WEBVTT

00:00:00.080 --> 00:00:05.520
we've got no shortage of videos on our channel about building expensive gaming

00:00:03.600 --> 00:00:09.040
machines but what about those of you who would rather spend your money on

00:00:07.279 --> 00:00:12.639
lttstore.com well for you guys we've got value

00:00:10.800 --> 00:00:16.480
machines and you can get a lot of gaming bang for your buck these days

00:00:14.639 --> 00:00:20.640
this right here is a test bench made up of the components from our 900 value

00:00:18.880 --> 00:00:27.519
gaming build guide a little while ago it's an 8 core 3800 x processor from AMD

00:00:23.760 --> 00:00:29.199
a 2060 super from NVIDIA and this b450

00:00:27.519 --> 00:00:32.880
board from MSI but

00:00:30.640 --> 00:00:38.160
what if i told you that i could get twice

00:00:35.120 --> 00:00:41.360
as many cores for my money and all i

00:00:38.160 --> 00:00:45.360
would have to do is scour aliexpress and

00:00:41.360 --> 00:00:47.600
ebay for refurbished xeon cpus and ecc

00:00:45.360 --> 00:00:56.000
memory and this the gian tech jinxa x79 dual socket

00:00:53.280 --> 00:00:59.600
bizarre chinese motherboard that definitely they're gonna be around to

00:00:57.600 --> 00:01:02.559
honor that three-year warranty on this video is brought to you by glasswire

00:01:00.960 --> 00:01:06.159
with glasswire you can instantly see your current and past network activity

00:01:04.320 --> 00:01:10.490
on your Windows or Android device use offer code Linus at the link below to

00:01:07.600 --> 00:01:15.480
get 25 off

00:01:18.479 --> 00:01:24.560
now just because we've got twice as many cores doesn't mean we're going to necessarily have twice as much

00:01:22.960 --> 00:01:29.040
performance so we're going to run some benchmarks on our ryzen system here to

00:01:27.360 --> 00:01:33.280
give us a point of reference for how much value we're actually getting

00:01:30.479 --> 00:01:38.560
because what's interesting about the components that i chose is that once you

00:01:35.360 --> 00:01:42.560
factor in the CPU motherboard and memory

00:01:38.560 --> 00:01:44.400
they are very nearly the same price

00:01:42.560 --> 00:01:51.600
so let's talk through the xeons we chose these are Intel's e5 2667 v2s and they

00:01:49.200 --> 00:01:58.000
are a little bit on the older side so these are ivy bridge and they're lga

00:01:54.159 --> 00:01:59.759
2011 and what that means is that

00:01:58.000 --> 00:02:04.240
you're not going to have the same kinds of aggressive boost clocks that you

00:02:01.680 --> 00:02:08.879
would find on modern processors even server ones so their base is about 3.3

00:02:06.799 --> 00:02:13.200
gigahertz and their boost is about 4 gigahertz but

00:02:10.720 --> 00:02:16.640
because Intel has been so stagnant when it comes to performance per clock

00:02:15.040 --> 00:02:22.080
improvements over the last four or five years we are expecting this

00:02:19.200 --> 00:02:26.400
to perform reasonably similarly to a modern CPU running at the same clock

00:02:24.319 --> 00:02:30.239
speed so when you factor in that we can have two of these for the price of a

00:02:28.080 --> 00:02:34.319
single ryzen 3800 it starts to look maybe pretty

00:02:32.400 --> 00:02:39.200
compelling our RAM is not that interesting these are just some ddr3

00:02:36.239 --> 00:02:43.120
1600 ecc server memory modules these are readily available on ebay but our

00:02:41.920 --> 00:02:48.640
motherboard is very interesting so you can find this

00:02:45.840 --> 00:02:53.920
on newegg for about 300 or if you're a lot smarter than that you

00:02:51.280 --> 00:02:59.519
can find it on aliexpress for about a hundred and thirty dollars and like that

00:02:57.200 --> 00:03:05.280
other bizarro chinese motherboard that we checked out a little while ago it

00:03:01.519 --> 00:03:08.080
appears as though jingxia is actually

00:03:05.280 --> 00:03:12.879
manufacturing these out of refurbished parts so it seems like the chipsets that

00:03:10.800 --> 00:03:18.959
they're using in them because it can't be x79 x79 was a single

00:03:16.560 --> 00:03:25.360
CPU-only chipset so it appears the chipsets are c-series ones that were

00:03:21.519 --> 00:03:29.040
pulled off of either dead or discarded

00:03:25.360 --> 00:03:31.200
server boards and then plunked onto

00:03:29.040 --> 00:03:36.159
this new design that they've done and it's a pretty interesting one so we've

