WEBVTT

00:00:00.480 --> 00:00:06.400
so here's the thing 18 cores if you're a performance fiend

00:00:05.040 --> 00:00:12.639
is pretty darn good in spite of the imac

00:00:09.599 --> 00:00:14.480
pro's woefully inadequate cooling

00:00:12.639 --> 00:00:18.320
but come on apple

00:00:15.839 --> 00:00:22.640
is this really the fastest mac that money can buy

00:00:20.000 --> 00:00:26.560
the imac pro i'm glad i asked

00:00:24.480 --> 00:00:29.279
because in fact no no it is not

00:00:27.920 --> 00:00:34.320
how do we know because we and by we i mean

00:00:32.719 --> 00:00:40.320
mostly that guy built the real

00:00:36.079 --> 00:00:43.200
fastest mac on the planet for ourselves

00:00:40.320 --> 00:00:45.840
and it's time for us to take it for a little

00:00:46.000 --> 00:00:50.399
spin get it

00:00:49.120 --> 00:00:54.239
that was bad i don't want to move it that much i don't want to break it

00:00:52.640 --> 00:00:59.440
speaking of things i don't want to break this transition to our sponsor private

00:00:56.559 --> 00:01:04.080
internet access that's exactly what you'll get with pia a safe and protected

00:01:02.160 --> 00:01:07.980
ip and it's got loads of other features as well check it out at the link in the

00:01:05.760 --> 00:01:16.240
video description

00:01:16.240 --> 00:01:21.280
okay so let's begin by going over what

00:01:19.280 --> 00:01:26.960
we've done here this is our asus dominus extreme motherboard

00:01:24.880 --> 00:01:31.360
and hanging off of it we actually have uh

00:01:28.479 --> 00:01:35.600
wait anthony is that two power supplies yeah even though the thousand watts

00:01:33.600 --> 00:01:40.320
sonic prime that's on my bench right now has more than enough wattage

00:01:37.759 --> 00:01:45.840
to handle all of this yeah the actual load of all these vrms during power up

00:01:43.840 --> 00:01:48.880
it's just way too much for it it trips the protection and

00:01:47.280 --> 00:01:52.799
i had to put the second power supply on there speaking of things we needed to do

00:01:50.720 --> 00:01:57.280
we needed to go with rgb memory of course right well of course yeah so it's

00:01:54.720 --> 00:02:02.880
not ecc but it does support it if we wanted it okay instead it's just six

00:01:59.360 --> 00:02:05.840
matching sticks of ddr4 3000 which is

00:02:02.880 --> 00:02:08.399
faster than the 2666 we have in our imac pro

00:02:06.799 --> 00:02:14.080
we could go faster but we don't have that many matching sets lying around so

00:02:11.120 --> 00:02:20.720
he mentioned that we could go with ecc memory and that's because the cuda grass

00:02:17.360 --> 00:02:23.520
is that underneath this noctua u14s

00:02:20.720 --> 00:02:34.160
cooler here is the biggest baddest xeon w workstation processor on

00:02:27.040 --> 00:02:36.560
the market the 28 core 56 thread 3175x

00:02:34.160 --> 00:02:42.319
which means that we've actually got a solid

00:02:37.599 --> 00:02:45.040
20 more threads on our system than we do

00:02:42.319 --> 00:02:50.400
on our upgraded imac pro and what's more not only do we have more

00:02:47.760 --> 00:02:54.160
ram slots than our imac pro by a factor of two

00:02:51.440 --> 00:02:59.040
we can also add more modern graphics cards like our radeon 7 here

00:02:57.440 --> 00:03:05.280
so just to be clear guys we have crushed

00:03:02.080 --> 00:03:06.959
the imac pro in every possible

00:03:05.280 --> 00:03:12.080
performance metric but that radeon 7 does mean that we

00:03:09.760 --> 00:03:15.120
needed to get the latest version of mac os working

00:03:13.840 --> 00:03:20.720
yeah that was a pain to get running and before i talk about it i just want to

