How many Chrome tabs can you open with 2TB RAM?
Summary
TLDRThis video takes a deep dive into building a high-capacity server system with 2 terabytes of RAM using specialized memory sticks from SK Hynix. It explains how load-reduced DIMMs and advanced memory architecture allow for such massive capacity, though it comes with performance trade-offs. The host tests the system's capabilities by opening thousands of Chrome tabs, eventually hitting practical software limits. Despite using only a fraction of the RAM, the test demonstrates the extreme performance potential of the system. SK Hynix's SSD is also featured as the boot drive. It's an engaging technical exploration of cutting-edge server hardware.
Takeaways
- 🖥️ The video explores the process of configuring a system with 2 terabytes of RAM, which typically requires exotic multi-CPU designs and specialized memory modules.
- 💾 Each RAM stick used in the system has a capacity of 128GB, allowing for a high total memory count.
- ⚙️ The system utilizes SK Hynix's load-reduced DIMMs (LRDIMMs) to overcome rank limitations, enabling more memory without performance degradation.
- 🚀 The memory in use operates at 2933 MHz, the rated speed of the AMD EPYC 64-core processor used in the build.
- 🧠 LRDIMMs reduce memory controller load by converting quad-rank memory to dual-rank, allowing more memory with a small latency penalty.
- 🖥️ The maximum limit for memory in Windows 10 Pro is exactly 2 terabytes, which matches the configuration in this test.
- 🌐 Chrome uses a large amount of RAM for tabs, with 5,000 tabs using about 132GB, but CPU usage becomes the limiting factor rather than memory.
- 🔄 The system starts to slow down significantly with over 6,000 Chrome tabs, reaching the practical limit of software responsiveness.
- 💻 Despite having resources available, the system becomes unusable beyond a certain point due to architectural limitations in Chrome and Windows.
- 💿 SK Hynix's Gold S31 SSD was used as the boot drive, which is part of their high-performance SATA3 SSD product line.
Q & A
What is the primary challenge in fitting two terabytes of RAM onto a single system?
-The challenge is finding a way to use exotic designs like quad or eight CPU sockets, as even server-grade motherboards typically top out around 16 memory slots. Specialized memory modules like those used in the video are required to reach such capacities.
What special type of memory is used in this system to achieve two terabytes of RAM?
-The system uses 128GB sticks of Load Reduced DIMMs (LR-DIMMs), which allow for higher capacity per module while maintaining stability and performance by reducing the load on the memory controller.
How does Load Reduced DIMM (LR-DIMM) technology work to handle large amounts of memory?
-LR-DIMM technology uses a buffer chip to allow the CPU's memory controller to operate in serial mode. This reduces the load on the controller, enabling it to handle higher-capacity memory configurations without significant performance degradation.
What are the two methods mentioned for increasing memory capacity per module?
-You can increase memory capacity by either adding more chips (placing them on both sides of the module or double-stacking them) or engineering the chips with more data areas or ranks (e.g., dual rank or quad rank modules).
What is the trade-off of using high-capacity memory like quad or eight rank modules?
-Using higher capacity modules can lead to performance penalties, as the memory controller has limits on the number of ranks it can handle directly. As more ranks are added, memory speed may be reduced, and fewer slots can be populated without performance degradation.
Why does the system use SK Hynix's Gold S31 SSD, and what are its features?
-The SK Hynix Gold S31 SSD is used as the boot drive for testing. It is a SATA3 SSD designed by SK Hynix for high performance and quality, with features like a 5-year warranty and availability in 250GB, 500GB, and 1TB capacities.
What is the significance of memory training when booting up a system with large amounts of RAM?
-Memory training is a process that occurs when booting a system with new hardware configurations, where the system tests and optimizes memory timing. The more RAM installed, the longer this process can take, which is why it can take several minutes for the system to start with large memory configurations.
What is the maximum memory supported by Windows 10 Pro, and how does the system perform with that amount?
-Windows 10 Pro supports up to 2 terabytes of RAM. In the test, the system had 2 terabytes of memory installed but only showed 0% memory usage when basic applications like Chrome were opened, as they did not demand much memory.
How did the system perform when loading thousands of browser tabs in Chrome?