00:03:32.640 --> 00:03:38.480
got dual pci express 16x slots here m.2

00:03:36.159 --> 00:03:44.480
which was not actually that common on older boards in the x79 era a couple

00:03:40.879 --> 00:03:47.519
more pci express 1x slots are two 2011

00:03:44.480 --> 00:03:50.080
slots and then up to eight memory slots

00:03:47.519 --> 00:03:54.640
giving us hopefully the full quad channel memory experience per CPU even

00:03:52.959 --> 00:04:00.239
though these are color coded as though they would be a dual channel for

00:03:58.000 --> 00:04:05.120
whatever reason wow i gotta say the i o on this thing is not impressive

00:04:02.799 --> 00:04:10.720
ps2 okay i mean it never hurts to have that but only two usb 3 ports dual

00:04:08.400 --> 00:04:15.599
gigabit Ethernet and that super basic three port 5.1 audio

00:04:14.400 --> 00:04:19.199
okay now something to note is that i've seen

00:04:17.600 --> 00:04:24.479
a fair number of people speculating online that oh because this motherboard

00:04:21.519 --> 00:04:32.160
runs an x79 chipset maybe you could put like extreme editions or core i7s in it

00:04:28.160 --> 00:04:34.320
but no the x79 chipset is just outright

00:04:32.160 --> 00:04:38.160
false advertising and if you want to run two cpus in this thing you will

00:04:36.000 --> 00:04:42.720
absolutely need to get xeons they're the only ones that have the extra qpi link

00:04:40.720 --> 00:04:47.680
enabled that allows the two cpus to communicate with each other and that is

00:04:45.199 --> 00:04:52.240
exactly one of the performance issues that i'm expecting to run into here

00:04:49.440 --> 00:04:56.560
because in a perfect world in a workload that supports it

00:04:53.919 --> 00:05:00.240
we're going to be able to leverage 16 cores

00:04:57.680 --> 00:05:05.440
against a mere eight course for our AMD system but the problem with that is that

00:05:03.440 --> 00:05:10.880
many workloads even heavily multi-threaded ones are not what's

00:05:07.520 --> 00:05:13.600
called pneuma aware so that is to say

00:05:10.880 --> 00:05:18.160
that they are not able to leverage memory that is directly connected to

00:05:15.759 --> 00:05:23.520
this CPU for a workload that's running on this one in very very simplistic

00:05:20.960 --> 00:05:27.840
terms so what that means is that you will basically in your worst case

00:05:25.199 --> 00:05:32.639
scenario only be able to use one of your cpus so that would be eight cores

00:05:30.000 --> 00:05:36.080
against eight cores and these are much much weaker cores

00:05:34.320 --> 00:05:41.039
all right my render finished in 12 minutes 28 seconds and i got sick so i'm

00:05:39.759 --> 00:05:48.000
gonna sound like this for the rest of the video let's go ahead and run a couple of games then we can compare

00:05:45.039 --> 00:05:50.840
against our ebay machine i mean that's the good news is i can

00:05:49.440 --> 00:05:54.639
talk like this in shadow of the tomb raider here

00:05:53.280 --> 00:05:58.720
i've gone with what i think are reasonable graphic settings for a

00:05:56.880 --> 00:06:04.840
machine of this budget so we're running at 1440p all high motion blur off and

00:06:02.880 --> 00:06:09.199
we're not indulging in msaa and you can see that we're right in

00:06:07.280 --> 00:06:15.280
the range of what i would expect for a solid gaming experience 80 90 100 FPS

00:06:12.800 --> 00:06:19.199
world builder type rts games tend to be quite CPU bound so i've fired up a i'd

00:06:18.240 --> 00:06:24.479
say a mid-level scenario in nano 1800 looking

00:06:21.680 --> 00:06:28.479
at about 75 FPS again this is medium settings this time with a little bit of

00:06:25.840 --> 00:06:31.280
anti-aliasing because this game tends to kind of have a lot of jaggies in it oh

00:06:30.319 --> 00:06:38.080
okay well whatever that's fine i got my numbers i'm using a pair of nhu 12s's

00:06:36.000 --> 00:06:44.240
for cooling but it's worth noting that any lga 2066 or 2011 compatible cooler

00:06:41.919 --> 00:06:47.360
will work on these cpus so if you're trying to keep your pricing

00:06:45.840 --> 00:06:52.800
close between the platforms you might want to get something a little bit more

00:06:50.639 --> 00:06:56.720
value conscious but these are really quick to set up on a test bench and

00:06:54.240 --> 00:07:00.240
stuff so i really like using them you know i've always felt like rocking

