WEBVTT

00:00:00.080 --> 00:00:06.160
Intel sponsored or at least they tried to i've never

00:00:04.160 --> 00:00:10.880
done this before but shortly after i arrived at Intel's israel development

00:00:08.080 --> 00:00:15.360
center i ripped up the previously agreed upon sponsorship check because the lab

00:00:12.960 --> 00:00:19.760
here is so far beyond anything i i've ever seen before in my life that i would

00:00:17.440 --> 00:00:24.080
have paid them to take a tour of it what you guys are about to see is the full

00:00:22.000 --> 00:00:29.279
monty not just the facilities and the equipment but the people who created

00:00:26.240 --> 00:00:31.920
many of my favorite cpus from core 2 duo

00:00:29.279 --> 00:00:35.920
back in 2006 all the way to 12th gen alder lake and

00:00:33.680 --> 00:00:39.680
somehow maybe they were all sick pr and marketing didn't get in the way giving

00:00:37.760 --> 00:00:44.480
me nearly unlimited access to the engineers lab techs and project leads

00:00:42.559 --> 00:00:48.160
you wanna know how the sausage gets made wrong video but if you wanna know how

00:00:46.160 --> 00:00:54.320
CPU gets made you've come to the right place oh just like our sponsor crucial don't

00:00:51.840 --> 00:00:59.359
just work faster work better crucial's dvr5 RAM is engineer for efficiency so

00:00:56.879 --> 00:01:04.159
you can load transfer and download files faster with less lag time and more

00:01:01.840 --> 00:01:08.360
efficiency get your crucial ddr5 RAM today using the link below

00:01:14.720 --> 00:01:19.360
creating a CPU takes a really long time

00:01:17.759 --> 00:01:23.520
from this wish list scribbled on a whiteboard by engineering to final

00:01:21.119 --> 00:01:29.680
delivery 12th gen core was about a 30-month process the first step is

00:01:26.720 --> 00:01:34.400
design that takes a year or more of hundreds of engineers working tirelessly

00:01:32.000 --> 00:01:38.000
to build a simulation of a working processor once they're

00:01:36.560 --> 00:01:42.560
pretty sure it's going to work the blueprints get handed over to the fab to

00:01:39.920 --> 00:01:45.840
produce a batch of es1 or alpha silicon that's going to take a few months due to

00:01:44.159 --> 00:01:48.880
the incredible complexity of the fabrication process but that doesn't

00:01:47.600 --> 00:01:53.280
mean they're going to be waiting around the simulated version of the chip can be

00:01:50.880 --> 00:01:59.040
used in the meantime to develop firmware software and testing tools and launching

00:01:56.720 --> 00:02:03.759
any new processor design let alone the four dies and six different

00:02:00.960 --> 00:02:08.640
configurations from nine to 240 watts of alder lake requires close collaboration

00:02:06.159 --> 00:02:13.360
with partners so at this stage Intel shares high level timelines and design

00:02:10.800 --> 00:02:17.440
considerations like expected thermals and power consumption of course there's

00:02:15.680 --> 00:02:22.000
no guarantee that what they get back from the fab is going to work in fact

00:02:19.920 --> 00:02:25.360
it's pretty much a given that it won't man

00:02:22.959 --> 00:02:30.480
i would give just about anything to be a fly on the wall in this room during a

00:02:27.760 --> 00:02:35.120
product launch this is the power on room and when the first es1 chips roll off

00:02:32.560 --> 00:02:40.640
the line Intel flies experts in every field memory BIOS networking you name it

00:02:38.640 --> 00:02:45.200
to here in israel where they all get together in this room with the goal of

00:02:43.360 --> 00:02:49.040
powering on the chip and booting an operating system sounds simple enough

00:02:47.440 --> 00:02:54.000
right but no even the best simulations can't possibly

00:02:52.080 --> 00:02:57.599
hope to account for all the problems that can occur in the real world and i

00:02:55.680 --> 00:03:02.319
give you my personal Linus tech tips guarantee that if i brought a black

00:02:59.519 --> 00:03:06.080
light in here i'd find sweat blood

00:03:03.280 --> 00:03:09.599
and tears on pretty much every working surface

00:03:07.200 --> 00:03:14.560
i worked more than 24 hours in a day to do to get this to make it happen it's

00:03:12.000 --> 00:03:18.400
really amazing i was even when my wife gave a birth with my little child it's

00:03:17.440 --> 00:03:22.480
even more than that you would say that

00:03:20.959 --> 00:03:27.120
once the most game-breaking bugs are ironed out usually through software and

00:03:24.400 --> 00:03:31.680
firmware workarounds that es1 silicon makes its way out to the validation lab

