Tiny computer with external GPU! (Lattepanda Alpha 2019 model + RX 550 GPU m.2)


This video is brought to you by Skillshare. Become a member for free for 2 months with
the link down bellow. Last year I showed you my love for this thing,
the Lattepanda Alpha. This tiny raspberry pi like computer hides a full
windows 10 PC with a laptop m3 CPU and 8 GBs ram, making it potentially one of the most
interesting tiny low spec computers out there. You can watch the original video I did on
this on the corner, but long story short I had received one of the new versions of this
computer which introduced what on paper is a slightly better dual-core CPU and to balance
limits to RAM to single-channel which kill integrated graphics performance, although
there was still a fair deal that could be achieved in terms of gaming. I ended that video by commenting on the m.2
ports on the back of the device and that I would add an external GPU and do that video
“soon”. That was 5 months ago. What happened? Lots of things happened. Turns out sometimes adding a GPU over m.2
is a bit a hack and not quite a seamless experience. Here is what I have learned so you can make
better decisions. On the back of the lattepanda there are two
m.2 ports. These are becoming more and more common as
the main port for expansions even on cheaper computers, with stuff like mini PCIe relegated
to older components. While each port will vary on the type of connection
it allows, they often have PCIe lane access meaning that they can be used for expansions
like SSDs and Wifi adapters and sometimes if you are lucky you can even get an external
GPU connected in there. Easy right? WRONG. You see there is no one type of m.2 port,
they come in 12 freaking variants depending on the location of the notch and while in
practice there are roles for each variant, also known as key, and therefore you will
rarely see most of them, not every cable, adapter and device that you see online specifies
which key does the cable use. Case in point I went and bought a bunch of
“m.2 to PCIe adapters” online mostly from Amazon Spain which had a very limited selection
without checking the images to confirm details. This what not a good idea. 90% of all the adapters available on Amazon
are from M-Key, which is the right one on the lattepanda to PCIe x4. Thanks to the magic of the PCIe interface
you can perfectly connect a full x16 GPU to a tiny x4 interface and have it work, just
with limited bandwidth, which is all the bandwidth that m.2 key M port can deliver anyway. But to connect a full GPU to the adapter you
need it to have an open-ended PCIe slot, not one of these slots that has a piece of plastic
blocking it. At the moment of purchase, I could not find
an open-ended adapter so I tried to have a friend help me cut a hole to get the whole
GPU in. This did not work well since the pins
are so close together it was extremely difficult not to damage them, even with some expert
help. So the solution was to bit the bullet and
get an open-ended adapter from china from Ali-express and leave my destiny in the hands
of the Spanish customs gods. Eventually I did manage to get my hands on
a couple of working ones. To make this adapter work without a riser
the lattepanda has to be set backwards which does make checking status light a bit difficult
but it does look really cool with the GPU on top. However, the GPU can not be powered from the
m.2 and requires external power through these pins. If you are industrious and crafty enough you
can modify a powerful enough supply to connect here. If you are not they usually come with a Molex
or SATA adapter to connect to a PC Power Supply, the one problem with that is since the motherboard
connector is not plugged in the power supply will not turn on automatically with the PC
and you will have to short a couple of pins manually which is easy but yeah, a bit of
a hassle to set up. Which is why I literally ended up plugging
it to the power supply of an entirely different PC I had running to test it. This is probably the most wasteful setup I
could have done. I am well aware. Another alternative that was surprisingly
readily available on amazon is the EXP GDC. If you are a long time fan of the channel
this might seem familiar. This a full PCIe base that connects to some
sort of PCIe access through… and HDMI port? Sure. PCIe over HDMI. The gentleman’s thunderbolt. That makes absolute sense. And you can find this extremely weird HDMI
cable implementation going to one of 3 things: mini PCIe which is what most old laptops use and
the one I used in my ancient video about connecting an external GPU to a netbook, to ExpressCard which you can find in another
set of older laptops and M.2 which is the cable I bought for this
video. On one side the EXP is its own contained unit
so you do not have to put the lattepanda in any weird position and it connects and powers
up a PC power supply easily without the need of shorting pins and is in a general way more
comfortable to work with. The M.2 connector cable is a bit too short
for the lattepanda so I can not securely screw it. There are adapters to deal with that but I
am done buying stuff from aliexpress for a while so I am just going to, you know… I am sure this is going to be fine. Now the real problem that you might have zero-ed
in is this m.2 connector is E-Key so I can only connect it to the other slot on the lattepanda
and this one has only two PCIe lanes vs the 4 of the M-Key connector. So even though you are plugging the GPU in
a full 16 lane connector in practice you are getting, at most, 2 and windows shows just one! What does that mean? That the amount of data the PC can get to
the GPU is limited and therefore it might not use a GPU to its full potential. Don´t worry, I am going to experiment more
with that on a second. Now you would believe with any of the two
adapters it just a matter of putting your GPU in, turning the lattepanda on and installing
drivers right? Nope. Wrong again. Oh no, that would have been too easy. Connecting a GPU over m.2 can be a bit of
a finicky process. With both adapters, the whole setup didn’t
exactly just worked with any GPU I tried. The Nvidia GT 1030? it doesn’t even boot. The Nvidia GT1050? It boots, and let me install drivers and then
epically bluescreens when I try anything else. The AMD RX 550? This was an ok sub $100 GPU choice not that
long ago as it stood between the gt 1030 and GT 1050 in terms of power and it is still
a great choice if you can find it for less than a GT 1030. More importantly for our experiment, it works