00:06:58.319 --> 00:07:05.280
two cpus is kind of like having like a gigantic hood scoop on your car

00:07:03.039 --> 00:07:09.599
like it it could be an indicator that there's lots of performance under the

00:07:06.960 --> 00:07:11.919
hood or it could be mostly for show and

00:07:10.560 --> 00:07:15.520
i guess we're about to find out which one we're dealing with today

00:07:13.919 --> 00:07:19.360
installing this on the bench is actually an excellent reminder of some of the

00:07:17.199 --> 00:07:24.240
things other than improved performance that newer hardware yields so you can

00:07:21.919 --> 00:07:28.000
see that we actually need two 8-pin CPU power connectors and the motherboard is

00:07:26.880 --> 00:07:31.440
physically quite a lot larger than our 3800x

00:07:30.400 --> 00:07:36.800
platform so yeah they might cost the same but this

00:07:34.880 --> 00:07:40.639
one's gonna cost more to run and it's gonna take up more space

00:07:39.199 --> 00:07:46.720
i don't know maybe it's so much faster it doesn't matter though let's find

00:07:42.840 --> 00:07:48.639
out just play pour it in and

00:07:46.720 --> 00:07:51.280
good luck everybody oh wow

00:07:50.080 --> 00:07:58.639
didn't even get a chance to go in the bio so we're kicking right to Windows boys and girls well i can't believe how

00:07:55.280 --> 00:08:00.080
painless this was we booted right up 3.3

00:07:58.639 --> 00:08:05.440
gigahertz base 16 cores 32 threads

00:08:03.759 --> 00:08:09.840
64 gigs around everything is just working so let's go ahead and uh what

00:08:07.840 --> 00:08:13.199
did we start with blender let's start with blender

00:08:11.120 --> 00:08:17.039
now what i want to see here is how many of our cores we end up using

00:08:15.280 --> 00:08:21.039
because i've seen blender struggle across multiple pneuma nodes

00:08:19.360 --> 00:08:26.560
but this is not that day or whatever he says

00:08:24.000 --> 00:08:30.319
we are getting 100 utilization an unfortunate thing though is we are

00:08:28.080 --> 00:08:36.000
running merely at base clock remember these older cpus didn't really have all

00:08:33.360 --> 00:08:40.240
core boost the base clock was the clock speed it ran at when you were pushing

00:08:38.000 --> 00:08:45.040
every single core i think this thing is gonna slay

00:08:42.880 --> 00:08:51.760
that AMD platform oh wow that is not the margin of victory

00:08:47.839 --> 00:08:53.600
that i expected 11 46 i mean it's faster

00:08:51.760 --> 00:08:58.480
but given that we've got double the core count and we were running at 100 across

00:08:56.560 --> 00:09:04.080
all of our cores running quad channel memory even though it's ddr3

00:09:00.800 --> 00:09:06.160
i was expecting more than a 42 second

00:09:04.080 --> 00:09:11.360
margin of victory here oh that does not bode well we're getting

00:09:08.480 --> 00:09:17.839
100 utilization in cinebench as well but it's clear to me that these cores are

00:09:14.480 --> 00:09:19.279
not as fast as our AMD ones i was not

00:09:17.839 --> 00:09:25.600
expecting this our gaming results are actually looking

00:09:21.600 --> 00:09:27.440
really similar so far what the heck

00:09:25.600 --> 00:09:31.519
like this is fine i am kind of impressed actually i haven't

00:09:30.000 --> 00:09:36.480
seen the final score like i don't know what our 95 percentiles are or anything like that

00:09:34.800 --> 00:09:42.560
but so far this looks shockingly viable

00:09:40.000 --> 00:09:48.240
all right let's have a look here okay so shockingly our average FPS is

00:09:45.839 --> 00:09:51.680
only like six percent behind that is not what i was expecting after how much

00:09:49.839 --> 00:09:57.279
slower these cores are but if you check the CPU game FPS the 95th

00:09:55.760 --> 00:10:01.839
percentile it's way lower it's more like 12

00:10:00.000 --> 00:10:07.680
13 lower i don't know i'm doing this math in my head but it's over 10 lower

00:10:04.720 --> 00:10:11.760
so that's more like what i expected just goes to show you guys though that modern

00:10:09.680 --> 00:10:14.320
games are really coming a long way in terms of utilizing more core as you can

00:10:13.519 --> 00:10:19.360
see we're utilizing at least seven of our

00:10:17.200 --> 00:10:23.760
cores pretty substantially although check this out remember i talked about

00:10:22.079 --> 00:10:27.440
programs not being aware of multiple pneuma nodes

00:10:25.200 --> 00:10:31.680
these cores were clearly untouched while only these ones were utilized