00:03:29.440 --> 00:03:36.959
where the goal is to achieve a much higher level of functionality and this

00:03:33.920 --> 00:03:39.200
place is absolutely

00:03:36.959 --> 00:03:43.120
wild so make sure that you're subscribed for the deeper dive that's coming there

00:03:40.799 --> 00:03:47.200
is so much crazy gear in here that i actually extended my trip to make an

00:03:45.280 --> 00:03:51.360
additional video about it which is another first but let's get you through

00:03:49.519 --> 00:03:56.239
the basics for now it's divided into multiple zones each of them focused on

00:03:53.519 --> 00:04:01.920
validating either a specific subsystem or some combination ciivt is right

00:03:59.680 --> 00:04:07.120
outside the power on room and this acts as a small scale testing ground for new

00:04:04.560 --> 00:04:11.599
fixes which could easily be rolled out multiple times a day while the lab is

00:04:09.519 --> 00:04:17.280
running full tilt as you might expect the core ip section where they validate

00:04:13.760 --> 00:04:20.160
the CPU cores themselves is the biggest

00:04:17.280 --> 00:04:25.040
and by an even bigger margin than you probably already realize this lab

00:04:22.720 --> 00:04:30.000
everything you just saw actually only contains the team that worked on alder

00:04:26.800 --> 00:04:32.720
lake's p cores or high performance cores

00:04:30.000 --> 00:04:36.880
the e-core or efficiency core validation lab is somewhere else entirely but if

00:04:35.199 --> 00:04:41.520
you keep wandering around in here you'll find basically everything else

00:04:40.400 --> 00:04:46.080
the mislabeled memory debug team faced some

00:04:44.320 --> 00:04:49.759
really unique challenges with alderley thanks to the four separate memory

00:04:48.240 --> 00:04:55.680
technologies that it has to support during the transition from ddr4 to ddr5

00:04:53.199 --> 00:04:59.680
external devices like Thunderbolt docks and usb drives need to be tested and

00:05:00.240 --> 00:05:05.919
this aisle is fondly nicknamed time square because it's full of all the

00:05:04.479 --> 00:05:10.800
different kinds of screens that you might want to connect to your computer

00:05:07.680 --> 00:05:14.400
high resolution high refresh rate HDMI

00:05:10.800 --> 00:05:17.759
2.1 you name it they've got it and

00:05:14.400 --> 00:05:19.440
sometimes all at once

00:05:17.759 --> 00:05:24.160
this alder lake mobile test bench was running 4k video on two high resolution

00:05:22.479 --> 00:05:27.759
external monitors plus the embedded displayport display to represent the

00:05:26.000 --> 00:05:30.960
built-in one on your notebook it's looking pretty good ladies and gentlemen

00:05:29.600 --> 00:05:34.639
the only thing missing around here is some quality desk pads from

00:05:32.680 --> 00:05:39.360
ltteststore.com what is this mice directly on the table

00:05:37.120 --> 00:05:43.759
all of these teams work in parallel as much as possible testing retesting

00:05:42.160 --> 00:05:49.600
making sure that someone else is fixed didn't break your own biz until finally

00:05:46.320 --> 00:05:51.280
we end up in sst stress and stability

00:05:49.600 --> 00:05:57.199
testing is where the smaller scale validation now gets run over longer

00:05:53.840 --> 00:06:00.080
durations and at a much greater scale

00:05:57.199 --> 00:06:03.680
most of the es1 bugs will be squashed or at the very least bound and then they go

00:06:02.000 --> 00:06:09.759
through the whole process again with more refined beta es2 silicon then rinse

00:06:07.360 --> 00:06:14.639
and repeat until finally the design is ready for prime time which doesn't mean

00:06:12.240 --> 00:06:20.080
that testing is over every chip that rolls off the line gets validated but at

00:06:17.360 --> 00:06:24.400
Intel scale quality control can't possibly be done by hand like this so

00:06:22.479 --> 00:06:29.919
let's go downstairs and see some really cool robots shall we

00:06:26.479 --> 00:06:32.080
the device under test in this case a CPU

00:06:29.919 --> 00:06:36.880
gets installed in what's called a tester interface unit which is kind of like a

00:06:34.720 --> 00:06:40.960
giant pcb that plugs into the back of the green machine and is covered in

00:06:39.199 --> 00:06:45.680
probe points that can be monitored using the oscilloscope in the back this tells

00:06:43.199 --> 00:06:50.479
the engineers exactly what's happening on even a single pin during the test

00:06:48.319 --> 00:06:55.600
then when it's ready the thermal head here which we met earlier presses down

00:06:53.680 --> 00:06:59.440
and the tests begin what you're looking at right now is basically the same test

