馃攳
DON'T Buy a Used Mining GPU! - $h!t Manufacturers Say - YouTube
Channel: unknown
[0]
- Stop!
[0]
No matter the price,
[1]
you should never buy a secondhand GPU
[5]
that was used for mining.
[7]
It could be as much as
10% slower for each year
[11]
it's spent mining cryptocurrencies
[13]
like Ethereum or Dogecoin.
[17]
At least, that's what the rumor
[19]
that has been spreading ever
since Palit Microsystems,
[22]
a large manufacturer of GPUs,
[24]
allegedly warned benchmark.pl
[26]
about the supposed dangers
of buying used GPUs.
[31]
But hold on a minute.
[32]
A GPU manufacturer only makes
money when you buy a new GPU.
[37]
Of course, they would wanna warn you off
[39]
buying a cheaper secondhand one.
[41]
So is there any truth to this?
[42]
To find out, we contacted
an acquaintance of mine,
[46]
a local, full time
[47]
crypto miner.
(fire crackles)
[48]
(crowd chatters angrily)
[49]
Yeah, I know.
[51]
I know.
[52]
But he's otherwise a totally cool guy.
[54]
So we got in touch with him,
and we borrowed three GPUs
[58]
that have been mining 24/7, 365
[62]
for one, two, and three years.
[66]
And I'll be comparing their performance
[68]
to our closest available equivalent GPUs
[72]
that have done pretty much nothing
[74]
but sit on the shelf in our store room.
[76]
This is an important video,
[78]
because when the inevitable
GPU price crash does happen,
[81]
it'll determine whether it's
worth it to pull the trigger
[84]
on a used GPU to finally turn the tables
[87]
on those filthy, filthy miners.
[90]
Oh, no offense, but by the way,
[92]
thanks for lending me these cards.
[94]
And thanks to World of Tanks
for sponsoring this video.
[97]
World of Tanks is the free to play
[99]
epic online combat multiplayer
game that features tanks.
[102]
Use invite code tankmania
at the link below
[105]
to get some great bonuses.
[107]
(upbeat electronic music)
[115]
We've talked about
performance degradation before
[118]
in this video, where I compared
a ten-year-old GTX 480,
[121]
that had been quite heavily
used, to a brand new in-box one.
[125]
And that time around,
our conclusion was that
[127]
if it's still running,
at the same clock speed,
[130]
its performance will be
identical to the day it was new.
[133]
However, if the silicon has degraded
[136]
to the point where it can't
run at those clock speeds,
[138]
it's gonna artifact or black screen,
[141]
rather than somehow lose performance.
[143]
So why are we revisiting this topic?
[147]
I'm so glad you asked.
[148]
Modern graphics cards,
ever since the GTX 680,
[151]
no longer run at a fixed clock speed.
[154]
And it's common practice for manufacturers
[156]
to advertise both a base clock speed,
[158]
as well as a boost clock speed
for what they would expect
[161]
the GPU to reach under typical conditions.
[163]
And this is where things
get really interesting.
[167]
All other things being equal,
[168]
the performance of a
processor, like a GPU,
[171]
is directly determined by how many
[173]
cycles it performs per second.
[176]
And with nearly a 20% difference
[179]
between the base and boost clocks
[181]
of say, NVIDIA's RTX 3080,
[184]
it's conceivable to imagine
a card that still works,
[188]
but with dramatically worse performance.
[191]
So let's meet our cards, shall we?
[194]
Starting with the ASUS
TUF RTX 3080 10G Gaming.
[198]
Now we were really lucky here,
in that we happened to have
[202]
(hand slaps)
[203]
in the box, to give you some
idea how little it's been used,
[206]
a TUF 3080 Gaming card that is identical
[209]
to the one that has been
running for 10 months
[211]
of continuous use mining cryptocurrency.
