WEBVTT

00:01:14.880 --> 00:01:21.640
hey sir they found it you better get over here now

00:01:24.840 --> 00:01:29.520
oh oh nevermind yeah no sorry false

00:01:28.000 --> 00:01:34.479
alarm uh yeah i thought they were gonna take the fast one

00:01:31.600 --> 00:01:34.479
yeah you have a good night

00:01:36.880 --> 00:01:45.759
this is the pc pro 32 000 worth of pc

00:01:43.119 --> 00:01:50.079
hardware so fast it'll knock your socks right out of your sandals let's take a

00:01:47.520 --> 00:01:54.560
look inside at the uh beautiful power supply provided by our

00:01:52.479 --> 00:01:59.520
sponsor c sonic who makes great power supplies for everything from entry-level

00:01:56.159 --> 00:02:02.520
gaming rigs to servers to monsters like

00:01:59.520 --> 00:02:02.520
this

00:02:10.640 --> 00:02:16.319
the 2019 mac pro is a thing of beauty and there's no denying that it's

00:02:15.280 --> 00:02:19.720
powerful ours here is equipped with 28 processing

00:02:18.879 --> 00:02:25.360
cores 768 gigabytes of memory it's got an AMD

00:02:22.800 --> 00:02:30.480
Radeon vega pro duo a graphics card so exclusive you can only get it in this

00:02:27.760 --> 00:02:36.480
machine and it's got one small problem macs aren't known for their bang for the

00:02:33.120 --> 00:02:37.360
buck so then even if i had a lot of

00:02:36.480 --> 00:02:42.879
bucks 33 000 of them to be precise here

00:02:40.080 --> 00:02:45.519
would i be better off with one of these or

00:02:44.239 --> 00:02:51.519
one of these with the pc pro our goal was to create a

00:02:48.400 --> 00:02:55.120
machine that beats the mac pro in every

00:02:51.519 --> 00:02:59.440
conceivable way starting with the case

00:02:55.120 --> 00:03:01.599
this inwin 2 2.0 is one of only 300

00:02:59.440 --> 00:03:06.560
units in existence and it's one of the few pc cases that i think is truly

00:03:03.680 --> 00:03:10.239
striking enough to take on the mac pro the outside is covered in five

00:03:08.239 --> 00:03:15.360
millimeter thick tempered glass that has been vacuum sputter coated to create an

00:03:13.040 --> 00:03:19.599
incredible one-way mirror that only shows off your hardware when internal

00:03:17.360 --> 00:03:23.280
lighting is applied the process is a little deeper than we have time for now

00:03:21.200 --> 00:03:27.280
but it's super freaking cool so watch this video from applied science later on

00:03:25.599 --> 00:03:32.560
all of that tempered glass is held in place by an aluminum frame that was

00:03:29.200 --> 00:03:35.440
welded by hand giving it the distinctive

00:03:32.560 --> 00:03:40.879
geometric look that it has now hearing words like welded by hand and

00:03:38.560 --> 00:03:45.120
vacuum sputter coating and one of three hundred you'd probably guess that this

00:03:42.879 --> 00:03:49.440
is worth a pretty penny and at two thousand dollars

00:03:46.640 --> 00:03:55.120
you'd be correct so for that price how good is it to build in not good very bad

00:03:53.680 --> 00:03:58.560
all right so hardware access then

00:03:56.959 --> 00:04:03.599
mac pro one pc pro zero now alex i can't help but

00:04:02.080 --> 00:04:07.040
notice that we've got two monitors hooked up to the pc pro and only one to

00:04:05.840 --> 00:04:10.560
the mac yeah the second one there is just for

00:04:08.959 --> 00:04:14.080
the post sequence so you know that it's actually doing something after you hit

00:04:12.080 --> 00:04:17.519
the power button of course it is i feel like i know how this is going to work

00:04:15.439 --> 00:04:21.359
out but let's do our power on boot time test

00:04:19.359 --> 00:04:25.919
okay fine i'll press that button just give it a second

00:04:24.160 --> 00:04:32.160
okay what did you do i just held it for longer the good news is that you've got

00:04:29.120 --> 00:04:33.120
remote access over ipmi so if you're not

00:04:32.160 --> 00:04:39.520
you know in the same building you can still remote into it over the network and like

00:04:36.960 --> 00:04:44.960
you know update the BIOS or you know restart it or whatever the bad news is

00:04:41.759 --> 00:04:47.600
it outputs to vga first and then

00:04:44.960 --> 00:04:51.600
displayport later all right boot time is admittedly not a strong point for the pc