00:06:57.680 --> 00:07:03.360
rig but designed for an engineer to stand in front of it and work on it

00:07:00.880 --> 00:07:08.000
manually and with rigs like these Intel can throw vectors at anything from a

00:07:05.680 --> 00:07:13.360
single transistor to a block of transistors to an entire subsystem of

00:07:10.800 --> 00:07:17.919
the whole chip or any combination of the above and they can do it at temperatures

00:07:15.440 --> 00:07:22.080
ranging from sub-zero all the way up to over a hundred degrees with the help of

00:07:20.080 --> 00:07:28.319
the thermal head synthetic tests are fine and good but ppv is the end game

00:07:26.240 --> 00:07:32.960
product platform validation takes place on pretty much every CPU off the line at

00:07:30.639 --> 00:07:37.440
this point and involves taking that theoretically working processor and

00:07:35.120 --> 00:07:42.400
installing it in an actual motherboard connected to actual storage memory and

00:07:40.479 --> 00:07:47.840
peripherals and checking to make sure that everything is working as expected

00:07:44.800 --> 00:07:49.680
these rigs can do all kinds of cool

00:07:47.840 --> 00:07:54.400
things that a customer would never encounter in the real world like they

00:07:51.840 --> 00:07:59.120
can even have the CPU go to sleep but still output to a display

00:07:56.879 --> 00:08:02.879
i'm sure there's a reason that they test that but

00:08:00.319 --> 00:08:07.759
it's not obvious to me this one's a little more obvious i

00:08:05.680 --> 00:08:12.160
love this room i mean it's all fine and good to have one processor that works

00:08:10.160 --> 00:08:16.160
with that one kit of memory that you validated it with but if you want to

00:08:13.840 --> 00:08:20.960
deliver a consistent experience you need way more data and robots like this one

00:08:19.120 --> 00:08:27.120
can help you get it whether you want to test a huge number of cpus with every

00:08:24.160 --> 00:08:31.360
possible configuration of memory or run countless chips through a test where

00:08:28.879 --> 00:08:36.719
they're using an external graphics card this is the place to do it and okay

00:08:34.080 --> 00:08:41.120
actually it's not countless Intel did disclose that over 10

00:08:38.880 --> 00:08:45.120
000 systems get powered on internally before they launch in order to hit their

00:08:42.880 --> 00:08:47.680
quality target and they never rest remember that timeline we talked about

00:08:46.720 --> 00:08:53.120
before if alder lake was started 30 months ago

00:08:50.640 --> 00:08:57.440
do the math that means meteor lake is almost certainly well underway and if we

00:08:55.519 --> 00:09:01.600
were to rifle around through the stacks and stacks of processors in this place

00:08:59.519 --> 00:09:05.360
we would almost certainly find raptor lake chips

00:09:03.040 --> 00:09:09.360
and more i mean they do everything from compute to communications to ai and

00:09:07.600 --> 00:09:13.040
security here and if alder lake has anything to go on all that stuff that's

00:09:11.279 --> 00:09:17.760
in development is going to be hecka exciting just like i'm excited to tell

00:09:14.720 --> 00:09:19.680
you about our sponsor nzxt with nzxt

00:09:17.760 --> 00:09:24.320
build getting a custom built pc is easier than ever it's available in the

00:09:21.600 --> 00:09:28.399
us australia germany netherlands france and italy and allows you to find a pc

00:09:26.320 --> 00:09:33.600
that fits your needs perfectly whether it be a custom a pre-built or a build

00:09:30.959 --> 00:09:37.440
kit just set your budget see how your pc will perform in your favorite games and

00:09:35.279 --> 00:09:41.760
build takes care of the rest you can even customize and upgrade your build

00:09:39.200 --> 00:09:46.720
with various nzxt case options and RGB lighting setups nzxt features

00:09:43.920 --> 00:09:50.640
transparent pricing with a flat 99 assembly fee in your local currency and

00:09:48.480 --> 00:09:54.720
they'll ship within 48 business hours with all your pc's components covered by

00:09:52.320 --> 00:09:58.000
one warranty plan nzxt will manage any problems that you have through their

00:09:55.839 --> 00:10:01.680
expert live chat for real-time help and troubleshooting there's even free and

00:09:59.839 --> 00:10:06.000
easy returns for any system that does not meet the build engine FPS

00:10:03.839 --> 00:10:10.399
performance guarantee within 10 so check out nzxt build today using the

00:10:08.320 --> 00:10:13.920
link in the video description if you guys enjoyed this video maybe go check

00:10:12.240 --> 00:10:18.800
out the fab tour because i think that's the only thing that we've ever done that

00:10:16.640 --> 00:10:22.079
can hold a candle to this tour of the research and development center