#the-water-cooler
1 messages · Page 121 of 1
And if you are using one for a recipe you don't really want extra crap on the screen cluttering it up or to be switching between apps all the time
Honestly a smartphone app sounds about the absolute worst way to use a scale for cooking
Put a phone jack on it and run a cheap-ass phone cable or some other flat cable to a dumb 16x2 screen you can mount somewhere out of the way.
Or make them wireless and battery powered
I started off the day with no opinion on it but now I'm quite convinced a small dedicated screen is the best option for interacting with a kitchen scale.
Make it a small disc with a screen in the middle, you set a bowl on it, once it sees the weight stabilize for two seconds it holds that until it's lifted away and another bowl is sat down
Would still be annoying when adding to a certain weight though
this needs ai!
have a bowl recognition and it guesses both the density and fill status
Make the scale jiggle the bowl and have it guess it from the sound
whats the point of the removable screen?
you can put it in a place that doesnt get dirty or covered
Uh, the whole convo we just had
i didn't read it
I just bought a big chungus scale from a company that sells warehouse equipment.
Needed something that would do 10kg @ 1g precision, which nothing in the "kitchen" department will.
As a bonus, if I ever need to count 400 nuts and bolts by weight, it has that functionality 😄
you seem to be pretty hungry for cake
Pizza dough
Empty bowl from my stand mixer is a couple of kilograms, wanted to put nearly 4kg of dough in it
thats a lot you can ruin with pineapple
so you need a scale that comments on what you're making based on the ingredients you're weighing?
😋 😋 😋 🤢 🤮
mm pizza
detection for qf would be faulty, always would detect it as pile_of_money
Know how to?
Once you’ve got the density, there’s not much value of weighing versus a measuring cup.
shouldn't you be boarding your windows and blocking your doors from the wave of south american's unarmed robbers sweeping your area?
I think I see its eyes in the water bowl
it's that 3d printed water bowl. looks good.
see tom see's it haha
Nightmare fuel, can't unsee now
big words coming from something that is commonly turned into a handbag or cat toy.
No support in this off-topic channel. Try #integrations-archived (read the channel descriptions).
I think it's mostly from plant matter
so drive electric then, boom roasted
@shy comet it's coming down
@clear ferry https://www.youtube.com/watch?v=eJ6aZ_rAzVM
This is Debian 12 with the Gnome Flashback desktop. This is normally installed alongside the Gnome and Cinnamon desktop. But Gnome Flashback can also be run as the only desktop. It is very light and will suit both old and new hardware.
Ethanol and electric, if they’re serious. Personally, I got the leather seats.
electric is generated by coal mostly which is made of dead animals
ethanol is made form grain which uses nutrients in the soil which are made from dead animals
USA is below 20% coal power. Anyway, I think it’s mostly from plants.
So what you're saying is that they're restricted to bamboo bicycles
But they can't have tires
bamboo is to keep pandas alive
every bambo bike kills 19 pandas
being a "true" vegan is somewhat impossible really.
ultimately folk all have different reasons/meanings to their veggie/veganism
my friend had a brooks saddle on his bike and was so prod of the hipster cred... then i pointed it out to him. "you know thats made form a cow right?"
Or for me: I had pork earlier. So errm, very good vegan yes yes
(I am not a veggie or vegan)
yes. you pick your battles for being a vegan
veggie is a different topic
veggie has a logic to it
rather, a consistency
dont kill animal. simple enough
will say though. much as I love a good bit of beef. Some of the plant burgers around these days are pretty good in their own right
see I'm opposite. Not a fan of chicken
I will eat it in a pinch, but really not a fan
yer missing out on portuguese style chicken. mmm
haha
actually, thats my lunch. chicken, bacon, and some form of cream pasta
hmm
ooh, no. rice, peas, crumbled bacon and chicken
mmmmm
haha
One minor issue I have, a lot of pure vegan diets rely heavily on nuts. Useful. Allergic.
would certainly restrict options a bit
i dont understan nut alergy people... peanuts are peas, cashews are fruit seeds, why would you be alergic to both?
hehe
hehe
nah I am lucky, no anaphylaxis
still a deeply unpleasant experience with what does happen. But not gonna die.
now i wait for pcb's
Are you running a wire out for every button?
in this case yes.
i will eventually do a V2 which will have a matrix, the charger on the board which makes room for a bigger battery, and a few other features
maybe a SNES variant, but there are issues with that (too thin)
i could also instead of a matrix do what the nes has which is a shift register
decided to start simple though
i also want v2 to have an input for player 2.
(right now p2 would need a usb controller)
eep
hehe
You could just buy the NSO NES controllers too
Can get them on eBay with a small markup if you don't have an NSO subscription
Or I could invite you to my family plan
Though their charging method is a bit annoying
They have SNES ones too
https://www.nintendo.com/us/store/products/super-nintendo-entertainment-system-controller/
O_o
im not sure those worth with retropie. need to check
i know some of the playstation ones do
player 2 will have some lag with wireless as well of course. but i dont think this is a competitive gaming rig
:p
ah. bluetooth and only work with nintendo stuff it seems
bah
They work with PCs
Never tried on a retropie
I use them on original consoles with bluetooth adapters
They can be a bit wonky having more than one connected to the same bluetooth adapter though
Some weird Nintendo crap.
i read up. they do work with retropie but they wont automatically connect on boot
but, maybe we can figure out how to fix that
also with my case being aluminium, it will block the native bluetooth.
What does it take to write LLM prompt engineer on your linkedin page.
A certain lack of real skills
more skills than the average invuencer
No
Well I just made a prompt for my anwsering machine and it takes some clear thinking
improved it. *
you are only entitled to add this to your linkedin if you also are a blockchain entrepreneur and nft enthusiast
crypto something, sparkle k8s, something something ai on the edge
LOL. blockchain currency are bad for the enviroment
no they are great, it kills humans
There is no such thing as a prompt engineer
you're just jealous. you cant make up a prompt for your voicemail system.
Such skill
its worth upwards of 250k a year i heard
250k rupees
serious 80's vibe with this park https://youtu.be/qFuPpHTtGrE?si=tu449Xnf5Qt7DGdS
須磨浦山上遊園
住所:神戸市須磨区一ノ谷町5丁目3番2号
カーレーター
「カーレーター」は、ロープウェイ山上駅から園内の山頂までを結ぶ珍しい乗り物です。昭和レトロな雰囲気の乗り場から2人乗りのカゴのようなものに乗り込み、振動しながらゆっくりと傾斜を進みます。その「乗り心地の悪さ」が人気で、まるでアトラクションを楽しむような体験ができます。
回転展望閣・屋上展望台
回転展望閣は、昭和33年に営業をスタートし、須磨浦山上遊園のシンボルです。3階建てのビヤ樽型の建物で、屋上には絶景を見渡す展望台があります。特に3階の喫茶店「コスモス」では、ターンテーブルによって床が360度ゆっくり回転し、神戸のパノラマを楽しむことができます。全てのテーブルが窓際に置かれており、ドリンクや軽食とともに不思...
Does whoever is in charge of the automations page in HA hate us?
Did we do something wrong?
This new UI is really awful
Why are things that used to be one or two clicks away buried under 4 layers of confusing menus
You can disable that
oh right, I reverted to 2024.02. I was super confused, which new layout? D:
I'm still on 2024.2.5, I haven't made an automation in a while though.