00:04:49.759 --> 00:04:55.360
pro here so mac pro 2

00:04:52.960 --> 00:04:59.520
pc pro zero admittedly this isn't going very well for us so far it'll get there

00:04:57.759 --> 00:05:04.560
now one of the toughest decisions for this build was actually the CPU

00:05:02.160 --> 00:05:09.520
Threadripper was very tempting because of its excellent single core performance

00:05:07.039 --> 00:05:14.400
overclocking capabilities and official support for Windows 10. the deal breaker

00:05:12.000 --> 00:05:22.560
for us was its maximum memory capacity of 256 gigs now for most applications

00:05:18.479 --> 00:05:25.039
256 gigs of RAM is a lot but the mac pro

00:05:22.560 --> 00:05:29.440
here can be kitted out with up to one and a half terabytes of memory and we

00:05:27.919 --> 00:05:34.919
couldn't allow ourselves to get beaten like that that's why we chose

00:05:32.080 --> 00:05:42.720
against AMD's wishes the epic 7502p a 32 core server processor with

00:05:38.960 --> 00:05:46.960
support for up to two terabytes of RAM

00:05:42.720 --> 00:05:49.199
now it adds 3 847.18

00:05:46.960 --> 00:05:53.840
to the total of our machine here but as far as we can tell it's lack of support

00:05:51.680 --> 00:05:58.479
for Windows 10 is just a validation bottleneck over at AMD and i'm quite

00:05:55.919 --> 00:06:02.639
confident that it will beat the 28 core mac pro in multi-threaded tests even if

00:06:00.960 --> 00:06:06.960
it falls somewhat behind in single threaded tests besides i mean

00:06:04.960 --> 00:06:12.240
i think it's fair to assume that anyone buying a 28 or 32 core processor is

00:06:10.639 --> 00:06:16.400
mostly concerned with multi-threading anyway right the main problem with

00:06:14.080 --> 00:06:20.240
choosing a server processor is that you're usually stuck with a server

00:06:18.080 --> 00:06:26.080
motherboard in our case the taian mainboard s8030gm2ne

00:06:23.840 --> 00:06:29.520
now to be clear we really like it and there's even another version of it that

00:06:27.520 --> 00:06:34.880
has dual onboard 10 gigabit network cards but the thing is we aren't exactly using

00:06:33.120 --> 00:06:41.520
it in its intended application so certain things like fast

00:06:37.759 --> 00:06:43.919
boot times robust io and fancy black

00:06:41.520 --> 00:06:48.960
pcbs have been sacrificed for things like reliability high quality pci

00:06:46.160 --> 00:06:52.560
express gen 4 capable traces and heaps and heaps of storage options let's go

00:06:51.199 --> 00:06:58.240
ahead and open it up that is a thick boy of a panel dang man

00:06:55.919 --> 00:07:02.800
there is something positively delightful about seeing 64 threads in task manager

00:07:01.440 --> 00:07:07.919
and check this out 1020

00:07:04.720 --> 00:07:12.080
gigabytes of memory however keeping a

00:07:07.919 --> 00:07:15.440
terabyte of RAM cooled is uh no laughing

00:07:12.080 --> 00:07:18.639
matter these 128 gig modules that we're

00:07:15.440 --> 00:07:21.120
using from sk hynix run freaking

00:07:18.639 --> 00:07:27.520
hot so we're actually using something that i haven't seen since like the skull

00:07:23.919 --> 00:07:30.560
trail days these active coolers running

00:07:27.520 --> 00:07:33.360
over top of our dual four dim

00:07:30.560 --> 00:07:36.880
banks of memory modules here and i had a small heart attack when i saw that this

00:07:35.199 --> 00:07:41.199
one was resting on the back of our graphics card but actually it's okay

00:07:39.120 --> 00:07:45.199
good job alex you covered it in electrical tape which is definitely an

00:07:43.120 --> 00:07:50.639
appropriate solution for a 30 000 machine do you want to hear why

00:07:48.479 --> 00:07:55.199
we really need those coolers let's just turn these off for a second there's an

00:07:52.479 --> 00:07:58.879
alarm that goes off at 95 degrees it's mad we gotta we got to plug the RAM

00:07:57.360 --> 00:08:03.039
coolers back in all right that's better we're down to

00:08:00.639 --> 00:08:08.160
about 50 degrees now now that alarm might seem obnoxious but

00:08:06.080 --> 00:08:13.520
all the monitoring on this motherboard is part of what allows it to run 24 7 in

