External GPUs

Some motherboard have a Thunderbolt port.
But at that point, you might as well ask why you aren't attaching it to a PCIe slot, which will have less latency.

You can't with an sff.

doesn't thunderbolt use pci-e connections

I'm not lazy, I'd get a job if I wasn't so lazy.

Yeah, like the rich. Literally a system designed to encourage laziness, corruption and criminal behavior.

Get a PCIe riser and cut a whole in the case. If you're using an external GPU you probably don't care about aesthetics.

>i would go to college, have great marks, do well in an internship and look for opportunities

>except i would rather sit at home and play games and post my opinions on the internet all day

>capitalism is working against me

>go to college
>inflated tuition and predatory loans

>entry level job
>5 years experience needed
>go to hundreds of interviews that waste your time when they will hire some random woman or the kid of an employee anyway

>end up unemployed and in crippling debt

this is reality

Fucking tempted at this point.

The connection is a PCIe bus. The connector is USB-C.

Wait, why would you have to cut a hole in the case? Also is performance affected by using a riser?

Oh, I guess you would have to cut a hole in the case for the cord. How are they normally oriented? Surely this isn't the intended design.

You'd cut a hole in the case so you can close it back up and keep somewhat proper airflow.

That can't be how these things are supposed to operate.

They aren't, but when you're dealing with an SSF and space is tight it's necessary. I'm more so thinking of a PCIe extension cable of sorts instead a riser.

This fucking pisses me off. None of this would even be an issue if computers weren't so goddamn expensive.

Then don't buy expensive computers?

The thunderbolt 3 ones should be as long as your motherboard allows pci-e over it. The main problem is that even the latest revision of thunderbolt only supports pcie 3x4 which will limit high end gpus. So I would wait until pcie 4 and a new revision of thunderbolt are available before trying them.

Actually that would also require a new laptop anyway. So you're just plain fucked.

That's the point though. If you don't buy an expensive computer you almost certainly are getting an sff, which is not upgradeable without severe augmentation.

Actually, I wanted to use this for a desktop. So triply fucked.

i used one on my thinkpad
jesus christ, I had to learn mandarin just to get it working

What's wrong with augmentation?
My shitposting is augmented

It's kind of a pity you can't use dual TB with them with one dedicated with In and the other an Out

It cost more money and effort. Why can't shit just work for a decent price? All of this shit is just a handful of goddamn sand and iron to begin with. Nothing but pure fucking profit. God damn it I hate capitalism.

theres like 0.5% performance loss, some reviewers experience none

7nm euv and smaller will cost shit tones as well and will never be as cheap as older bigger less complicated nodes
Rip 28nm and older

I'm achieving within 15% of an internal card (geekbench score). On a 3 yo laptop with integrated graphics. I had the card laying around and the egpu enclosure cost me $125.

You dont get 60%, youd get closer to 80 to 90%. You've never used one.

>comparing to igpu
stupid idiot redditor

>Not using your thinkpad's PCIE slot

>they're garbage.
You've never actually used one, have you?

Reading comprehension, not even once.

15% of the cards internal equivalent. Work on your depression you contrarIan retard.

>this is reality
Maybe your reality...

> what, you dont like shit? I bet you've never even tasted it.

>Being this ass mad...

Is that with the external card outputting to a separate monitor or is it to the laptop's own screen? What resolution or refresh rate?

How the hell did you get an EGPU enclosure so cheap?

>Inferior connection

You can have your GPU more than 6 meters away from your actual computer and it will have no impact on performance or latency.

>they will always be bottlednecked to 60% performance because they have an inferior connection.
Can any GPUs even exceed an x4 pci-e connection?

All these external GPUs use either thunderbolt or USB-c which are never as good as being connected to a motherboard

See for yourself.

youtube.com/watch?v=q5xvwPa3r7M

the card itself connects to the monitor, the laptop connects to the egpu via one cable.

The egpu also has a 2.5 and m.2, so effectively I've given my laptop extra storage options as well.

You have a significant performance drop if you have it drive your laptop's screen. That 15% hit for using an egpu on an external monitor could very well turn into 50%

Does PCIe over thunderbolt work with all devices and the normal drivers?

con't. drives a 1440p monitor.

Don't use it for gaymes, I use it for editing, but I have tested it in gookbench to ensure the GPU is working (which is how I calculated my perf hit compared to internal pcie cards) and the editing times have improved.

Yes, but unless you use macOS, you need to boot your system with the expansion chassis already on.

>Can any GPUs even exceed an x4 pci-e connection?
Yes, but it's not necessary for it to function (most stuff are done on the GPU, not copying memory back and forth).

>All of this shit is just a handful of goddamn sand and iron to begin with.
>Nothing but pure fucking profit.

Yeah shit's fucked. Google "DIY 10nm silicon fab"

I'd love to see an eGPU as part of a docking station or something.

Attached: Screenshot 2019-05-05 at 21.25.54.png (1062x1626, 3.24M)

I bet you look funny. You can overcome that by being friendly and not smelling bad, but if you can't do one of those things then you should just resign yourself to the neetlife. I've put it in a helpful graphic for you, pic related.

Attached: TRIANGLE OF SUCCESS.png (680x564, 16K)

That's not what he's talking about you monkey.

My Alienware laptop has a cool eGPU port that only Alienware containers will work in

What's this?