I don't know when it was added
Whew. That was a journey and a half.
For anyone playing along at home, the solution was to download the Windows11 Install Media Creation Tool, have that download the full win11 23h2 installation and make an .iso from that, mount that iso and then install the update that way.
that's a complicated way to install gentoo
What I don't get is why has the automations page been redone a couple times in the last year but there's still no folders or tags available
Lose the training wheels and write your automations in YAML. Then you can organise them however you wish.
If I wanted to waste time googling for field names and writing YAML I wouldn't be using a GUI.
the automation UI is far superior to writing yaml
isn't there a jsonschema for them? Then you can just have vscode autocomplete it for you
Yea, 'stick to the yaml' misses half the point of HA. If not for the GUI I wouldn't be here, lol. I'd still be doing my own thing.
VSCode has an HA extension with auto complete.
node based editor could be fun though
agreed, its a dumb response
with how much services learned to have in/out
Though auto-complete is a PITA most of the time. I end up with duplicates like -platform: state -platform: state most of the time
yea, silly things like that happen too often :/
Maybe somebody will make vscode as an addon with intellisense already set up for HA
I've recently really had issues with having things like functionName()() in vscode 😦
shouldn't be too hard. There's way to just host it for browser consumption
I'm not going to, but that does sound like something somebody would make.
This add-on is provided by the Home Assistant Community Add-ons project. Visual Studio Code, accessible through the browser. About This add-on runs Visual Studio Code, allowing you to edit your Home Assistant configuration straight from the web browser and can be embedded straight into the Home Assistant frontend UI. Visual Stu...
I don't run HAOS, so 🤷
I have HAOS running in a VM, I've yet to encounter anything I wanted to do that that didn't do.
its a good combo. Running a full VM does have quite some overhead, but on my nas thats not really a problem
I have pfSense and HA in VMs and Frigate/my dashboard monitors along with another dozen random services running in containers on a 5700g for mine.
Even then it's not doing entity names.
I also feel using vscode is a bad fix when the problem is 'needs a sortable tag'
I have grafana and influx and vlc and esphome and whisper/piper/openwakeword installed in HA as addons, they're all used HA-exclusively and I like them being contained in their own VM with HAOS.
Though I might move whisper to a container on the host so I can use my GPU for it. Not sure if the apu on a 5700g would be worthwhile for that.
I run a k8s cluster either way
I could experiment with HAOS in kubevirt
But ehhh
Do they even have amd support for whisper? Oder is it one of the vids only things?
I'm running it cpu side right now and it's just a little slow for what I'd like
But right now the llm I'm using is huge and slow too so
What are you doing up there?
I needed a better view, but I just got stuck :/
Trying to spot the best garbage cans in the neighbourhood I bet
Spider Coon, Spider Coon
the videos showcasing people going through the trouble of swapping out old device barrel jacks etc to use a usb c instead
$0.37 adapter on AliExpress, probably
I've done that for a couple things.
It's convenient.
Did it with a GBA SP, did it with a flashlight, a portable heated mug I have, and something else I can't recall right now
The last two were because they were damaged, the GBA was because the GBA chargers are annoyingly short
I did the usb-c mod with a gba-sp aswell, nearly killed the mainboard in that process.
dont they have different pin outs/footprints/
They make adapter boards
PCBs with half through-holes on the sides to solder to the gba and a usb-c port already on top
I need the competitive edge
I had to patch a couple pads on mine
I ripped like 3 traces up removing the original port, they were like wet paper
idk what they did or if it was because it was so old but those traces lifted so easily
But I only ripped like 2-3mm off of all of them so they bodged easily
The GBA I have had chemical damage on the shell when I got ahold of it so idk if that's related
The shell just kinda crumbled
Not a pet!
This beautiful and unusual wild cat had to be rescued and rehabilitated. He is humanized, which means that he is used to people, but he is not tame, which means that he won't bite anyone, but he also doesn't let anyone approach him or pet him. The big surprise was when he decided to approach the carer, wanting to befriend her, even th...
I didn't even know that was a thing! I would so have one as a pet
Uhh, nice. One of the features I was missing in my usage is coming for paperless-ngx https://github.com/paperless-ngx/paperless-ngx/pull/6094
I got ROCm/whisper working on my 5700g's integrated gpu.
It appears to be rather pointless though, it's like 10% faster than doing it on cpu
Not even, sometimes it's like 30% slower.
Yea that was pointless
nice. and bummer D:
It's also really unstable and keeps crashing the gpu requiring a reboot to resolve
Which since this is on my pfSense box, I think I'm done playing with this.
@tidal bronze think I found where i want to move to. https://youtu.be/X2wLMp67Cls?si=Mpv7eN0kZ_lrKJPM
An abandoned McDonald’s on an island in Alaska has turned into a time capsule. The island of Adak in the Bering Sea used to be home to an army base built in 1942. It was decommissioned in the early '90s and most people left the remote location. Photographer Chris Luckhardt documents places long after they’ve been abandoned. He set out to photogr...
pop of 33 people. sounds just perfect
youll have to pay me to eat the crap from mcdonalds
Is anyone else starting to get unreasonably angry at cloud products? I feel like it's planned obsolescence, and that you pay extra for it. I'm shopping for a new fridge/freezer atm and all of the top options are cloud only so I refuse to buy them.
and it's frustrating because I love tech, I want the fancy fridge freezer with variable temperature zones and that you can tap on the glass to see inside, like, it's cool. But not when it has forced cloud. We're getting aircon too and that's cloud by default and I have to build my own board to fix it
That's absolutely true. It's a way to make you keep paying for the product after purchase or lock you into their "ecosystem".
BUT it is also very conventient for most people.
one can only hope that the EU forces more interoperability and ownership
Most People ™️ don't want to set up a server for local control. They want to just install an app on their phone and have it work even off the network.
But then again Most People ™️ don't need to ever connect to their freezer either. That's a set-and-forget thing.
big tech wants to force you into their ecosystem so you become so dependent on it that moving to a different manufacturer will seem like too much effort and so you stay 'loyal'
Learning from Apple
ofcourse apple and google are by far the worst examples of this
I wouldn't mind being able to see the insides of my fridge on my phone while at the grocery store
but I also don't need that
ye, there's a reason the products are generally cloud enabled. Not just for the milking.
Indeed, it's horrible, I also think that good easy to use products can be made local. What's to stop the fridge from being Zigbee? works out of the box with all the popular stuff like Amazon echo, just job done. I think they should follow standards or at least give some way for the product to function when they drop support
which company commonly included in "big tech" sells fridges? o.0
Samsung, LG
Neither are commonly included in big tech. Though Samsung probably should
FWIW, I'm mostly on building products that are defensive against local access like that
though they are for B2B, and they are generally actually like the security we can provide by locking it down.
And want to have a support relationship either way, so they don't really care if we bill them 20€ for support, or 15€ for support and 5€ for cloud^^
For me, I feel that any product that relies on cloud integration has an unknown future expiry date. The company will, at some point, stop supporting it, and then features that you paid for will be gone with basically no path for repair or restoration of those features. So I refuse to buy such products
don't ever buy a "smart" fridge or freezer. they are nothing but headaches and will be a fault of a "bad" appliance before the core of what makes those things work fail
I have an underground propane tank that I want to track. Long term, I plan to head down the path of connecting a meter directly to the gauge. Short term, I want something to remind me to check the gauge and input a value so I can track it over time. My ideal scenario is that I receive an email or mobile notification asking me for the current fill%. I then enter a number and that number is stored somewhere in a database or spreadsheet that I can see how it is changing over time. Any ideas?