00:08:10.960 --> 00:08:18.319
a server type environment besides we got to protect these modules because they're

00:08:15.599 --> 00:08:22.319
over twelve hundred dollars each for a total of nine thousand seven

00:08:20.080 --> 00:08:27.440
hundred and ninety two dollars of memory alone in our system pivoting to our

00:08:25.120 --> 00:08:31.199
least expensive component if we were to 100

00:08:28.639 --> 00:08:37.599
push everything in this system we'd be drawing close to 1300 watts now normally

00:08:34.959 --> 00:08:42.320
we throw a sea sonic prime platinum 1300 watt power supply in rigs just because

00:08:39.839 --> 00:08:47.600
we can it'll run most gaming machines without even turning on the fan here we

00:08:44.959 --> 00:08:52.240
actually need the entire 108 amps on the 12 volt rail and 13 20 watts of

00:08:50.160 --> 00:08:57.040
continuous power now in real world operation you almost never actually run

00:08:55.040 --> 00:09:00.160
every component of a system at a hundred percent so we're not expecting to see

00:08:58.640 --> 00:09:04.959
more than about a thousand watts of power usage at a time partly because of

00:09:02.640 --> 00:09:09.760
how the gpus are configured more on that later but it is considered best practice

00:09:07.760 --> 00:09:15.040
to over spec your power supply a little because they do degrade a little over

00:09:11.600 --> 00:09:17.360
time also upgrades happen and

00:09:15.040 --> 00:09:22.080
a failure especially in this case can destroy a lot of expensive hardware so

00:09:19.680 --> 00:09:27.279
this unit has a 12 year warranty and only took a 270 byte out of our budget

00:09:25.040 --> 00:09:35.360
which sounds totally reasonable in this context back to expensive stuff now this

00:09:30.080 --> 00:09:38.720
right here is a quadro rtx 8000 NVIDIA's

00:09:35.360 --> 00:09:42.480
top GPU with a staggering 48 gigabytes

00:09:38.720 --> 00:09:44.800
of vram and a price tag of 5500

00:09:42.480 --> 00:09:48.000
just a single one of these should be able to put a vega pro duo in

00:09:46.880 --> 00:09:53.440
its place naturally we got two

00:09:51.040 --> 00:09:58.720
but there's more to it than that joining the two together is an envy link bridge

00:09:56.080 --> 00:10:03.600
which is basically a 100 Gigabyte per second data super highway that allows

00:10:01.120 --> 00:10:08.640
the two cards to not just communicate with each other but actually share

00:10:06.399 --> 00:10:15.120
memory giving digital artists data scientists and engineers

00:10:11.080 --> 00:10:16.800
96 gigabytes of effective vram in a

00:10:15.120 --> 00:10:20.800
single workstation that is it'll give them that if their

00:10:18.640 --> 00:10:25.200
boss will sign off on the expense report so we've got two quadros and an envy

00:10:23.279 --> 00:10:28.560
link bridge which means that nvlink should just be

00:10:26.320 --> 00:10:33.440
working right you might think that by looking at NVIDIA's literature the only

00:10:30.720 --> 00:10:37.920
problem is that this guide contains almost zero useful information

00:10:36.320 --> 00:10:43.120
run a command to check if it's working you say how and where do i run this

00:10:40.800 --> 00:10:46.959
command fortunately for us the guys over at puget systems put together a guide

00:10:44.959 --> 00:10:50.959
for how to get nvlink going which they created through trial and error without

00:10:48.640 --> 00:10:55.839
any documentation from NVIDIA first problem when enabling nvlink on

00:10:53.120 --> 00:11:01.760
quadros the display output actually gets disabled so we needed another GPU of the

00:10:59.040 --> 00:11:06.240
same generation or newer in our case this rtx titan here is going to act as

00:11:03.920 --> 00:11:10.640
our display output then we have to do some command prompt stuff to enable

00:11:07.920 --> 00:11:14.720
nv-link on the gpus and after a reboot we can run the nv-link tester that puget

00:11:12.640 --> 00:11:18.800
systems actually made to ensure that it's all working and check the bandwidth

00:11:16.640 --> 00:11:22.160
between the cards huge thanks to those guys for putting all that together i

00:11:20.320 --> 00:11:25.680
have no clue how NVIDIA expects their customers to figure these things out

00:11:23.680 --> 00:11:30.959
otherwise in a nutshell everything i just said means that this rtx titan here

00:11:28.320 --> 00:11:35.200
is just a really pretty golden display output bringing our total spend on