-The system was able to open 6000 Chrome tabs using only around 10% of its available RAM. However, performance started to degrade due to architectural and software limits, not memory constraints, as the system became sluggish despite having more resources available.
What were some of the performance bottlenecks observed during the test with 6000 Chrome tabs?
-The main bottleneck was not memory or CPU capacity but the architectural limits of the software, particularly Chrome and Windows. Despite having plenty of RAM and CPU power left, the system became nearly unusable due to the strain on the browser and operating system.
Outlines
💻 Massive 2TB RAM System Overview
In this segment, the speaker introduces the concept of fitting two terabytes of RAM into a single system. They emphasize the unusual nature of this configuration, which typically requires high-end server-grade motherboards with many memory slots. The speaker highlights the use of SK Hynix's 128GB RAM sticks to achieve this capacity, and briefly mentions the use of an SK Hynix Gold S31 SSD as the boot drive for testing. The explanation focuses on the technical aspects of memory architecture, including server DIMMs and strategies to increase memory capacity through dual and quad-rank modules.
⏳ Memory Training and Initial Boot Delays
The second part discusses the booting process of a server motherboard with a significant amount of RAM, mentioning that the first boot can take a long time due to memory training. The speaker humorously references how it could take up to ten minutes for the system to turn on, followed by a wardrobe change joke. They reveal that the system eventually detects 2TB of memory and 64 cores running at 2933 MT/s. The speaker touches on Windows 10's memory handling limits, noting that the system runs at zero percent memory usage, even when opening Chrome with multiple tabs.
🌐 Testing Chrome's Limits with Massive RAM
This section shifts to testing Chrome's performance under heavy load, focusing on how many tabs the browser can handle with such an extensive RAM setup. The speaker experiments with opening hundreds of tabs and tracks how RAM usage steadily increases, but notes that Chrome's memory footprint remains minimal in the beginning. Despite having plenty of resources, the system's performance gradually deteriorates as more tabs are opened. Eventually, the test reaches a point where Chrome consumes significant CPU cycles and RAM, leading to slowdowns and unresponsiveness even though they haven’t fully maxed out the system’s memory capacity.
Mindmap
Keywords
💡RAM
💡LRDIMM (Load-Reduced DIMM)
💡Memory ranks
💡ECC (Error-Correcting Code)
💡AMD EPYC
💡SK Hynix
💡Memory training
💡Chrome tabs
💡Memory density
💡Windows 10 memory limit
Highlights
Introduction of a system with two terabytes of RAM and the complexity behind its setup.
Special memory modules used: 128GB sticks from SK Hynix to achieve two terabytes of RAM.
Explanation of traditional registered server DIMMs and how they operate in parallel with the CPU's memory controller.
Introduction to quad-rank and dual-rank memory modules and the challenges they pose for memory controllers.
The use of Load Reduced DIMMs (LR DIMMs) to reduce the load on memory controllers, enabling the use of higher capacity modules.
LR DIMMs running at 2933 MHz with ECC (Error Correction Code) and memory buffer chips to support serial operation.
Limitations in motherboard memory slots: even with 16 slots, only certain configurations of dual or quad-rank DIMMs can be used without reducing speed.
Demonstration of booting the system with 2 terabytes of memory and setting up Windows 10 Pro, which has a maximum memory limit of exactly two terabytes.
Experiment on how many Chrome tabs can be opened, starting with 20 tabs and progressing to over 6000 tabs.
At 1000 tabs, Chrome used 132GB of RAM, with noticeable slowdowns but no impact on the available memory.
At 5000 Chrome tabs, CPU usage hit 69%, but only 10% of RAM was used, highlighting system inefficiencies.
At 6000 tabs, the system became nearly unresponsive, indicating limitations of the software rather than the hardware.
Memory usage hit close to 200GB when rearranging the tabs, but performance continued to degrade.
The practical limit of Chrome and Windows architecture was reached, not the physical memory of the system.
Conclusion: Even with 2TB of RAM, real-world usage is limited by software like Chrome and Windows, not hardware constraints.