I'm trying to get rid of as much friction as possible so I will actually do it.
Hmm, I wonder if I can use a power sensing smart socket on my scanner to figure out when it's scanning and kick off automations based on that 😂
you can use an input number and write to that using a service called from an automation. #automations-archived can help with that, there is no support in this channel
I do something similar with my washing machine
And you can do more general pattern recognition with a bit of machine learning
Which I did for my coffee machine lol
Thanks for the input. I was trying to be broad in my question since HA might not be the best answer for what I am trying to do. Wasn't sure if I am missing some easy Google Sheets add on or something that will get it done.
I dont see why you would need google sheets
I'm taking a wild gander that we have the same ISP in the office that I have at home
I have 5.4ms latency to a switch in the office
google reads this discord.... youtube just served me an LTT video on this mod
I am already using my phone, no matter what, cause the recipe is on there, and so are the timers
I know of at least one googler reading this chat 😂
I've heard length(feckoff) is defined < 20 so 19 more feck offs
feck off ?
can't find anything on the topic anymore :/
but my don't recommend button doesn't do anything anymore I believe
I definitely have more than 20 channels blocked on Youtube, probably between LTT and Simon Whistler alone
I don't believe that the thumbs-down button does anything
Youtube music just will not get the hint that I do not want to listen to Bruno Mars. At this point must have hit the feckoff button on 50 or more songs
How dare you not follow the trend set by the majority?
How will the pure ML models cope? D:
oh linkedin is so much worse. The suggested posts especially
Pretty sure at this point the algorithms are just payola by another name.
At one point that was a scandal. Now it's business as usual
It's the matter-of-fact-ness of that that makes it
What did you do last week? Well, this happened...
"happened"
Now that's what I want to see under suggested posts
i did that with mr beast... it replaced it with religion videos.... (nothing against mr beast, but i dnever clicked on in my life so why are you making half my feed page that)
I used to get a ton of MrBest too
Don't like middle earth anymore?
Here's a better video on that city
Welcome to America's modern-day ghost town - a place completely frozen in time.
Once a bustling city of 6000 residents, today only an estimated 77 year-round residents remain. Once home to one of America's most important Naval Air bases during the Cold War thanks to its strategic position at the edge of Asia (closer to Tokyo, Japan than Seat...
tl;dr: A town was made for monitoring the USSR during the cold war, after 1991 it basically died overnight
On the closest island off Alaska to the USSR
And now some private company owns it all and tries to bring in tourists
I guess that's what the 33 remaining people support 🙂
More or less, yea. Most of the city was crumbling.
and they apparently don't eat McDonald's
When they left they just left
There's a three-scoops-of-BS "documentary" on Netflix called "Pirate Gold of Adak Island" which if you ignore the manufactured drama and fruity nonsense has a reasonable amount of history of the area
don't think Aleutian Outfitters owns it. they just bought some of the abandoned buildings and run the tourist aspect
good video tho
really?
people are going to end up not even knowing how to talk to one another without ai assistance
like some high-tech Cyrano de Bergerac
think i need to increase my camera resolution. frigate can't detect this pesky racoon in my back yard
Just think tomorrow's doctors will likely float through schooling in such fashion.. exciting times!
I've been partying with medicine students before AI. I'm worried enough already...
maybe we will get lucky and just skip to this https://youtu.be/3M4WNh_82BA?si=XryFkRr2xEnYcH4o
Robot Chicken is an American adult stop motion-animated sketch comedy television series created by Seth Green and Matthew Senreich for Cartoon Network's nighttime programming block Adult Swim. The writers, most prominently Green, also provide many of the voices. Senreich, Douglas Goldstein, and Tom Root were formerly writers for the popular acti...
It's easy to weed out the cheaters
you mean after enough people die and they catch on?
no, already during their education
no idea what they actually have to do but I have 0 doubts that if there is an easy way some are going to take it
like the people who think ai is implemented as a bunch of if/else statements lol
Well, to some extent it probably could be, not very dynamic though
the model doesnt have any if else statements. What ever infra they built around it could have, but not related to the AI itself
But a lot of people also think AI is incredible and has a mind of its own
My 6400 GPU cluster should be online in about a month now
Then to deploy two more of them
So a single socket system then
I'm still waiting for the confirmation about the superspine upgrade, just 8192+4096 cables plus 64 new switches and 12k trancievers
HPC go brrrrrrrrrr
yeah its sitting under a desk
we have a HPC cluster as well that they just built, I have no clue what the specs are
must be peanuts for @clear ferry
Quite impressive specs for 2001

Except for the number of GPUs I actually have more cores and memory in my garage, albeit older generations
One of those nodes actually looks usable 😂
The last one yes
Still pretty small in terms of cores and memory, might actually be limited by the amount of memory unless it has all nvme
I rolled out 8 x 8A100s recently, and they had 128 cores and 2TB each
seems like mechE has had an upgrade
No GPUs, older workloads I guess
I still know a company running a HPC load on windows, with infiniband and scheduler
Application is from the late 90s I believe
no clue what any of this means but FYI
Operating System: Rocky Linux 8 (100% bug-for-bug compatible with Red Hat Enterprise Linux®)
Cluster Management: TrinityX
User Management: LDAP with TU/e Active Directory for Authentication
Module system: Lmod
Software build system: EasyBuild
Remote web access: Open Ondemand
interesting
shit looks like something out of a movie. crazy https://youtu.be/sweWMxpTwKg?si=p1h9Em_N1u-ZsAAG
In the US city of Baltimore a major bridge has collapsed after being struck by a cargo ship.
The ship caught fire on impact, then sank.
Multiple vehicles fell into the water below -- and Baltimore police say rescue efforts are underway.
The Francis Scott Key Bridge
Subscribe to our channel: http://bit.ly/AJSubscribe
Follow us on Twitter: http...
I've actually driven across that bridge
here is the whole clip. can actually see the ship strike the pier 5 mins in. https://youtu.be/SDceU9x58vc?si=mDYUGqYAsESBA4mi
Francis Scott Key Bridge Collapse after being struck by a large container ship.
Location: Baltimore, MD
Date: March 26, 2024
Info: Webcam Footage from the Port of Baltimore, livestream of incident and current situation available on the Port’s YouTube channel. Uploading captured footage.