00:11:32.959 --> 00:11:39.279
graphics up to thirteen thousand four hundred and ninety nine dollars

00:11:36.800 --> 00:11:43.839
ridiculous fortunately the rtx titan is also the fastest gaming GPU on the

00:11:41.279 --> 00:11:47.440
market so naturally for our first test we're gonna see just how well these two

00:11:45.680 --> 00:11:51.360
computers that are completely not designed for gaming fare against each

00:11:49.760 --> 00:11:56.240
other in gaming ready for the drag race yeah all

00:11:53.519 --> 00:11:59.600
right run benchmark three two one go they're both running Windows by the way

00:11:57.600 --> 00:12:05.360
just so you guys know that's a lot more FPS yep

00:12:02.560 --> 00:12:09.120
oh i'm sorry yep it's a lot more and you can talk all day about how this is

00:12:06.959 --> 00:12:13.760
gaming and it's not you know for gaming or whatever but more GPU powers more GPU

00:12:12.079 --> 00:12:18.399
power that would be clearly beneficial oh nice

00:12:16.320 --> 00:12:24.320
that was a pretty one-sided fight but a w is a w so it's 2-1 for the mac pro

00:12:22.480 --> 00:12:29.360
let's go for the mac pro's main draw though content creation to test this to

00:12:26.720 --> 00:12:34.639
its fullest our video editor mark who is admittedly a mac guy put together a

00:12:31.839 --> 00:12:40.639
benchmark for us in premiere and final cut pro to really test if the better

00:12:37.600 --> 00:12:43.600
optimized mac pro which does by the way

00:12:40.639 --> 00:12:47.839
have the accelerator card in it or the much more powerful pc pro is able to

00:12:46.079 --> 00:12:53.519
take the win so we've got a multi-platform audio visual render kit

00:12:50.399 --> 00:12:55.120
or markbench as we're calling it so we

00:12:53.519 --> 00:13:00.560
weren't kidding when we said that the i o on this thing is terrible

00:12:57.519 --> 00:13:02.639
it's got two usb 3 ports

00:13:00.560 --> 00:13:07.920
a management network interface that that's not like a network port you can

00:13:04.000 --> 00:13:09.440
actually use vga serial and then two

00:13:07.920 --> 00:13:13.040
gigabit network ports and then it could also have two more 10 gig net reports

00:13:11.040 --> 00:13:18.160
lots of networking that's uh what is that now three one for the mac pro yep

00:13:16.240 --> 00:13:21.279
wait so i see there's a usb card right here is there a reason we didn't install

00:13:19.519 --> 00:13:25.360
this in the bottom slot yeah we bought pretty cheap ones and

00:13:23.120 --> 00:13:30.399
they were causing stability issues mark bench also known as benchmark oh i love

00:13:28.320 --> 00:13:36.800
myself so here it is the afterburner card versus

00:13:33.279 --> 00:13:38.880
two rtx 8000s worth of GPU muscle

00:13:36.800 --> 00:13:44.480
three two one go keep in mind that this is uh multiple

00:13:41.760 --> 00:13:50.320
8k clips animated with you know effects attached to them

00:13:46.959 --> 00:13:52.320
oh so you did the same project yeah in

00:13:50.320 --> 00:13:54.880
both pieces of software i thought we were just doing a simple encode but

00:13:53.680 --> 00:13:59.760
we're not oh this is way more interesting then so

00:13:57.120 --> 00:14:02.959
we're using a hundred percent CPU right now

00:14:00.560 --> 00:14:06.800
this is using 90 GPU

00:14:04.079 --> 00:14:13.600
uh these puppies are up at 80 degrees it is using them perfect okay are you done

00:14:09.680 --> 00:14:17.680
oh yeah oh yeah it's done i'm at 45

00:14:13.600 --> 00:14:19.279
that's thrashed oh dang

00:14:17.680 --> 00:14:22.480
i was not expecting that result it wasn't either the scripts were oh this

00:14:20.959 --> 00:14:26.639
is the first time you've seen it yeah it's the first time i've seen it oh that's hilarious okay the script's

00:14:25.360 --> 00:14:31.360
written as if the mac wins mark bench oh okay

00:14:29.360 --> 00:14:36.079
the problem with a lot of videos about the mac pro though is that people that

00:14:33.279 --> 00:14:41.120
make videos about the mac pro are mostly well video creation professionals in

00:14:38.800 --> 00:14:44.800
real life video editors who might observe that the mac pro is so

00:14:43.040 --> 00:14:49.199
much faster uh make up only a small portion of the

00:14:47.120 --> 00:14:53.760
people that need massive computing power for their work engineers and scientists