00:10:29.600 --> 00:10:35.760
oh lordy now this

00:10:33.440 --> 00:10:42.959
is more like what i'd expect as a worst case scenario 42 to 43 FPS looking at

00:10:39.920 --> 00:10:45.120
exactly the same scene so what this

00:10:42.959 --> 00:10:49.360
tells us is that shadow of the tomb raider was doing an excellent job as we

00:10:47.200 --> 00:10:53.839
saw of spreading out the load across course helping to compensate for the

00:10:51.519 --> 00:10:56.480
poor per core performance of this machine

00:10:54.800 --> 00:11:01.760
this game not so much so when you consider that we

00:10:59.120 --> 00:11:07.120
have twice as many cores here versus our ryzen 7 and our multi-threaded

00:11:04.880 --> 00:11:12.079
applications still ran at about the same speed it stands to reason that if we're

00:11:09.360 --> 00:11:17.920
extremely CPU bound we're gonna run at something close to just over half of the

00:11:15.519 --> 00:11:21.360
speed ouch let's just have a look at our task

00:11:19.360 --> 00:11:26.240
manager here you can see here it's actually utilizing more cores to a

00:11:24.800 --> 00:11:31.600
lesser degree but it's still running like hot garbage

00:11:29.360 --> 00:11:37.680
so our conclusions today then are pretty interesting one is that if you have a

00:11:35.040 --> 00:11:42.640
particular workload that benefits from lots and lots of cores refurbished

00:11:40.079 --> 00:11:46.959
server or old workstation hardware on a used marketplace like ebay can be a

00:11:45.279 --> 00:11:51.200
fantastic bet you're going to give up some power consumption but we did get

00:11:49.360 --> 00:11:57.360
noticeably better performance in both blender and cinebench and if you manage

00:11:54.320 --> 00:11:59.200
to find some ddr3 dimms in a dumpster or

00:11:57.360 --> 00:12:04.320
something like that which does happen from time to time that can tip the

00:12:02.000 --> 00:12:08.079
balance significantly in favor of a solution like this

00:12:05.920 --> 00:12:12.800
but what we also know is that if you want something that's going to perform

00:12:09.920 --> 00:12:16.959
predictably like consistently across applications be they single threaded or

00:12:15.279 --> 00:12:21.680
multi-threaded something that is architecturally newer even though it

00:12:19.360 --> 00:12:25.839
does come with the premium of being brand new is definitely the way to go

00:12:24.399 --> 00:12:29.680
not to mention that of course when you buy new hardware you get a warranty to

00:12:28.079 --> 00:12:35.600
go along with it so if you have the exact right use case

00:12:32.160 --> 00:12:38.720
honestly this looks not bad for everyone

00:12:35.600 --> 00:12:41.839
else there's something more modern like

00:12:38.720 --> 00:12:45.120
a 3800x

00:12:41.839 --> 00:12:46.720
or there's our sponsor freshbooks

00:12:45.120 --> 00:12:49.760
freshbooks is the small business accounting software that's custom built

00:12:48.399 --> 00:12:55.200
for how you want to work and it's a simple way to be more productive more

00:12:52.079 --> 00:12:57.519
organized and to get paid faster cash

00:12:55.200 --> 00:13:02.160
flow is so important for a small business so whether it's people

00:13:00.079 --> 00:13:05.519
not responding to your invoices in time freshbooks lets you check to see if

00:13:03.839 --> 00:13:09.920
they've actually seen the invoice or whether it's just if it's so complicated

00:13:07.680 --> 00:13:13.920
and tedious that you don't feel like sitting down and dealing with it that

00:13:11.920 --> 00:13:18.160
can be a huge problem so try out freshbooks you can create and send

00:13:16.079 --> 00:13:21.920
professional looking invoices in just 30 seconds you can set up online payments

00:13:20.079 --> 00:13:25.519
with just a couple of clicks to get paid up to four days faster and you don't

00:13:23.920 --> 00:13:30.000
have to take my word for it you can try it for free at freshbooks.com

00:13:28.320 --> 00:13:33.600
tech tips for 30 days we're gonna have that linked in the video description

00:13:31.920 --> 00:13:36.959
just make sure you put Linus tech tips in the how did you hear about us section

00:13:35.760 --> 00:13:41.360
so thanks for watching guys we've actually got some more great server

00:13:38.800 --> 00:13:46.399
content upcoming check this out i'm gonna be building myself a new storage

00:13:43.040 --> 00:13:50.000
server we got that 512 gigs of RAM 32

00:13:46.399 --> 00:13:52.399
epic cores and over a hundred terabytes

00:13:50.000 --> 00:13:55.639
of NVMe storage get subscribed