Low quality bait

yes

Attached: IMG_20190505_232846.jpg (3904x1054, 785K)

modular components are the future and once USB4 comes in you're going to see a lot more builds like this

Attached: intel nuc.jpg (1280x720, 89K)

you need all three of those fucking points, and that's the absolute minimum.

Just 22Gb/s of transfer speeds with PCIe over TB3 and a shit mobile cpu are a massive bottleneck for any relevant GPU.
And there will be no USB eGPUs,ever. It doesn't work that way.

Look up the specs for USB4 before making stupid posts

All they did was rename Thunderbolt?

So they've integrated Thunderbolt 3 into the spec? TB3 is still capped at 22Gb/s of PCIe transfer speed.

Why be autistic and use a non-argument argument?

Not sure why you would claim that PCIe link width (thunderbolt is not "capped", it's the equivalent of a x4 link) is a bottleneck. It would only be a bottleneck if you're transferring a lot of data to and from system memory (RAM).

>TB3 is still capped at 22Gb/s of PCIe transfer speed.

Which isn't a bottleneck

Attached: lle.png (591x394, 92K)

Of course it's capped
thunderbolttechnology.net/sites/default/files/Thunderbolt3_TechBrief_FINAL.pdf

It's a massive fucking bottleneck for modern games and GPUs which make more use of the available PCIe 3.0 x16 bandwidth. It's a stupid tech for poor people. I have used a 1060 in a TB3 enclosure and it's very much an inferior experience, can't imagine how limiting it would be with a more powerful GPU.

>That 15% hit for using an egpu on an external monitor could very well turn into 50%
Yeah, that's what I'd expect. It's pretty shit though, if you need an external monitor too I feel like such setups don't have too much appeal. At that point I'd rather skip the external enclosure, buy a cheaper laptop focused on portability and put the money saved towards building an actual desktop computer instead of fucking around with eGPUs.

>thunderbolttechnology.net/sites/default/files/Thunderbolt3_TechBrief_FINAL.pdf
It's not capped, it's literally what I said. Its specified to be up to x4 lanes.

>It's a massive fucking bottleneck for modern games and GPUs which make more use of the available PCIe 3.0 x16 bandwidth
But they don't.... If you'd have said machine learning tasks, you would have had a point. But for games, it doesn't matter because there isn't a lot of memory transfers between RAM and GPU memory.

Attached: Screenshot 2019-05-05 at 23.55.53.png (678x110, 25K)

Scroll to the bottom

Attached: Screenshot from 2019-05-05 23-58-28.png (998x535, 84K)

Sounds like you fucked up because a 1060 doesn't even come close to saturating that. You probably used the internal display on a laptop which does have a bigger hit.

You are missing the point, the cable has only enough copper wires to support 4 PCIe data lanes. Since PCIe x16 lanes are compatible with x1, x2, x4, and x8, that's not being capped, it's just not support. I don't know if you've seen other external PCIe cables before (such as Mini-SAS HD or iPass connectors, but a x16 link will require 4 or 2 cables. 22 Gbps is the maximum throughput of x4 PCIe.

Then you don't want an eGPU. Getting one only makes sense if you want to beef up an ultrabook or an older gaming laptop, IMO.

Nooooo, PCIe 3.0 x4 has almost 32 Gb/s of raw throughput (data + 128/130 parity coding). 22 Gb/s is a cap that was introduced by shintel

>PCIe 3.0 x4 has almost 32 Gb/s of raw throughput
Bandwidth != Throughput

It's entirely dependent on the game. Dirt Rally for example gets half as many FPS, on a 1440p monitor (which should effectively remove any cpu bottleneck). I've noticed that the newer the game the more use of the PCIe link it makes, so whenever I wanted to buy a game I'd have to look up how well it works on an eGPU. And more often than not that's not possible.

The maximum throughput of PCIe 3.0 x4 is 32 Gb/s, they literally gimped it.

See

DIRT rally would wan't to disagree with you. EGO engine has massive amounts of draw calls and is quite inefficient in it's post pipeline. While the bus isn't fully saturated, lower bandwith still introduces higher latencies where GPU stays still waiting to be fed data and that massively lowers the framerate. Just to say, at 1440p with the same details I get almost double the frames with a proper PC (still the same 1060 6GB).

So then fire up CUDA-Z and see that it manages almost full 32 Gb/s (when you take parity into consideration).

I'd really like to see actual benchmarks on this, but I'll have to take your word for it. I just think that people tend to falsely believe that modus operandi is to saturize the PCIe bus, but that rarely occurs. When it comes to data transfers, the most significant bottleneck is GPU memory speed when you miss the GPUs L2 cache anyway.

Higher bandwith gives you more speed so you benefit even without full saturation. It's just like roads irl.

wasted quints

That literally makes no sense. Bandwidth only becomes an issue when you saturise the link. Unused capacity doesn't "help" or "give you more speed".

You're maybe thinking about the latency, which of course is slightly higher when the GPU sits physically further away?

That is two remote GPUs, one in each computer, appearing local on a third computer.

(You)

>if you drive as fast as you can on an empty freeway, even more lanes means that you get to your destination faster
How does this make any sense to you?

Nice post faggot I know who you are

wasted

Attached: 13b.gif (680x498, 224K)

This, so much this!

checked

Attached: 1546827909269.png (386x366, 183K)

nice trips shame youre such a fag lol

Band WIDTH. Having 5 lanes doesn't matter when it's just you and a cop behind you with a speed limit of 60mph. What a waste of quads