00:14:51.760 --> 00:15:00.480
financial analysts medical professionals and many others require this kind of

00:14:57.199 --> 00:15:02.320
power so for them how is the performance

00:15:00.480 --> 00:15:05.920
between these two machines going to differ we're going to find the biggest

00:15:03.839 --> 00:15:10.800
differences in applications that support NVIDIA's cuda GPU compute exclusively

00:15:08.959 --> 00:15:15.760
and will not be able to take advantage of the AMD gpus found in the mac pro for

00:15:13.360 --> 00:15:21.839
instance solidworks visualize only supports cuda so on the mac pro it took

00:15:18.720 --> 00:15:23.279
several times as long as it did on the

00:15:21.839 --> 00:15:29.199
pc pro meaning our score is now even to look at

00:15:27.600 --> 00:15:34.480
some other workloads we're leaning on the spec workstation benchmark

00:15:31.440 --> 00:15:36.959
in many of the CPU tests that are either

00:15:34.480 --> 00:15:43.279
single core dependent or shorter burstier multi-core workloads the mac

00:15:39.279 --> 00:15:46.160
pro comes out ahead so 7-zip blender

00:15:43.279 --> 00:15:50.800
python compiling and simulations that are just out of necessity single core

00:15:48.800 --> 00:15:55.360
like stress analysis because it can't be paralyzed that nets the mac pro 11

00:15:53.440 --> 00:16:00.480
points the only thing though is that the mac pro only won these CPU benchmarks by

00:15:58.160 --> 00:16:06.240
a little bit whereas in anything that required GPU muscle the pc pro

00:16:03.519 --> 00:16:09.680
absolutely wrecked house sometimes doubling the mac pro giving it a total

00:16:08.880 --> 00:16:13.360
of 13 points

00:16:14.800 --> 00:16:20.160
sort of i mean here's the thing guys cheerleading any kind of inanimate

00:16:18.480 --> 00:16:24.399
object is silly and making the right decision is

00:16:22.320 --> 00:16:29.920
always going to come down to your particular workload and budget so let's

00:16:27.519 --> 00:16:34.320
get back to practicality as cool as the pc pro is

00:16:31.440 --> 00:16:39.680
i don't think anybody should buy this i mean we did find a way to spend

00:16:36.880 --> 00:16:45.199
32 000 on a pc but this configuration isn't going to

00:16:42.320 --> 00:16:49.519
excel at any one task like a machine that was built with more focus would

00:16:47.600 --> 00:16:54.560
like say we wanted the fastest rendering machine possible we would sacrifice a

00:16:52.399 --> 00:16:58.639
bunch of the RAM and spend that money on additional GPU muscle maybe we do need

00:16:57.279 --> 00:17:03.199
loads of RAM for big financial calculations well in that case we could

00:17:01.040 --> 00:17:07.360
ditch some GPU and get a server board that doubles our maximum memory capacity

00:17:05.280 --> 00:17:12.160
to four terabytes probably the most egregious waste of money is of course

00:17:09.280 --> 00:17:17.120
the case which is 100 form over function and is 10 times the

00:17:14.640 --> 00:17:21.520
cost of a less flashy and more pragmatic one though in fairness to us we were

00:17:19.520 --> 00:17:26.160
trying to compete with apple here when in rome right and besides at the

00:17:24.079 --> 00:17:32.000
end of the day we achieved exactly what we set out to do look at this thing

00:17:28.480 --> 00:17:32.799
the pc pro is a nod to excess and in

00:17:32.000 --> 00:17:37.440
that it excels absolutely where the mac pro

00:17:35.840 --> 00:17:42.000
destroys your wallet while making massive performance compromises because

00:17:39.200 --> 00:17:46.640
mr cook and mr wong can't sort out their differences the pc pro evaporates your

00:17:44.640 --> 00:17:51.360
bank account while being the most op system around and yeah it doesn't excel

00:17:48.960 --> 00:17:56.000
at one single task and definitely has its own problems but i think it's pretty

00:17:53.440 --> 00:18:00.799
epic pun intended and i want to thank csonic for sponsoring this video and you

00:17:58.880 --> 00:18:05.600
for watching it if you guys are looking for another fun over-the-top mac themed

00:18:03.440 --> 00:18:09.600
pc video maybe check out our hack pro series by the time this is out we should

00:18:07.280 --> 00:18:14.960
be done filming the finale and you're going to be able to binge watch the

00:18:11.120 --> 00:18:14.960
entire thing very soon