Developing story, casualties unknown at this time a...
the way the bridged collapse is unreal
I am more scared about the comments that think this is an issue with the bridge
this will prompt some drastic change to how the piers are protected. there is no doubt
They will invent more concrete concrete, as opposed to the currently used ambiguous concrete
people freak.. someone see's $$ and the next thing you know tax payers are paying out the ass for some "clever" design that in theory will prevent future incidents like this from happening and those designs will cost an arm and leg
like canadian sheds that have fancy recessed lighting. over priced for what needs to be done
Reminds me of that one bridge in India which collapsed multiple times
this will be closer to that scene where those workers watched that kaiju break through the massive wall they just built in seconds
those supports do look flimsy as hell
Those boats plow through anything, might as well not make it more expensive as needed
If it's not a boat, someone on a mobility scooter will get em
just odd to see it doesnt have any protection at all
not much you can do.
that bridge even had what looks like power line markers placed on the same side that the ship had to miss prior to striking the pier
there's a lot you can do
if only money grew on trees
seems like top part of the ship is what hits the struts first
that was easily preventable
it all was preventable if the captain of that ship didn't fuck up
it's going at snail's pace
those things are like 1300ft in water.. they take a whole lot of space to maneuver but they had plenty to set themselves straight but rather than playing it safe they probably "hoped" it was going to work out
thats like what? 8 lanes? and maybe only half the span across water lol
don't think bridge is high enough off the water to accommodate cargo ships either
yeah, well obviously you could scale it up
its 12 lanes
but I dont think the bridge ever stood a chance lol
no. pretty much king kong sitting on your house
qq, i`m get core update 2024.3.3, but Drag and drop not work, my UI dosnt changed
You know those bridge designs are consisdered "fracture critical" and will collapse with only a single point of damage
#frontend-archived can help, this is the HA free channel 😉
The whole bridge is under tension, breaking a single link causes those kinds of bridges to literally tear themselves apart
It's..
Like it's their whole design
There is a reason new bridges are all suspension bridges
They probably were using tuya relays on the boat
boatassistant
raspberry pi2
Apparently the boat had put out mayday warnings
And so additional traffic had already been stopped entering the bridge
Yikes, why didn't they pull the rest of the workers...
Probably in process of doing so
As it is they refined 7 missing down to 6 because one was found. Not to make light of it but maybe that one ran faster
Unironically if I lived on a houseboat, that'd have to be the hostname for my HA setup
What's not clear right now is I've seen one statement that only 2 people were plucked from water, another that a few vehicles definitely went in, another that people from vehicles are accounted for and another that 20 were rescued. It's a tad confusing what's going on
But it'll probably take quite some time to pick through all that
You could definitely see the traffic dissipate when you watch the longer video above
It was pretty busy a few mins before
Quite the slow motion disaster
Interesting that they still had people working on it
Apparently it had lost propulsion
But with a boat, that's kinda what you use to slow down
Like I know someone above suggested slapping some concrete around the bridge pillars, the reason they don't is simple. There's footage out there of what happens when ships that big slam into a dock. The concrete isn't winning.
The ship certainly doesn't win either. Mutually assured destruction.
F'in physics
The concrete industry will tell you that more concrete would help! They just need the check to fill up the whole channel and that won't happen ever again
also why didn't the bridge have a gun???
If the current goes that way it's not even going to slow down
Even in stationary water, using propulsion to slow itself down takes maaaaaaany miles
Yup it's wild how efficient moving on water is
A dock is not meant to protect bridge pillars
I was just surprised this bridge had no protection at all
Which is very different from bridges here
They easily carve through meters of that stuff
At the speed that thing was going, looked more like it kissed it
Hence previous statements on they need miles to decelerate properly
In nautical terms that thing was doing 90 in a 30 zone
And it was increasing in speed
8 knots apparently at time of impact, not anywhere near its top speed, but ya think about how much it weighs, it'll take more than just a tire hanging off the edge of a dock
I've got 7.3-8.6 estimates according to some news sites here
I'm sure they'll figure out for failure analysis but I don't think the end result between those two extremes is much different
Bridges have always been fragile structures, hard to build redundancy into them, they have that problem of more structure thrown at strengthening them is more shit for them to hold
Ships can demolish a concrete dock
As far as we can tell so far, the ship did suffer onboard mechanical and electrical failures
A very very unfortunate accident
And no doubt there'll be further reviews into whether this sorta thing can be prevented in future
Redundancy is easy, just building a second bridge nearby. Still doesn't help the poor ones on the collapsing bridge.
Ah I see the discrepancy, I didn't realise the thing was leaving, so yeah, 3 knots when tugs released and accelerating to leave port
At least the ship is still afloat. Hopefully it stays that way
Part of the bridge is sitting on it
Yeah. Gonna be an awkward recovery
And of course knock on economic impact: blocking the port
Looks hairy to drive around it
Might not be deep enough outside of that main channel anyway
The type of impact structures that Stefan linked earlier aren't necessarily designed to withstand the impact but to push the ship off from hitting the pier itself. Redirecting the energy wouldn't require anywhere near as much force
It's "only" 15m deep there
Can see in one of the videos BBC uploaded the ship approaching the bridge, lights going out, then they come back on and the thing kicks out some smoke
And impact
Being realistic, probably not much hope of the 6 missing turning up alive
Unfortunate but a surprisingly good outcome tbh
Yeah, if there was no warning at all or it was a busier daytime occasion or something
Could have been so much worse than it is
They supposedly dragged their anchor for a bit. I wonder if it had any impact at all
Francis Scott Key Bridge collapses after being struck by a large cargo ship.
Credit: @brick_suit5223 on X
thats why it slowed down before collision
I think the anchor is why it steered right into the pier
cause that sucker cut hard
https://www.youtube.com/watch?v=qZbUXewlQDk Longer video, not zoomed in
Dali Lost Power Outbound from Baltimore
What's Going on With Shipping?
March 26, 2024
In this episode on the Red Sea, Bab el-Mandeb🍺, Gulf of Aden, Indian Ocean, Somali Basin, and Strait of Hormuz, Sal Mercogliano - maritime historian at Campbell University (@campbelledu) and former merchant mariner - discusses the allision between MV Dali a...
I saw a time lapse and it was chilling to see a car race across the bridge 'just' before the collapse
That driver probably needs a good cleaning of their seating and pants
They said people were missing but I didn't see any cars on the bridge. Unless maybe some couldn't be seen from that angle, maybe going the other way. Or maybe there's also foot traffic, though it was pretty late I believe
They located a couple cars that sank via infrared cameras
I saw a lot of lights from workers cars on top of the bridge at least
Good thing it was probably closed for regular traffic
46 people dead. wow.
Oof that many? I was hoping it was closer to six or so that was mentioned earlier 😦
Where's the 46 figure from then?
46 dead, 2 missing i mean
implying they have recovered 46 bodies
hm. can no longer find the article saying 46. maybe it was a typo
where are you pulling this number from? your butt?
so far the count is 8. two have been rescued and 6 are still missing
yeah
i think the 46 i saw was an article comparing other bridge disasters. i hate the way news is presented.
think you got it from this lol https://en.wikipedia.org/wiki/Summer_Street_Bridge_disaster
On November 7, 1916, a streetcar loaded with passengers ran off the open Summer Street Bridge, a drawbridge, into Fort Point Channel near downtown Boston, Massachusetts, United States. Forty-six passengers were killed, making it the deadliest disaster in Boston's history until surpassed by the Cocoanut Grove fire in 1942.
nope. it was ohio
every ones gotta make a garbage story about their own bridges to get clicks
when the article is presented on google is says key bridge.. and 46 dead
Google definitely... knows stuff
Last night's gem, while reminding myself about fibre optic transceivers
Yea you don't want to starve or over feed the fiber
Google search is so done. It won't be too long before the AI systems make it completely obsolete
You can't find anything with it these days without using modifiers like "site:" etc or google scholar instead
SEO was already bad enough, but when you combine that with AI produced crap articles that get boosted to the top it's over man
Even site: is cooked. So many useless websites polluting your results because they bought a .yourcountry domain.