and is stable over the connector so we will proceed with that. So now, let’s test exactly what are we sacrificing
when using the slower but more comfortable port. I wish I could say something like half the
bandwidth means, 30% less FPS or something but it is not as simple, especially when working
with budget GPUs, and will massively depend on the game. For any game, there are a number of things
that could be the limiting factor in its performance on a specific GPU. IF all other elements of the GPU are doing
fine and the CPU is not causing a bottleneck then the bandwidth is 100% going to be a problem. Here is a great example. CSGO continues to be a relatively light game
in terms of GPU power so the rx550 is going to do fine. To my surprise the addition of a dedicated
GPU really allowed us to use the CPU to its full potential and it turned out to be enough
for testing scenarios on this game and that means that for sure the limited bandwidth
is going to be the problem. So the community benchmark map
over the 1 lane e-key adapter? 69 FPS average. Not bad at all. Over the 4 lane m-key adapter? 125 FPS average. As the people on my discord would say, big
oof But hold on, because matters change dramatically
on any game that forces the GPU at all. Case in point I run the benchmark on the original
Tomb Raider, which tends to be light on CPU usage and can make this GPU sweat a bit more
when on the high settings which still look pretty amazing even know. Since now we are not dealing with potentially
hundreds of frames per second the difference in bandwidth barely makes a difference. So for the framerates I expect from the games
I will test I am going to stick with lower bandwidth adapter, just because it makes the
setup way easier to handle and I am tired of using a weird system to power it up. I am sorry. If there an m-key adapter for this base I
will be sure to find it eventually. Especially considering that the CPU is still
the main limiting factor for pretty much everything I tried except CSGO. Case in point here is rocket league at 1080p
on a higher graphical preset at a 60 FPS cap. The RX 550 is having no trouble while the
CPU sees some heavy use and occasionally can not keep the 60 FPS lock. This is still a terrific result and an absolutely
wonderful experience for this game that I can see myself returning to. However, I could not help but notice that
for a lot of videos out there doing this same experiment with the older version of the Lattepanda
where getting better framerates. This is a bit odd given that the CPU on this
one is supposed to be an upgrade and the result was consistent no matter which m.2 adapter
or game I used. So what’s happening? I am not sure. The world of low TDP mobile CPUs is at times
weirdly inconsistent. It did still overperform my expectations. For example, here is Apex Legends at the 1080p. The m3 CPU was bravely capable of maintaining
over 30 FPS on this game, even during combat which, while unideal for a competitive shooter,
is enough to actual fend for myself as you can see. This is not precisely the lightest of competitive
shooters so this is definitively a great result. If anything, I kinda wish I had set a more
reasonable FPS limiter to improve the overall pacing. Overwatch, on the other hand, tends to be
somewhat lighter than Apex while also being more manageable at lower framerates. Considering how badly the game did when I
tried playing it on the integrated GPU and how I was unsure if the stuttering was caused
by a lack of bandwidth on the integrated graphics or lack of power on the CPU I can now very safely say that the CPU was
not the main problem. The RX 550 was an excellent match for Overwatch
as I could use medium preset graphics at 1080 with still plenty of space left over. The CPU was shugging along keeping the game
anywhere between 30 and 60 FPS but in general, the game felt just playable enough. Definitely a leg over what I expected. So should you get an adapter and start connection
GPUs to mini-computers with m.2 ports? Depends. You need to be aware that this whole thing
is a bit of a hack so you have to be comfortable with random things not working and having
to fiddle around with power supplies but if you are comfortable with that or crazy enough
to even build your own custom power supply for the adapter then owning something like
this base is actually a pretty interested investment. After all, I have seen members of the community
and fans of this channel build some really cool stuff using similar methods and the results
speak for themselves. I am definitely trying this on other devices
soon. Now if the idea of using a mini PC as your
mobile work station which can occasionally use an external GPU to freelance, work and
find your fortune around the world you might be interested in acquiring the skills
to work from anywhere. Like this class on how to write better Javascript,
a tremendously useful programing language, the one programming language that ever actually
got decent at and one that took me surprisingly far. This class is hosted by Christian Heilmann,
a superstar on web development. Funny story, years before I started YouTube
I did a bit of volunteering for Mozilla and I had a chance to briefly meet Christian
personally when he worked there and I can certify that any course made by him
is worth looking at. This is one of many courses you can find on
development, entrepreneurship, freelance, animation, graphic design and many, many other
cool skills at Skillshare, this video’s sponsor. Skillshare is a community for creatives and
lifelong learners with classes that include a combination of video lessons and a class
project. Skillshare has classes to fit your skill level
and most classes are under 60 minutes with short lessons to fit any schedule. If you become a member of its community you
gain unlimited access to thousands of inspiring classes, with hands-on projects and feedback
from a community of millions. An annual subscription is less than $10 a
month which is incredibly affordable, especially when compared to pricey in-person classes
and workshops aaaaand the first 500 of my subscribers to
click the link in the description will get 2 free months of Premium Membership so you
can explore your creativity. This is a limited offer so get it while it
is hot and get started on your digital nomad dreams. Thank you to them for sponsoring this video
and you for watching.

Leave a Reply

Your email address will not be published. Required fields are marked *