00:03:18.319 --> 00:03:26.560
stress that this is not a tutorial so first i used foxlet's new mac os

00:03:23.519 --> 00:03:29.280
simple kvm as a jumping off point after

00:03:26.560 --> 00:03:32.959
trying a couple of different approaches including running linux logical volume

00:03:31.040 --> 00:03:38.400
arrays with optane acceleration which didn't work i eventually settled on the

00:03:35.040 --> 00:03:40.480
current setup 27 cores 54 threads 42

00:03:38.400 --> 00:03:44.720
gigabytes of ram and two one terabyte samsung 970 pros in apple

00:03:43.040 --> 00:03:48.560
raid mirror because apparently this wasn't complicated enough for me as it

00:03:46.239 --> 00:03:48.560
was

00:03:49.599 --> 00:03:56.080
now unfortunately while i wanted to set up optane the reason that didn't work is

00:03:53.920 --> 00:04:00.640
because the mac os installer straight up refused to boot with opting in it even

00:03:58.080 --> 00:04:04.799
our ssd 800ps that i tried to use so so just with octane anywhere in the system

00:04:02.640 --> 00:04:09.040
the mac os installer is just like no no all right okay then other than

00:04:07.200 --> 00:04:12.400
that it was pretty much a standard install yeah basically

00:04:10.959 --> 00:04:15.840
there's some extra setup and some pitfalls with running apple rate on

00:04:14.000 --> 00:04:20.079
mojave but i worked through them and got the latest

00:04:17.519 --> 00:04:24.560
beta installed and ended up doing direct pass-through of not only the two 970

00:04:22.000 --> 00:04:29.199
pros but also the quantia 10 gigabit nick that's in here and our radeon 7

00:04:27.600 --> 00:04:33.040
and all of it just worked because it's all got drivers in mojave

00:04:31.280 --> 00:04:35.759
all right so i think we've teased you guys enough at

00:04:34.479 --> 00:04:40.320
this point let's do some benchmarks so why don't we

00:04:38.000 --> 00:04:46.320
start with a staple apple benchmark geekbench 4

00:04:42.800 --> 00:04:48.720
okay so our imac pro got a very

00:04:46.320 --> 00:04:51.360
respectable multi-threaded score of over 53 000

00:04:50.479 --> 00:04:58.880
but our hackintosh creamed it by over 20 000

00:04:56.160 --> 00:05:06.000
points to put that in perspective that's almost as much as adding an entire

00:05:02.320 --> 00:05:08.639
core i9 macbook pro to the mix

00:05:06.000 --> 00:05:14.160
and when we look at opencl and metal performance

00:05:11.039 --> 00:05:16.880
okay so that's interesting here our

00:05:14.160 --> 00:05:22.240
radeon 7 actually managed worse performance than the vega 56 in our imac

00:05:20.800 --> 00:05:27.440
pro now given that when we threw the radeon

00:05:24.560 --> 00:05:33.360
7 on the imac pro we got similar numbers it looks like aside from yes losing some

00:05:31.520 --> 00:05:38.800
of our performance thanks to overhead from our hypervisor the radeon 7's

00:05:35.759 --> 00:05:41.039
driver just isn't optimized for whatever

00:05:38.800 --> 00:05:45.440
geekbench is throwing at it here i mean i guess that's why this version

00:05:42.960 --> 00:05:49.520
of mojave is in beta to further investigate then we wanted to fire up

00:05:47.600 --> 00:05:53.919
another gpu test so let's hit it with luxmark now that's not

00:05:52.800 --> 00:05:58.960
bad now we've at least doubled our

00:05:56.000 --> 00:06:04.319
performance over the vega 56 which is quite impressive considering that the

00:06:01.039 --> 00:06:06.639
radeon 7 is a 64cu part

00:06:04.319 --> 00:06:11.280
and considering that our imac pro gets the same level of performance out of a

