00:07:02.639
is exactly what we would expect if they were brand new of course

00:07:00.000 --> 00:07:08.479
that doesn't mean that it's bs because we still have one more test a

00:07:05.039 --> 00:07:12.160
gtx 1060 that has been mining for four

00:07:08.479 --> 00:07:13.919
years okay i was way off in the intro

00:07:12.160 --> 00:07:18.160
for this generation both of our cards are rated for the same clock speeds but

00:07:15.840 --> 00:07:21.039
with different cooler designs though both of them should help them reach

00:07:19.520 --> 00:07:25.599
pretty much the full potential of the silicon this a non-tech review from 2016

00:07:23.680 --> 00:07:29.280
shows a founder's card like this one performing within about five percent of

00:07:27.440 --> 00:07:35.360
an ASUS strix card that has a beefier cooler so if mining for four years did

00:07:32.240 --> 00:07:38.400
anything to our zotac card's performance

00:07:35.360 --> 00:07:39.680
it should be pretty evident in the

00:07:38.400 --> 00:07:45.160
results let's take them down we're looking at an average of 18.35

00:07:45.840 --> 00:07:48.240
and

00:07:48.960 --> 00:07:56.879
conclusion 1847 versus 1830

00:07:55.120 --> 00:08:03.919
basically the same so okay then using a GPU still doesn't

00:08:01.440 --> 00:08:09.599
appreciably affect the day-to-day performance of said GPU that is unless

00:08:07.280 --> 00:08:14.960
pallet is admitting that their cards die really quickly or something i don't know

00:08:12.240 --> 00:08:18.960
probably not so it's purely binary then it still works or it doesn't work

00:08:16.960 --> 00:08:21.520
anymore go ahead and buy mining cards end of story right

00:08:20.639 --> 00:08:26.720
well not quite it's clear from their condition that

00:08:24.319 --> 00:08:32.159
while these gpus might have been heavily used for mining they've also either been

00:08:29.440 --> 00:08:37.760
very regularly cleaned or kept in a basically dust free environment and

00:08:34.800 --> 00:08:42.159
because modern gpus boost higher in some cases much higher when they run at lower

00:08:39.760 --> 00:08:47.279
temperatures the condition of the cooler is critical to the performance of the

00:08:44.080 --> 00:08:50.240
graphics card so a broken or clogged up

00:08:47.279 --> 00:08:55.519
or otherwise hobbled heatsink could easily cause a 10 performance hit or

00:08:53.200 --> 00:08:59.040
more that's why it is so important to carefully inspect the condition of the

00:08:57.200 --> 00:09:02.720
cooler of a graphics card before you make a purchase they can be a real

00:09:00.640 --> 00:09:06.320
nightmare to track down second hand if you need to replace them and depending

00:09:05.120 --> 00:09:11.760
on how much you're spending the remaining lifespan of the GPU before it

00:09:08.959 --> 00:09:16.240
reaches that binary doesn't work state could be a major concern as well i mean

00:09:14.160 --> 00:09:20.080
if you're paying half the price for the same performance brand new

00:09:18.000 --> 00:09:26.480
i say yolo it and buy another one if it dies but if you're only saving a few bucks

00:09:23.279 --> 00:09:28.800
well i'd recommend this video about why

00:09:26.480 --> 00:09:33.200
electronics die where luke explains how even components with

00:09:30.560 --> 00:09:38.160
no moving parts can be gradually worn out by invisible forces like thermal

00:09:35.920 --> 00:09:42.800
expansion and contraction and even the physical flow of electron through the

00:09:40.320 --> 00:09:47.760
tiny copper wires in a microprocessor with that said a quality GPU would have

00:09:45.839 --> 00:09:52.959
been built to last much longer than its warranty period and in my opinion the

00:09:50.080 --> 00:09:57.920
flood of used gpus that is undoubtedly coming could be the tide change the pc

00:09:55.440 --> 00:10:01.519
gaming community has been waiting for so the bottom line

00:09:59.200 --> 00:10:05.519
know the risks but don't let anyone scare you off of a good deal with

00:10:03.440 --> 00:10:09.920
baseless fud and don't let me scare you off our

00:10:07.279 --> 00:10:14.720
sponsor world of tanks world of tanks is a free to play multiplayer online game

00:10:11.920 --> 00:10:18.640
that is well free to play but skills win the battles not money it

00:10:17.040 --> 00:10:21.600
features over 600 tanks from the mid-20th century that are made to be

00:10:20.160 --> 00:10:26.640
historically accurate including legendary shermans churchills tigers and

00:10:24.480 --> 00:10:30.240
more and has 40 battle arenas that feature historic locations all over the

00:10:28.399 --> 00:10:34.480
globe there are a hundred million players worldwide so why are you waiting

00:10:32.640 --> 00:10:38.560
play on your own and master the art of armored combat or join team clashes and

00:10:36.399 --> 00:10:44.000
climb the leaderboard with invite code tank mania you'll get the excelsior tier

00:10:41.040 --> 00:10:48.720
5 tank 250 000 credits seven days of premium access plus three rental takes

00:10:45.920 --> 00:10:53.680
for 10 battles each if you guys enjoyed this video you might enjoy our uh

00:10:50.720 --> 00:10:56.640
turning a mining GPU into a gaming GPU where we had that headless card we

00:10:55.040 --> 00:10:59.519
managed to get it running that one was pretty cool