[214]
Now that is less than a year,
[216]
but if the rumored claims
from Palit are to be believed,
[219]
it should already have dropped
[221]
a measurable amount of performance
[223]
compared to this otherwise identical card.
[226]
Let's get it on the test bench.
[228]
Because we're comparing like for like,
[230]
that is to say that both graphics cards
[232]
in each round of our testing
are using the same GPU,
[235]
our setup can be very straightforward.
[237]
All we're gonna do is run FurMark
[239]
until we reach a plateau in temperatures.
[242]
Then, we're going to
log the GPU clock speed
[245]
for a period of one minute
and take the average.
[247]
That'll tell us about what
the card is boosting to.
[250]
(clock ticking)
One down,
[251]
and while our results seem fine,
[253]
we won't know for sure
until we compare to a card
[256]
that doesn't have a layer
dust all over the fan.
[258]
Here comes GPU number two.
[261]
Divided by 1950 equals.
[263]
Oh.
[265]
Well, our mining 3080 did end up slower,
[269]
but it turns out it was only by 0.8%.
[272]
So that is well within
the margin of difference
[275]
that you might expect just
picking up two brand new 3080s
[278]
off the shelf and running
them against each other,
[281]
but that doesn't prove
degradation isn't real.
[284]
Let's move on to something
[285]
that's been running
for quite a bit longer.
[289]
Representing our, approximately two years
[292]
is this RTX 2070 Gaming Z 8 gig from MSI.
[297]
It's actually only been
mining for about 18 months,
[301]
but in theory, if it's
supposed to lose 10%
[303]
performance per year, this one should be
[306]
quite a bit slower than an unused card.
[308]
Okay, I had planned to show you guys
[310]
the results for the 2070s
after I had run both cards,
[313]
but this first one is so
interesting that I couldn't wait.
[316]
In spite of it being rated
[318]
only for a boost clock of 1830 megahertz,
[321]
this MSI card is still running
at well over 1900 in FurMark
[327]
Definitely interesting,
even if I'm not sure
[329]
how valid this data point will be.
[331]
Because while we did manage to get
[333]
perfectly matching
cards for our 3080 test,
[336]
for our 2070s, we were not quite as lucky.
[340]
I mean, we'd have loved to just
[342]
buy a perfect match for the one
[344]
that was given to us by the miner.
[345]
But, the problem with that
is that most used GPUs
[349]
on the market are so-called,
[350]
because, well, they've been used.
[354]
(audience gasps)
And the whole point
[355]
of our test is that we want
to compare to something
[357]
that is in like new condition.
[358]
So even if we found a
GPU that someone said
[361]
hadn't been touched, we'd be
just relying on their word,
[365]
as opposed to any kind of empirical data.
[368]
So what we decided to do was
go with a non-matching GPU
[372]
that we know has spent most
of its life on a shelf,
[375]
even though it is rated
for a lower boost clock.
[378]
If anything then, this
should make it even easier
[381]
for our mining card to
maintain its performance edge.
[384]
And this card also ran well
above its rated boost clock.
[390]
So, I guess what that means is
that our test doesn't tell us
[394]
really anything about performance
degradation due to mining,
[397]
but rather that Nvidia's
sophisticated GPU boost technology
[401]
has never really cared much
[402]
about what the rated clock
speed from the board partner is.
[405]
So both cards are behaving as they should,
[408]
and even their relative performance,
[410]
given that they're both rated
[411]
for about 100 megahertz
difference in boost,
[414]
is exactly what we would expect
[417]
if they were brand new.
[418]
Of course, that doesn't mean that it's BS.
[422]
Because we still have one more test,
[425]
a GTX 1060 that has been
mining for four years.
[429]
Okay, I was way off in the intro.
[432]
For this generation, both of our cards
[434]
are rated for the same clock speeds,
[435]
but with different cooler designs.
[437]
Though, both of them
should help them reach
[439]
pretty much the full
potential of the silicon.
