AMD LET ME BUILD THEIR NEW 80 VEGA SUPERCOMPUTER…
Linus Tech Tips
·Linus Tech Tips
·2018-05-06
·
1,180 words · ~5 min read
0:00
We are here at the Enterprise Data
0:03
Center/ Supercomputer Lab at AMD's Sunnyvale headquarters where they have
0:09
got the definition of an all hands on
0:12
deck project going on here. They've flown in their techs and partners from
0:17
around the world. Even Mark Hurst,
0:20
corporate vice president of Radeon Technologies Group, is getting in on the
0:25
action, planning the power runs for one of the most exceptional supercomput
0:30
systems ever created. When it's full of
0:35
20 servers, this single rack is going to
0:38
be capable of one path flop of full
0:43
precision performance. And AMD
0:47
estimates, it's an estimate, they're not done building it yet, that they'll be
0:51
doing this at around half the cost of a
0:54
competing solution, not to mention more than double the flops per
1:01
watt. And how are they doing such a
1:04
thing? Well, as crazy as it might
1:09
sound, AMD is actually letting me assemble one of the systems. So, uh,
1:16
well, I'm going to show you guys exactly
1:27
how Synergy allows you to share your
1:31
mouse and keyboard between multiple computers at once. Check it out now at
1:36
the link in the video description.
1:39
So, pretty much everything you guys see in this room arrived about 12 hours ago
1:44
at 3:46 in the morning from AMD partners
1:48
like Invente, Samsung, and Melanox and
1:52
has been undergoing assembly and testing non-stop since then. So, it can be
1:57
unveiled at Sigraph in less than a week.
2:01
So, this video is as much a behindthescenes look at the hard work
2:06
that goes into the glitzy polished show
2:10
demonstrations as it is about this
2:14
system. Who am I kidding? We're here for the beastly hardware. So, this in front
2:19
of me is what AMD and Invente are calling the
2:24
P47. And it's this that allows them to
2:27
cram up to 640 CPU cores. That's,280 threads if you
2:34
prefer up to 40 terabytes of memory and
2:39
up to 80 Radeon Instinct Vega GPUs into
2:44
a single rack with some room for
2:48
high-speed networking like Infiniban. At
2:52
a total of about 35,000 watts of power consumption, this
2:58
is about as power dense as you would want to go with air cooling using a rear
3:04
door cooling system like what we saw at the SFU Supercomput recently. So, let's
3:11
take a closer look at it. At the front, you've got some storage bays like you'd
3:16
expect to find on a server. So, I don't think AMD or Invecting anybody to use
3:21
these for more than just like a basic scratch disc. And then on the back is
3:27
the reason for it because that's where the 100 gigabit Infiniband connections
3:33
go that handle job sharing between
3:36
servers as well as access to high-speed
3:40
dedicated storage machines that are elsewhere on the network.
3:44
Moving inside, at the heart of every P47
3:48
is the brand new AMD Epic server
3:52
processor. This one is 32 cores on a
3:56
single CPU with a 180 watt TDP. 32 cores
4:02
at 180 watts. But that's actually not
4:06
the whole story behind what gives P47
4:09
its stunning efficiency. Each epic can
4:13
handle up to two terabytes of DDR4
4:16
memory running in an 8 channel configuration at speeds of up to 2666
4:21
MHz. So that's great for virtualization
4:24
uh rendering and machine intelligence. Our machines have 512 gigs of RAM each.
4:31
But, and this is the key, each epic can
4:36
also handle 128 PCI Express
4:44
lanes. 128 lanes. So, that's why a
4:48
single Epic can be hooked up to up to six or even seven GPUs with a full PCIe
4:58
16X Gen 3 interface. dedicated to
5:02
everyone without a need for a second processor or a PLX chip. And that's with
5:08
plenty of bandwidth left over for high-speed NVMe storage or high-speed
5:14
networking. And all that's critical
5:17
because the real star of the show today is the Radeon Instinct
5:22
Mi25 based on AMD's Vega architecture.
5:26
These puppies right here are actually responsible for the bulk of that pedlop
5:32
of processing power. Every MI25 is a
5:37
full fat Vega consuming 300 watts.
5:40
Actually, similar to the Frontier Edition that already exists and the RX
5:45
Vega that's coming soon. But unlike
5:48
those, it lacks video outputs entirely
5:53
and a fan since it was designed to be
5:56
installed in specialized machines like this one where the cooling is
6:01
integrated. And this is sick. The entire
6:05
16 gigs of HBM2 memory can be accessed
6:10
by another device on the PCI Express
6:13
bus. This allows high-speed communication between discrete Radeon
6:18
Instinct MI25s or even other PCI Express
6:23
devices that are connected to the system like even an NVMe storage device. And
6:28
this is without a proprietary connection
6:31
like ND link and all this high-speed
6:35
communication enabled by Epic, which you can kind of think of as both a CPU and
6:41
the world's biggest PCI Express switch
6:44
is at the core of a new opensource
6:48
architecture that AMD is championing
6:51
called Rockum. In a nutshell, the idea
6:54
is that every part of an all AMD server
6:59
system is linked at extreme speeds in a
7:03
nonproprietary manner, so the user can
7:06
customize and scale up their supercomput
7:09
as they see fit. Want more performance? Just add
7:14
more racks. And while
7:17
$50,000 might seem like a lot for a
7:21
P47, when you consider that AMD's
7:24
initial estimates put them comfortably in a position to claim better
7:28
performance per dollar for modern workloads like financial modeling,
7:33
climate science, and AI, not to mention
7:36
better performance per watt. You can see where their confidence about uh really
7:43
shaking up the top 100 supercomputer
7:46
list comes from and I wish them the absolute best
7:52
of luck. Do you have two computers for some
7:56
reason? Maybe one of them's PC and one of them's Mac or Linux or whatever.
8:01
Well, Synergy lets you solve the problem of having separate keyboards and mice
8:06
for them once and for all. You can share one mouse and one keyboard between two
8:12
or even more computers, so you'll never get confused again. They've got basic
8:17
and pro options for synergy with a one-time payment. And features include
8:22
things like clipboard sharing between the computers, dragging and dropping
8:26
files between the computers, the ability to set up hotkeys, and more. Use our
8:31
link in the video description and get 50% off Synergy
8:36
today. So, thanks for checking out our video here at AMD Supercomput Lab. And
8:43
uh if you guys dislike this video, you can hit that button. But if you liked
8:46
it, hit the like button, get subscribed, maybe consider checking out where to buy
8:50
the stuff we featured at the link in the video
8:53
description. Also down there, we've got a link to our t-shirt store, which has
8:56
cool shirts like this one, as well as our community forum, where you can go
9:00
and talk about all the cool stuff you saw today.