Transcripts
two
terabytes of ram
how on earth do you pack two terabytes
of ram onto a single system i mean it
sounds crazy to say it out loud
and it is a little crazy i mean unless
you're looking at exotic quad or eight
cpu socket designs even a server grade
motherboard like this one from gigabyte
typically tops out in the neighborhood
of 16 memory slots so step one then
is we are going to need some very
special memory a single stick of this
ram
has more capacity than your entire
system one of these
is 128gb
of memory and naturally to hit two
terabytes as advertised in the video
title we are going to need
a lot of them
sk hynix is also promoting their new
gold s31 ssd every component of the gold
s31 sata3 ssd was produced built and
designed by sk hynix to meet their
quality and performance standards we're
going to be using this as our boot drive
during our testing and i'm going to have
a link to where you guys can learn more
in the video description
[Music]
i've got to ground myself here i'm
pretty sure sk hynix isn't sending me
more of these if i zap them so get ready
for a technical deep dive guys
traditional registered server dimms
operate in parallel such that
effectively each of the individual
memory chips on the module is wired
directly to the memory controller that's
built into the cpu so that's where all
the traces that you see running between
your memory slots and your cpu socket
come into play
now there are two ways that you can
increase the capacity of a memory module
you can add more chips by putting them
on both sides or by double stacking them
like we see here
or you can actually engineer the chips
themselves with more data areas or
ranks
a dual rank memory module might have
exactly the same number of visible chips
soldered onto it but it's effectively
like smushing two single ranked modules
into a single slot and a quadrant module
is like stuffing four modules into a
single slot
which is pretty cool if you need more
capacity but
it causes a small problem
the memory controller on a given cpu has
a limited number of ranks that it can be
wired into directly before performance
starts to drop off or it reaches a hard
limit so you might run into cases in the
server world where even though a
motherboard has you know let's say
16 slots you could populate all of them
with dual rank modules or only half of
them with quad rank modules and then
running at a reduced speed there's not
much point in doubling your memory
density if you cut away half of your
expansion slots and some speed right
like that's classic one step forward two
steps back
so we're getting around that problem
today using what's called load reduced
or lr dimms and there is a lot of
engineering packed into these they run
at a blistering 29 33 megahertz so that
is the rated speed of our amd epic 64
core processor and they have both
a ninth chip in each one of these rows
for ecc or error correction
and an extra memory buffer chip that
allows the processor's memory controller
to operate in serial mode this serial
operation causes a quad rank dim to load
the memory controller like a dual rank
dim or an 8 rank dim to operate like a
quad
now it comes with a performance penalty
but if you find any matching places
i'm not talking to you siri
what what even is this what are you
talking about
now it comes with a performance penalty
but if your workload requires a ton of
memory taking a small latency hit is a
lot better than not having enough ram at
all
set a good example
i love these sockets where you just
screw the cooler right into the
socket just need a boot ssd so all the
memory volatile or otherwise in our
system is actually made by sk hynix in
korea they're one of only a handful of
actual memory manufacturers
in the world and they've been making
ssds for years but it's only recently
that they branched out of just system
builders and enterprise users into
retail
so let's get windows 10 loaded up on
this puppy
moment of truth time server boards lack
a lot of the creature comforts of
consumer boards so we're just going to
short a couple pins turn it on
like i'm standing here eagerly waiting
but i'm not actually expecting it to do
anything anytime soon
every time you boot up a motherboard
with a new hardware configuration it
needs to go through a process called
memory training and the more ram you
have the longer it takes so this could
take like
10 minutes to turn on the first time
it may have taken long enough for a
wardrobe change lttstore.com
but we're up
2 million 97
152 megabytes of memory running at 29 33
million transfers per second
64 cores of processor goodness and i
actually got a cool tip from gigabyte
that apparently if you pop in here and
ignore all of this warning
stuff right here you can actually
overclock your memory on this platform
so it turns out the maximum limit for
installed memory for windows 10 pro
happens to be
boom you guessed it exactly
two terabytes so this is a bog standard
windows 10 install with
zero percent
memory usage
i could probably even open up
chrome and it would still be zero
percent before we proceed with our
entire test though i do plan to find out
just how many tabs we can handle i want
to have a look at what performance looks