00:06:08.800 --> 00:06:15.360
radeon 7 when connected via an external enclosure so again this seems to point

00:06:14.319 --> 00:06:19.039
to driver optimization rather than raw

00:06:17.680 --> 00:06:23.759
performance now let's change gears here and go ahead

00:06:21.520 --> 00:06:29.280
and hit our cpu with some good old-fashioned cinebench

00:06:27.440 --> 00:06:35.039
three two one go this is not even going to be a

00:06:31.680 --> 00:06:38.479
fair fight not really no

00:06:35.039 --> 00:06:40.639
i mean there are so many extra threads

00:06:38.479 --> 00:06:45.919
on this thing so to be clear we actually gave up a full

00:06:43.600 --> 00:06:51.759
hyper-threaded core for our host linux operating system but that still leaves

00:06:48.720 --> 00:06:54.560
us with a very significant performance

00:06:51.759 --> 00:07:00.000
advantage compared to the 18 core imac pro not to mention that our base clock

00:06:57.520 --> 00:07:05.360
is actually much higher which for a heavily threaded workload like this one

00:07:02.000 --> 00:07:07.360
makes a huge difference okay we're done

00:07:05.360 --> 00:07:14.479
nearly 11 000 in cinebench r20 and are you even

00:07:10.880 --> 00:07:17.280
done yeah just finished 6 800.

00:07:14.479 --> 00:07:21.599
we're over a third faster than the fastest computer that apple sells with

00:07:20.319 --> 00:07:26.080
mac os with that said though if we look at

00:07:23.840 --> 00:07:32.560
single threaded performance in cinebench the achilles heel of our xeon shows up

00:07:29.599 --> 00:07:36.080
because it's got so many cores intel had to do some

00:07:33.840 --> 00:07:42.479
careful tuning of the power profiles of this thing and our fastest boost clocks

00:07:39.360 --> 00:07:44.479
are much lower than the imac pros which

00:07:42.479 --> 00:07:48.880
means that without overclocking we'll always be at a disadvantage in these

00:07:46.720 --> 00:07:52.720
kinds of single threaded workloads that's a bit of a bummer but uh

00:07:51.440 --> 00:07:56.879
hey overclocking on this platform does

00:07:54.720 --> 00:08:00.639
actually happen to be a thing for now though

00:07:58.560 --> 00:08:03.840
let's move on to some more testing let's fire up blender and give that a go

00:08:02.800 --> 00:08:06.720
all right ready set

00:08:08.720 --> 00:08:14.800
it's so funny so you have to be the one that actually does all the work to build

00:08:12.960 --> 00:08:18.560
this thing but then when we're like drag racing i get to

00:08:16.879 --> 00:08:23.360
drive it well i got to drive the faster one last time so that's true you did

00:08:20.560 --> 00:08:26.479
actually i noticed that it's not actually a competition we're

00:08:24.960 --> 00:08:29.440
just we're just thrilled this thing works at all

00:08:28.240 --> 00:08:34.800
it's interesting there's actually not a whole lot of air coming out back here like there is but it's like not super

00:08:33.440 --> 00:08:38.959
high you say that as though you don't know that's just imac things

00:08:37.360 --> 00:08:42.479
yeah but this cooler is a lot better than the imac

00:08:40.479 --> 00:08:45.839
yeah but like is it a noctua

00:08:44.480 --> 00:08:50.640
okay then that would be a fun video oh

00:08:48.800 --> 00:08:54.560
cutting open the back of an imac and just going like full like

00:08:52.880 --> 00:08:59.680
you know muscle car hood scoop like yep yep

00:08:57.920 --> 00:09:04.000
is that a video do you guys want to see that anyway in the meantime let's get back to

00:09:02.240 --> 00:09:10.320
this um blender was not a good time for our poor

00:09:06.800 --> 00:09:12.959
imac pro we are just over 40 percent

00:09:10.320 --> 00:09:18.720
faster with our hackintosh on both the bmw and classroom test which is actually