Pinterest can be >50% of my search results some times, even though it brings absolutely nothing to the table
bing was good for some time, now its just a replica of google
I like to use yandex from time to time
If your goal is to avoid made-up bullshit, relying on ChatGPT might not be your best move
It's a risk I'm willing to take
Bing or copilot or whatever they branded it these days is just incredible for some stuff. Saves a bunch of time if you can trust what it makes up lol
Supposedly they're applying some sort of fact checking on top of the normal AI response
It's much better than ChatGPT (3.5 is the free one?) and provides sources so you can double check
For now you have to manually check but it's not far-off from being reliable from my experience
I'm terrified by the people who will be relying on AI to fix their code or help them write commands in a cli
Thankfully my wife is sane enough to deny users to use copilot to access to their internal code
AI that gives you recommendation by running locally only, sure, learning by reading your internal code and sending it off site, yeah I'm gonna pass
🤷 I'll give my corp code to our corp overlords
Even better, it learns by reading rubbish code
And gives rubbish advice

Aka reading my code
Okay grandpa
But.... I am the cloud
better?
I always wonder why companies are so hellbend on protection their shitty code
Like what stuff?
I let it proofread some things I wrote to correct grammar and stuff. It is a language model after all, that's what it's good at
But I don't trust it with anything content related
business™️ secrets™️
mixtral + ollama + vscode continue plugin is what I'm using, it's all self-hosted.
It's not GPT quality but it's good enough to ask questions to and to ask to summarize files
It can also auto-comment and auto-write code, but the auto-written code in my experience is mostly useless
Also 24GB VRAM is about the minimal to make it usable
But I imagine you have access to more
I get really good results out of mixtral with a 32k context size but I also only get like two tokens a second off of a 4090 with it
Which means commenting a full file is not the fastest, but it can do it
I'm getting really tempted to get another GPU for my office pc and take it's 3090ti out and put it in my vr pc with the 4090 so I can get 48gb vram but I also really don't want to
huh? is there some fancy SLI trick that makes VRAM additive?
You can just add the vram together if they're on the same system and they'll share over the pci-e bus
I think the higher end cards get advantage from nv-link but I only have one card with that and I'm not buying another
They're not making huge 1TB VRAM GPUs to do training on, they're clustering a bunch of smaller cards.
pci-e gen 4 x16 is ~32GBPS, the vram on a 4090 at full power is ~21GBPS. There's a latency hit but no bandwidth hit.
I can ask my ML-infra colleagues, but looking around a bit, that's not how it works 🤷
I've also not seen insane memory requirements for training tasks, but insane hour requirements
which then translate to memory, since the model is replicated to tons of GPUs and thus your 24GB model now needs 240GB, but also simply runs on 10 GPUs at once
You can ask them, that's definitely how it works.
You only get one 4090 of compute but you can get the vram from both
You could just google it instead of asking them, too.
I did google, and the only method that showed combining them uses both computes and transfers part of the data inference is run on, not the model weights
https://github.com/ggerganov/llama.cpp/pull/1607
Here's where it was added to llama.cpp last May
https://www.reddit.com/r/LocalLLaMA/comments/142rm0m/llamacpp_multi_gpu_support_has_been_merged/ Reddit thread
https://medium.com/@yash9439/run-any-llm-on-distributed-multiple-gpus-locally-using-llama-cpp-2ff478a0dc3c Article on it
https://github.com/meta-llama/llama-recipes/blob/main/docs/multi_gpu.md Page on multi-gpu training
https://github.com/ollama/ollama/blob/913306f4fdc264c76fdd36f40298376680973ecd/docs/gpu.md?plain=1#L34 Page on ollama configuration going over multi gpu configuration
Yea, looks like nvlink does decrease latency significantly but isn't required
Anyways, LLMs are basically built on the back of multi-gpu compute/vram. There's no way to train an LLM from scratch on a current single GPU.
At least not given our lifespans
Looks like there's methods to distribute parts of the compute now too
But not all of it
the first link at least confirms my view of things, let's read some more
yea, nothing suggests that this uses all GPU's memory as single memory space for a compute GPU
But that it splits the weights across GPUs and copies data around between them
I don't know how it confirms what you're saying, there's people in the comments happy that they can now run larger models on gpu.
At a quite huge performance loss. The initial PR you linked statese 70% better performance at 4times theoretical compute
https://github.com/ggerganov/llama.cpp/pull/1607#issuecomment-1565428195 Literally the top comment thanks him for adding the ability to use more VRAM
https://github.com/ggerganov/llama.cpp/pull/1607#issuecomment-1565428195 It was the difference between this guy running a model or not
https://old.reddit.com/r/LocalLLaMA/comments/142rm0m/llamacpp_multi_gpu_support_has_been_merged/jn8fqf9/ Reddit comment confirming it
https://www.reddit.com/r/LocalLLaMA/comments/15zx322/ideal_setup_for_dual_4090/jxo0zl6/ Guy explaining it more
It doesn't replicate the data across each gpu
Yea, the transfers add latency and therefore slow it down compared to running a model that fits on one card but a lot of models just don't fit on one card
But right now I'm sharing system ram with my gpu to use it and that's even slower
Though I am seeing reports that a 3090 and 4090 can't work together which might kill my plan
You know, you could try and understand what I'm telling you
You said, quoting, "and thus your 24GB model now needs 240GB" which is false.
[04:56]ongy: I did google, and the only method that showed combining them uses both computes and transfers part of the data inference is run on, not the model weights
which is also what I re-iterated later on.
But it's not something you'd do usually. As it's a nasty performance hit
What do you think people are running these 60-100GB models on? Either multi-gpu setups like this, or just sharing system ram.
There's not exactly a lot of options for consumers to get high vram gpus
I frankly don't care about consumer ML
Then you're not even having the same conversation I'm having.
🤷 if you want to tell me that two horses are as good as a car because you cannot afford the car
Never said they were the best money could buy, just said they'd be better than sharing system ram.
But, hey, reading is hard.
no, you said that it pools ram at effectively the same throughput
Nope
and behaves like a higher memory card (roughly) with just one side's compute
I said the pci-e bandwidth is close and they can share VRAM while noting there was a hit
But, I get it, reading is hard. I'm done here.
yea, you should practice it to understand the sources you link to 🙂
Says the guy who claimed it'd replicate all the data on each card
if you want to do it proper 😂
Again, you completely missed the context of what I'm running and are for some reason assuming I have a hundred grand of equipment to play with at home.
But if you cared enough to even read the PR, you'd realize that there's no sense of the inter-GPU links being fast enough
and things being split in both compute and memory
no, I don't care too much about data vs. model parallelism
I care about your miscategorizations of how it works
For practical purposes an end-user can add the vram together for local compute using libraries that split it over multiple memory pools.
That's just fact
You're being pedantic and whiny arguing about categorization.
I'm talking about running a model at home for basic code completion and commenting, man.
🤷 not like you'll ever have to care about the difference
I'm not wrong though 😂
Yea, you are, because you are ignoring the context around what was said and are trying to treat it like a line in a technical manual.
You were wrong, you doubled down on it, when I pointed out what I meant by what I said you got whiny and defensive about categorization. You've contributed nothing except for pedantism and pointing out a different method of using the GPUs exists, completely missing I was clearly talking about one specific way.