[441]
This AnonTech review from 2016
[443]
shows a Founders card like this one
[445]
performing within about
5% of an ASUS Strix card
[448]
that has a beefier cooler.
[450]
So, if mining for four years
[452]
did anything to our
Zotac card's performance,
[455]
it should be pretty
evident in the results.
[459]
Let's take them down.
[460]
We're looking at an average of 1835.
[464]
(clock ticking)
[466]
And,
[469]
conclusion.
[470]
1847 versus 1830.
[475]
Basically the same.
[476]
So, okay then.
[479]
Using a GPU still doesn't
appreciatively affect
[483]
the day-to-day performance of said GPU.
[486]
That is, unless Palit is admitting
[488]
that their cards die really
quickly, or something.
[491]
I don't know.
[492]
Probably not.
[493]
So it's purely binary, then.
[495]
It still works, or it
doesn't work anymore.
[497]
Go ahead and buy mining cards.
[499]
End of story, right?
[500]
Well, not quite.
[502]
It's clear from their condition
[504]
that while these GPUs might've
been heavily used for mining,
[507]
they've also either been
very regularly cleaned,
[511]
or kept in a basically
dust-free environment.
[514]
And because modern GPS boost higher,
[517]
in some cases much higher,
[519]
when they're run at lower temperatures,
[520]
the condition of the cooler is critical
[523]
to the performance of the graphics card.
[525]
So a broken, or clogged up,
or otherwise hobbled heat sink
[529]
could easily cause a 10%
performance hit or more.
[534]
That's why it is so to carefully inspect
[536]
the condition of the
cooler of a graphics card
[538]
before you make a purchase.
[539]
They can be a real nightmare
to track down secondhand
[542]
if you need to replace them.
[544]
And depending on how much you're spending,
[546]
the remaining lifespan of the GPU
[548]
before it reaches that
binary doesn't work state
[551]
could be a major concern as well.
[553]
I mean, if you're paying half the price
[556]
for the same performance brand new,
[558]
I say YOLO it, and buy
another one if it dies.
[560]
But if you're only
saving a few bucks, well,
[564]
I'd recommend this video
about why electronics die,
[567]
where Luke explains how even
components with no moving parts
[572]
can be gradually worn
out by invisible forces
[575]
like thermal expansion and contraction,
[577]
and even the physical flow of electrons
[580]
through the tiny copper
wires in a microprocessor.
[582]
With that said, a quality GPU would have
[586]
been built to last much longer
than its warranty period.
[588]
And in my opinion, the flood of used GPUs
[591]
that is undoubtedly coming
could be the tide change
[595]
the PC gaming community
has been waiting for.
[597]
So the bottom line, know the risks,
[600]
but don't let anyone scare you
[602]
off of a good deal with baseless FUD.
[605]
And don't let me scare you off
our sponsor, World of Tanks.
[609]
World of Tanks is a free to
play multiplayer online game,
[612]
that is, well, free to play,
[614]
but skills win the battles, not money.
[617]
It features over 600 tanks
from the mid 20th century
[619]
that are made to be historically accurate,
[621]
including legendary Shermans,
Churchills, Tigers and more,
[625]
and has 40 battle arenas that feature
[627]
historic locations all over the globe.
[629]
There are a hundred
million players worldwide,
[631]
so why are you waiting?
[632]
Play on your own and master
the art of armored combat,
[635]
or join team clashes and
climb the leaderboard
[637]
with invite code tankmania.
[639]
You'll get the Excelsior
tier 5 tank, 250,000 credits,
[643]
seven days of premium access,
[644]
plus three rental tanks
for 10 battles each.
[648]
If you guys enjoyed this video,
[649]
you might enjoy our turning a
mining GPU into a gaming GPU,
[653]
where we had that headless card,
[655]
we managed to get it running.
[656]
That one was pretty cool.
Most Recent Videos:
You can go back to the homepage right here: Homepage