00:09:17.120 --> 00:09:23.360
what's weird is this is better than you'd expect looking at the thread count

00:09:20.800 --> 00:09:27.680
alone and again that's our higher base clock in action here

00:09:25.040 --> 00:09:31.920
finally let's round out our testing with a little bit of handbrake so we fired up

00:09:30.000 --> 00:09:36.680
a float plane video here and we're going to go ahead and use our ltt fast preset

00:09:37.519 --> 00:09:42.480
let's face it the small head start i had isn't going to make a difference at the

00:09:40.959 --> 00:09:46.160
end of this nope so here's the thing even though

00:09:44.560 --> 00:09:51.680
there's diminishing returns when it comes to h.264 transcoding and adding

00:09:48.800 --> 00:09:58.080
more course our hackintosh still pulls off a very respectable 40 second lead

00:09:55.360 --> 00:10:05.519
over the imac pro bringing us then to our final test

00:10:02.080 --> 00:10:06.720
final cut pro

00:10:05.519 --> 00:10:14.320
now that is the only actual clear loss for our

00:10:11.200 --> 00:10:17.279
hackintosh today as it turns out final

00:10:14.320 --> 00:10:21.600
cut's background renderer is not highly threaded and is very

00:10:19.200 --> 00:10:27.680
bursty that is to say it results in regular spikes of cpu utilization which

00:10:25.279 --> 00:10:32.880
means then that our imac pro's higher boost clock wins the day by a wide

00:10:30.079 --> 00:10:38.000
margin and it only adds up as more and more of those spikes have to happen on

00:10:35.279 --> 00:10:42.320
our xeon w3170x again something we could potentially

00:10:40.000 --> 00:10:48.160
overcome with some overclocking before we go any further though we need to do

00:10:45.360 --> 00:10:52.959
something to validate our approach here because in theory you're not losing a

00:10:50.959 --> 00:10:56.640
ton of performance using virtualization at most you know

00:10:54.800 --> 00:11:01.519
one to three percent but we don't have any real world

00:10:58.399 --> 00:11:03.920
validation of this so what we need to do

00:11:01.519 --> 00:11:08.480
is shut this down and compare the results that we just got in mac os

00:11:06.480 --> 00:11:14.000
to our bare metal hardware running in windows in as many cross-platform

00:11:11.040 --> 00:11:17.360
benchmarks as we can find to see how big the difference is

00:11:15.519 --> 00:11:21.519
unfortunately when we did this testing it was all over

00:11:20.399 --> 00:11:26.959
the place even with the two extra threads that our

00:11:24.560 --> 00:11:32.959
bare metal machine has available so on windows occasionally we're

00:11:29.920 --> 00:11:35.360
eclipsing our mac os score in cpu tests

00:11:32.959 --> 00:11:39.040
like our handbrake transcode and in cinebench

00:11:36.480 --> 00:11:45.279
but then our windows machine drops off significantly in blender and v-ray which

00:11:42.480 --> 00:11:50.000
you would think would be very similar benchmarks to cinebench

00:11:47.519 --> 00:11:54.240
as for our more gpu-focused tests like luxmark

00:11:51.200 --> 00:11:56.880
well that shows that there is a bit of

00:11:54.240 --> 00:12:00.720
performance left on the table even with direct pass-through although

00:11:58.880 --> 00:12:04.720
this again could be down to driver optimization

00:12:02.079 --> 00:12:08.480
yeah it kind of smells to me like it's a little bit of a scheduler issue in

00:12:06.160 --> 00:12:12.480
windows it's something we've seen before so ironically

00:12:10.399 --> 00:12:15.440
that means our real world performance is actually better

00:12:14.079 --> 00:12:19.519
in the vm now to be clear we're not saying that

00:12:17.600 --> 00:12:24.399
there's magic performance that you get by running your operating system in a

00:12:22.399 --> 00:12:28.959
virtualized environment in fact quite the contrary it's just that