It doesn't exactly take a genius to see I wasn't talking about running a huge 10 gpu training cluster
I was wrong, I found the way it's working, pointed out your model is utterly wrong. You trippled down on remaining wrong
I'm right though, nothing I said was wrong. Oversimplified at worst.
You just want to be pedantic
You trying to claim that it doesn't need to copy the data form one GPU to the other (that's inferred on, not the model) is plain wrong
Your idea that it only uses one GPUs compute is also wrong
I never said it didn't need to copy it
ofc it needs to copy it
It'd need to do a DMA request and pull it into the local compute just the same as from system ram, the vram is just faster at it
what do you think it copies?
Who the fuck cares, that's irrelevant to what I was saying.
ok, now I understand how shallow you are thinking about this D:
If you just want to prove you know more than me then just say so, I'll agree. But this pedantic "I'm intentionally missing the point of what you're saying because I want to seem smart" crap is absurd.
I don't know a ton about LLMs, I've had ollama set up for like two weeks.
wasn't intentional. Just different way of thinking about things. E.g.
04:55]Netham45: You only get one 4090 of compute but you can get the vram from both
makes me think of using one side's compute only. While I suspect you meant that hte performance improvement is close to none over fitting it in one GPU.
It's a huge difference to me, but I get why you don't care about the difference.
I did correct that part, I thought it was the entire compute was limited to one GPU but it looks like only part of the compute process limited to one gpu.
But, yea, in the context of 'I want to run mixtral q6' I don't really care about how it's done. I'm not at that level with them yet.
Anyways, are you sure you cannot combine 3090 with 4090? 🤔 I'd suspect it works, just not NVlink, since that's super generation specific
4090 dropped nvlink anyways
I'm not sure, I saw one discussion that said that and haven't found anything else
nvidia and their bloody market segmentation =.=
It just as may have been them misconfiguring it
a bit expensive to "just try it. LOL"
I already have a 3090ti and a 4090, though I don't have a case or psu big enough, or overwhelming desire to put them in the same box.
yea, the bigger/professional machines just have one type of GPU either way. Since they usually are bought as single specced system
Also some limitations on k8s, which is used a lot for managing the workloads. Though those are lifted soon™
my latest GPU I own is a Radeon HD 6950 😂
I have more fun things to play with at work though...
I got a 3090ti for VR but nvidia bugged out their displayport implementation on specifically the 3090ti ( https://nvidia.custhelp.com/app/answers/detail/a_id/5382 ) so I got the 4090 instead and put the 3090ti in my office PC.
we have some 3090s in headless servers^^
I only ever game on one at a time so I was thinking of starting ollama on both and setting up a load balancing proxy to redirect api calls to whichever one is under less load, then I could basically always have an llm accessible while not having gaming affected
I really wish there was a way to pass a pci device over infiniband
That's sounds like a world of pain
RDMA not good enough?
I don't believe my cards support gpudirect rdma
More nvidia market segmentation
Have you tried that on your 3090s?
I heard there's a driver hack to get vGPU working on them, maybe there's a similar one for gpudirect
Not that I know of.
they are good enough to run inference on device. And for training we use stupid sized clusters in cloud
I set up soft-RoCE between two test boxes and got rdma working over that but didn't actually try the gpus since it seemed like they wouldn't support it
And the test boxes weren't the boxes with the gpus in them, heh.
Figured if it worked I'd get some mellanox cards and set up infiniband
FWIW, my understanding of the ollama code, it uses vertical slicing first, then tensor parallelism if it needs to (terms from https://huggingface.co/docs/transformers/v4.17.0/en/parallelism)
If you can fit everything into the vertical slicing, it could in theory use fast networking links between nodes. But that would need application level support which isn't there
Now I kinda want to setup a training cluster at home to play with
But GPUs are €€€
I was going to look into trying to tune a pre-existing model
I have a prompt for controlling HA as a voice assistant that Mixtral does well but none of the smaller models handle
Just place the computers next to one another
Plug both GPUs in one
But have the other donate some power 😂
Hmm
If I got risers I could put the gpu in an enclosure and move it between systems
That might be workable
I failed setting up voice for HA
Stupid incompatibility between things
Maybe I'll just buy a compatible smart speaker...
I set up the wyoming protocol stuff with whisper/piper/openwakeword
I haven't configured openwakeword yet but the other two work well
We've done the computers with crossing cables before...
It works, but it's a huge mess
Also not sure if I'd be worried about common ground with high power tasks. It was only SATA back then
I got whisper to work on the igpu on my radeon 5700g server with ROCm and it performed worse than the cpu on average, lol.
Yea, I got stuck at trying to setup the mic array box^^
I wanted to get the software where I wanted it before I bought any hardware so I haven't tried the wake word stuff at all, just been using it from the app on my phone
I've had a spare raspi with a mic-array around
From when I first read about rasspy
I have a USB-C screen hooked up to my PC that runs a dashboard/shows security cameras, it's got speakers built in and a USB C 2.0 passthrough on the side. I was going to get a mic to plug into that.
And if that works I'll worry about satellites later
I think that failed on HA not being quite ready at the time.
And then when I recently wanted to set things up, I Realized that the software side needs 64bit OS for the features I want, but that mic-array's driver kernel panics on 64bit kernels 🤦
I run HA in a different room than the one I'd care about for voice^^
Same, I ran the cable from my server through a wall and wall-mounted the portable screen.
I still need to clean up the hole I poked for it though, I was going to 3D print a flange for it
... I have an existing ethernet link between my office and the living room.
I also still have a blue yeti in storage. I might just hook that up to the cluster node I already got running in the living room 🤔
I just think the pickup pattern isn't quite right for smart speaker applications
I have two snowballs and they sucked when I tried them with almond
Way too close of a pickup field
boom arm above my sofa 😂
I could get another of https://rode.com/en/microphones/on-camera/videomic-go-ii which I use for VC
I didn't want to have a mic in frame but still have better quality than my webcam (or god forbid, the laptop's internal)
I got a wireless modmic I'm happy with the quality from
apt-x ll so it's low bandwidth but more than good for voice
Assuming you mean voice chat by VC
But it's also visible and magnetized to the side of my headset
Heh, I have a standalone bluetooth receiver that outputs toslink that the modmic can connect to, I was a bit surprised the first time I accidentally paired them and my voice boomed out my speakers.
Video chat/conference. But same difference
you guys are streamers? what your channels?
I don't stream, I sit in meetings at work a lot though
Enough to warrant getting a setup that lets people hear me clearly
screw that.. builtin mic or they can just hear static imo
Work paid for a headset for me, I just didn't like it.
It looks pretty ridiculious, it's got all these light-up blue strips on it
least they didn't make it a requirement. I've had friends tell me they are forced to use whatever their job gives them no matter how shitty it is lol
prob some response to all that stuff you can buy to "fake" you being active and working while at home
My requirement is to have something that works, there's no specifics.
How do you fake talking to someone over a headset?
was refering to other things. but the "requirement" aspect spilling over to all parts
schools here require the kids to use thier school issued laptops etc. can't even use your own computer lol
I got banned from using school computers when I was in highschool
Idiots didn't know how to take a security report
i wouldn't be shocked if your not handd a laptop starting at grade 3 thes days
Probably
I would have been a lot more organized in school if I'd been able to do it all on a computer, ngl
Tablets during some points in 1st grade
They're in 2nd now and speed math is now on those instead of "mad minute" like we used to do
Good, honestly
Kids don't need to waste time learning antiquated methods like pencil and paper.