like with our configuration it's worth
noting that our 64 core processor
actually doesn't have smt enabled so
it's only running 64 threads it's a bit
of an idiosyncrasy of this board right
now and i haven't spent any time
figuring out boom
52.44 seconds so we're not looking at
some kind of you know crippling
performance difference or anything like
that so we got to figure out a good way
to do this so brandon hit me with some
websites youtube
all right b h photo video i just got
some neopets in there you know linkedin
what are you talking about brandon what
are you doing on linkedin
oh she's sure yeah hey there he is
nope this is a terrible linkedin profile
all right it seems like he's
seems like he's sticking around
new grounds oh i love it and we'll head
over to the mkbhd store this is my
favorite uh 10 million subscriber
commemorative merch right here chrome is
actually using two and a half gigs of
ram right now doing
functionally nothing
so
it's a hog still it just doesn't make a
dent
it doesn't even register it's flatlined
so let's do it let's open all 20 tabs in
a new window yes i'm sure all right we
jumped from 5 to 5.8
now i wanna find out if cycling through
them actually increases it so we're
gonna control tab over through all these
tabs and see if it jumps
it's gonna slow us down quite a bit in
terms of how many tabs we can open per
minute here if we have to do that
and it looks like we don't
hey that's great news
so now this just becomes an exercise in
how fast can i click this is super dumb
i want a piece of paper
you know i'm going to start tallying how
many tabs i'm opening i need paper
update for you guys
at 200 chrome tabs we are now at 10 gigs
of ram and 1 usage 600 tabs 700 tabs 900
tabs 1 000 tabs
5
this is interesting neopets seems to be
our heaviest website first you take my
youth then you take my cpu cycles we are
up to 69 cpu usage well i think we're
gonna run at a cpu before we run out of
memory here guys we are over one stick
of memory use now
132 gigs chrome at least admits that
it's using over 100 gigs of ram now
power usage very high 5 000 tabs here we
go boys rent hey we're at 10 ram usage
now
so that's sick
but what's interesting is that in spite
of us having resources available to us
the system is getting slower and slower
and slower to the point where even when
this ad managed to load in i couldn't
help but notice it was running at a
super crazy low resolution look at this
here here's a perfect example it just
froze while it was playing back that's
what it looks like opening a hundred now
when we already have 5000 open in the
background was like opening up the first
500 when we did it now this is
interesting i was going to rearrange the
tabs to shoot the thumbnail and we got a
spike in cpu usage and i'm not sure if
memory actually changed but
check this out a bunch of these tabs
that were already open before are
reloading now so it seems like
interacting with the window
actually gets it to refresh but that
doesn't change our ram usage and since
that's what we're after i guess that's
still okay uh-oh
oh no
we might have hit our limit
brandon
brandon i'm rolling
we might have hit our limit brendan
but i want to get the thumbnail
if i have to open them all again to get
the damn thumbnail it's gonna suck
six
thousand
we're at the point now we're doing
almost anything in chrome on the system
i'm trying to combine a tab i
accidentally dragged out of a browser
window back into it and it just
it's not responsive enough to do it but
what oh hey
it did it
okay
so we got that
i mean gimp's managing to open up so
that's cool
we are just shy of 200 gigs of memory
usage now
and that's just from rearranging all the
tabs we were at closer to like 185
weren't we this is so weird check this
out this up here says amazon
but mkbhd is also selected and this is
an apple website
so strictly speaking we may not be at
the limit
but from a practical standpoint
i think we've reached the limit
oh no i think i just did the wavy thing
where everything minimizes
the system is completely unusable at
this point
look at them all go a right click
three four five six seven 25 20 there it
is
just shy of 30 seconds just to right
click
so
we only managed to use about a tenth of
our ram
half of our cpu
but we seem to have reached
architectural limits of the software
itself whether it's chrome or windows
and 6000 tabs is well beyond what is
actually usable in google chrome i hope
you guys enjoyed coming along with us on
the ride massive shout out to sk hynix
for sponsoring this video sending over
two terabytes of memory even though we
didn't manage to use it all
this time and of course providing their
gold s31 ssd for us to boot off of it's
available in 250 gig 500 gig and one
terabyte capacities it comes with a five
year warranty and if you're looking for
a high performance say ssd you guys can
check it out it is at the link in the
video description so thanks for watching
guys hope you had as much fun as we did
with this insanity
and we will see you in the next video if
you enjoy more server type content like
this we actually built a crazy nas a
little while ago you can check out that
video down below
Посмотреть больше похожих видео
5.0 / 5 (0 votes)