00:12:26.639 --> 00:12:33.360
what we've discovered is that unless we run this bare metal hardware as a

00:12:31.600 --> 00:12:37.680
hackintosh we're not going to be able to quantify the difference

00:12:35.200 --> 00:12:41.360
in mac os now i'm not saying that's impossible it's just that part of the

00:12:39.680 --> 00:12:44.880
reason that we had wanted to do it this way with a vm in the first place was we

00:12:43.760 --> 00:12:49.519
wanted to avoid the fragmentation of hardware and

00:12:48.079 --> 00:12:53.519
the way that apple is constantly updating mac os in ways that can brick

00:12:52.000 --> 00:12:57.040
compatibility now a vm could still have the

00:12:55.040 --> 00:13:01.279
compatibility bricked but at the very least it's just

00:12:58.959 --> 00:13:06.160
one set of things to maintain instead of every motherboard on the planet

00:13:03.519 --> 00:13:09.279
anyway bottom line is this is very experimental right now this is not

00:13:08.079 --> 00:13:15.360
something that we're recommending that you go do for yourselves unless you've got a ton of time to kill i mean

00:13:13.440 --> 00:13:18.800
this motherboard has a market segment composed of a few

00:13:17.440 --> 00:13:24.160
dozen people at best and when anthony said this was a

00:13:21.440 --> 00:13:28.639
complicated setup he wasn't kidding i'm just going to let this list of pitfalls

00:13:25.839 --> 00:13:31.360
and problems scroll by wow

00:13:30.240 --> 00:13:35.519
but there is still even more potential here

00:13:33.440 --> 00:13:42.480
because unlike our imac pro we can overclock we can add more memory and

00:13:39.040 --> 00:13:44.480
uh well if this gpu slot wasn't dead we

00:13:42.480 --> 00:13:48.800
could even add more than one graphics card to our hackintosh

00:13:46.480 --> 00:13:52.720
but since support for the radeon 7 is still as a filming in beta we're not

00:13:51.200 --> 00:13:57.120
going to press our luck any further than this besides i think um

00:13:57.360 --> 00:14:02.880
we've got enough future plans for this that there's

00:14:00.800 --> 00:14:06.480
another video in here freshbooks is the small business

00:14:04.560 --> 00:14:10.639
accounting software custom built for how you want to work it's a simple way to be

00:14:08.320 --> 00:14:15.040
more productive more organized and to get paid faster with freshbooks whether

00:14:13.360 --> 00:14:18.320
you're a freelancer or a small business owner you can create and send

00:14:16.639 --> 00:14:22.000
professional looking invoices in less than 30 seconds you can set up online

00:14:20.399 --> 00:14:26.639
payments with just a couple of clicks to get paid up to four days faster you can

00:14:24.399 --> 00:14:30.959
see when a client has seen your invoice to put an end to the guessing games and

00:14:29.120 --> 00:14:35.199
you get access to freshbooks's unbelievable support like literally you

00:14:33.279 --> 00:14:40.240
call them on the phone and someone picks up no support tree no return calls

00:14:37.519 --> 00:14:43.920
nothing like that just friendly help when you need it for an unrestricted

00:14:42.320 --> 00:14:48.160
30-day free trial just go to freshbooks.com tech tips we're going to

00:14:46.160 --> 00:14:51.040
have that linked below and enter linus tech tips in the how did you hear about

00:14:49.760 --> 00:14:57.040
us section so thanks for watching guys if you disliked this video you can hit that

00:14:54.639 --> 00:15:01.920
button but if you like seeing the fastest mac on the planet

00:14:59.920 --> 00:15:04.480
then hit that like button get subscribed and maybe consider checking out where to

00:15:03.120 --> 00:15:07.680
buy the stuff we featured at the link in the video description

00:15:06.160 --> 00:15:11.279
also down there is our merch store which has cool shirts like this one and that

00:15:09.279 --> 00:15:14.240
one and our community forum which you should totally join