I run across people asking about if laptops/tablets are more eco friendly than pencil and paper when it comes to like schooling etc over the grand scale and you always see the argument about how trees are getting chopped down but they never touch on the recycling and lasting impact aspect of it lol
Putting an environmental price on it is hard
Computers take a lot to make but they allow things to be done without creating paper waste, they allow things to be done much quicker, and they allow for things pencil+paper just doesn't.
But from an organizational standpoint computers are so much easier
Misplace a paper file in an office? Fuck. Misplace a file on a PC? Run a quick search, it's found.
in a learning aspect. pencil and paper still wins out tho
I feel I went to school during the divide of computers overtaking paper, I learned cursive (was the last year to in school) but by my senior year I was expected to turn in everything electronically.
which really is kind of the point
No particular learning method works better for everyone and, personally, I don't see what pencil+paper offers over a computer
I would've rather they went over how to use evernote and stuff like that in school versus go over how to use a paper planner and organize a binder
I use note taking applications still, I haven't touched a binder in years
they have done studies that in general for most your brain soaks up more when you relay information through writing vs typing
but as you said, computers just offer more stuff they can be utilized
thus why schools do the tablets/laptops
They've got to be a godsend for teachers, too
Grading 150 assignments on a PC would be so much easier than grading them on paper and dealing with crappy handwriting
yea well the handwriting itself is often graded to a degree
not to mention an indicator for some things teachers really should look out for
Which should not be the cade
Grading handwriting should have been removed at least 10 years ago
cursive is kind of useless but normal print is critical
I don't think we have had grading of handwriting since the 70s
Heck, kids have had a dedicated tablet / laptop since 2003 here
.s/critical/also useless/
I find it hard to believe even in other countries that there is no grading of handwriting during the lower grades at all.
you pretty much learn to write the same time you learn to read
lets hope you don't lose your tongue during an apocalypse then lol
??? Why would that be related to handwriting skills?
not referring to cursive but your handwriting needs to be readable lol
thats what they grade you on.. difference between scribbles and print
least they use to. no idea what they do these days
how does that relate to the apocalypse now?
no tongue, no talk.. apocalypse usually means no typing... didn't think that was a confusing statement
And no. The only thing I've written by hand since I left education is my name. And very seldomly my address.
I should just print stickers for that...
With my meds, if I can't type I'm dead already either way 😂
the level of apocalypse that destroys all electronics would have some worse issues than spoken language for me
I still take notes on paper so im happy I can write properly
I prefer notes on paper then using something like the note app on the phone
Why take them twice if you can just start with something better?
I effectively need to be able to search my notes.
I could try to write in a way OCR can cope with. But I'm faster with a keyboard than doing that by hand
as long as it's readable is all that matters imo.. only office type jobs tend to push for better
My notes involve a lot of math and drawings. I tried doing that on a tablet, couldnt get used to it
I take a lot of notes
On my remarkable
Best purchase last 12 months
Great for presentations too
My colleague also loves his. But I'm unsure if I want one. Essentially never use my wacom which would allow many similar features while I'm at my desk
I mostly got the first one I have to write the support in sway^^
second one was intended as toy. And for some types of notes. But ehh
I use the remarkable a lot in the field while walking around customer sites
So the Wacom isn't comparable to me
^^ I tend to just do that on my phone. I don't usually take lots of notes outside meetings AND away from my desk
Fun thing about the remarkable during presentation is doing the drawings before and then just use the colour marking highlight during the presentation
I wish
Thankfully we also have a lot of generations that don't know what the sumerian empire or Reaganomics was like
Ironically we had overhead projectors and smart boards in the school at the same time
Early 00s
We also had "mother tongue" teachers in school over broadband in the late 90s
Oh no technology
smart boards (backlit and projector driven), simple projectors and overheads all at the same time^^
Yeah
We had ones that had touch screen displays at some point
But those just got replaced by tablets pretty quickly
wait, no that thing wasn't backlit. It was a normal blackboard (the white variant, not green with chalk) and a cam that then tried to be smart 😂
yea, that one sucked.
The projector based ones were decent.
Though at that time they worked similar to a wiimote. Which allowed to do some silly things
I just managed schools at the time, but dang it was a lot of stuff at once
I rolled out wireless on most of our 55 schools in 2005
That was a nightmare with reflection on metal lockers and old brick schools
Was also fun trying to maintain windows update over 128k lines to the island schools in 2006
All inner school had 10+Mbit
And I had 50 at that point, so 128k gave me proper 1998 vibes
Or US 2021 vibes

and now you are just spoiled with speed
and in another 50 years people will be reminiscing about how slow 10g was
We already do
It is 2010 speeds after all
I haven't rolled out 10G in over a decade
Also why you can buy 48port 10G for $50
which one? 🤔
Nexus 5k
I just ordered one
40G is also reasonably cheap
Or 56G if you run ipoib
that's an empty SFP chassis? Still quite nice
Even just the first google hit is 180€
That's 32port, right?
^^ I could drop some at work, but they'd get mad at some point
I dont own anything that can even do 100g
Is that just ebay? 🤔
Yes
that reminds me, I wanted to check storage if we have a network card left over at work D:
wut?
you could run some nvme over fiber at those speeds. but unless you have SAN, I don't think there's a reason to
I dont have any usecases for that
I'd have to split my compute and storage
Then it might make sense to run VMs off of something like
Instead of iscsi
Iscsi is fine for most cases
As long as you have a proper sync storage
Optane in front, rust in the back is great
For anything at home 1g is fine. 10g would be better but only my pc can do that
1G 
I dunno, once you introduce video and various smart home stuff. 1G isn't enough.
I do think people swapping out to 2.5G and not just skipping to to the next level is dumb
those dont tend to use that much bandwidth either unless you actually move files through the net
I doubt that
wish that were true. but i can see the bottle necking in my router from time to time lol
course I also pipe mjpeg camera feeds around
must be very high bitrates
just alot of seperate feeds
the only reason why i have and am still upgrading to 10gb is that backups will take too long to transfer around otherwise
i sure do not like my wifi being slow because of that :^)
kinda weird that multi gb isnt the standard for most switch things yet
cost vs gains, just not there yet
pretty much any device comes with a 2.5g port
just not them switches, its similarly expensive to buying 10gig gear
vast majority of consumers dont even have a switch
the era of the computer nerds
screw wifi
less
checked pricewatch, routers with 10G ports start at about 500 EUR... yeah good luck lol
video feeds should really be on separate networks either way. As much as possible
if they can afford the latest iphone they can afford a 500 buck router ;D
especially as the 10g ports are like 20g on m1 
just get L3 capable switches and do some SDN 🙂
ubiquity, the apple of networking
if you let a third party repair it drops back to 1G/s
Even my 4k hdr pure rip lotr is only 96MBps
So should play fine with a bit of buffering over gigabit
So for normies who use streaming services they probably peak out at 15MBps
In absolute worst possible scenario
With pure 4khdr
you sure that's MB/s and not Mb/s?
Yes
Regular 4k movie is about 50Mb/s, or even lower
for SDR yeah
Can I have your opinion on https://www.kickstarter.com/projects/urgreen/ugreen-nasync-next-level-storage-limitless-possibilities
https://www.amazon.com/NETGEAR-Nighthawk-Quad-Stream-Quad-core-Compatible/dp/B01M12RE4A this router was maybe $500 new
The Nighthawk x10 AD7200 Smart Wi-Fi router is industry's fastest router for media streaming with Plax media server. The x10 delivers 802.11AC/ad Wi-Fi for ultra smooth 4K streaming, VR gaming and instant downloads. With a powerful 1.7 GHz Quad core processor and quad stream architecture, you can...
hmm, 600$ for the 6bay 🤔
has only one 10G port
Sth recently reviewed cheap 2.5/10
Unless you need >1Gb on a single link I'd personally go with multiple AC/AX routers spread around for coverage
oh, there's an all flash.
I wish they said something about the OS and with lack thereof, hackability
overpriced. plus ugreen has no buisness being on a kickstarter
every dedicated nas costs more than building a cheap-ass PC in a case and chucking drives in it
Even a PC case with hotswap bays is pretty cheap
its for a friend who is probably not completely techy enough to do an own NAS project
this didn't specify multiple 10G ports tho
qnap enters the chat
What is the difference between the 4-bay and the 4-bay Plus? 🤔
my qnap was like $200, $350 with drives
Is it one of the USB-C enclosures?
Their NAS's don't go that cheap, at least not now.
Except for their multi-drive USB-C enclosures
the cheapest 2 bay is $100+ more than a mini pc with same specs and you only gain the two bay hot swap out of it
https://www.amazon.com/QNAP-TS-433-4G-US-Quad-core-Processor-Diskless/dp/B0B8M5WQ1Y this is the newer one to the one I have
The TS-433 is small in size and remains high-performance, cool, and quiet with an effective cooling fan. Powered by a ARM® Cortex-A55 quad-core 2 GHz processor, built-in 4 GB RAM, one Gigabit LAN port, one 2.5 Gigabit LAN port, and supports four SATA 6Gb/s hard drives, the TS-433 provides dependa...
but power usage is gonna be lower on the nas than on the PC
i disagree
mini pc is like less than 13w
*same specs
CPU and with that the ports
non-plus has 2x2.5GB, Plus has 1x2.5, 1x10GB
you gonna use a mini PC to build a DIY nas tho? how many drives you fitting in that thing
I priced out a DIY NAS a while ago that's about $80 more than that and gets a real CPU
https://www.amazon.com/AMD-4600G-12-Thread-Unlocked-Processor/dp/B09Z5NH6SJ - $115.72
https://www.amazon.com/Gigabyte-A520I-AC-Interfaces-Motherboard/dp/B08F7BHDLY/ - $99
https://www.amazon.com/KCMconmey-Internal-Compatible-Backplane-Enclosure/dp/B09WQC44B3/ - $79.99
https://www.amazon.com/TEAMGROUP-T-Force-3200MHz-PC4-25600-Desktop/dp/B08PJNVWNZ/ - $35.99
https://www.amazon.com/SilverStone-Technology-FX350-G-Connector-SST-FX350-G/dp/B0B2ZRW27R - $96.99
https://www.amazon.com/gp/product/B0CMPZQS81/ for four internal drives - $35.99
Total - $463.68
CPU Ryzen 4600G
Memory 16GB DDR4
Hotswap Drive bays 4
Internal Drive Bays 1 2.5"
M.2 slots 1 internal
I mean considering mani have built in m2 and a slot for ssd then you have like 4 usb3 ports.. infact some even have the thunderbolt ports
Though idk if the cooler will fit, might need a shorter one
so in other words multiple points of failure, no/very limited internal drives, and a mess of cabling? no thanks
how are those points of failure?
well find your rocker and get back on it
I have had 2 external drives going for over 5+ years.. not worried one bit
I see the enclosure as a point of failure, the cable as a point of failure (as its external), and the USB port/bus it's connected to as well
your saying usb is a failure over a traditional sata connection?
no I'm saying because it's outside of the case, that it's more likely to get bumped or pulled
do you understand points of failure Sonoma?
stop letting your cat play with your shit then?
then how does it get bumped?
then my cat can't play with it, or anyone else
If you can't keep your drives from getting knocked around that's on you, tbh
I guess you never go in your cabinet to add a cable or anything then
I don't flail around wildly while doing so
compared to yours, yes Sonoma, I am a large man
my hands are slightly smaller than Shaq's
hey if I had small hand it would look very very odd on my 6'5" body
Sonoma not everyone has hobbit hands, I'm not from middle earth you know
i would be more worried about your hands being too big lol
Externals are fine, just don't stand them up, lay them down on the wide side.
I use externals, but I don't use them as my primary data storage lmfao
Only external I have anymore is on my Wii
I have one connected to my nas for nightly backups of my home dir
I Don't backup my media at all
how did you manage to not bump it or pull it out by now?
I made a braket for it
surley you must have yanked that damn thing dozens of times
pretty sure that braket cost a hell of a lot less than some devices with a hot swap bay lol
it cost me nothing because I fabricated it myself from scrap
there you go
took me 20 mins and the dremel I already have
nope
just because I use one external drive for backups that gets unplugged when they are done... doesn't mean I put all my shit on externals and hope for the best
lol hope for the best
I'd be the most sketched out over the USB part of it personally
Had too many failed controllers and data corruption issues over the years
exhibit a
the usb port/bus.....
hmmmmm
and exactly what did you think I based my choices from? the back of a cracker jack box?
😛
never seen that. ever
well middle earth is not the world
bad luck is just that bad luck
hey there's a reason we have MTBF
it's averages and percentages
but that doesn't mean because you've had good luck so far that you won't have bad
Oh, I bet the issue here is Sonoma never transfers anything bigger than a JPEG.
So he never sees weird USB issues
but it seems when it does bite you, it will be glorious Sonoma
if a plex server counts as a jpeg.. sure lol
watching 1960's shows in 360p don't count
Plex is definitely closer to a jpeg than an actual intense random workload.
ok nasa engineer
ok Mr "write once read many is the same as actively using a drive"
I store other stuff too
I bet Sonoma doesn't network boot from his usb drives lmfao
because I can network boot from my NAS
😄
but it's not a work server so not gonna make it out to be more than what it really is lol
well my NAS does everything I need it to
and burns like 12W of power
so you do your sd card media server, and I'll keep what I have
you do that
and I'm getting a new NAS this week too
trash finally make it's way to washing up on shore huh
you mad bro?
He's always mad
the mega yachts don't come here to navigate thru trash lmfao
just congratulating you on your beach find
yeah cuz I'd totally use a NAS I dug out of the breakwater covered in sand right?
mega yachts don't come there to buy computer equipment either lol
I'm shocked you didn't think I used quantum coconuts
I would, getting that thing fixed up and running would be some serious nerd cred
i didn't think you were that advanced yet lol
oh? you're aware of all the tech businesses based here that outfit them?
or i might have
or did you know we have the largest dry dock in this hemisphere to repair large yachts and ships?
is there 1 listing in your yellowpages? oh sorry.. meant town billboard directory?
The guy from KC talking about boats is 🤣 🤣 🤣
