A discussion surrounding RAM price hikes and repairability

Discussion
41

No… :smiley: - it’s already partly open - I allow SSH in (on a non-standard port) to a specific host… Even so - I get port scanned and as I’m running fail2ban - I often see lots of IP addresses trying to SSH using the non-standard port (and getting jailed / banned)…

I tried another service - and - that service kept crashing daily with “too many login attempts” so I turned that off… I looked into trying to get fail2ban to also monitor that - but it looked too hard so I gave up…

How would I even tell a legit connection from an It’s FOSS member IP address?

That’s the sole purpose of it - for me - and my family - i.e. missus and two daughters at home…

I used to watch that content directly from my PC - e.g. browse the NFS shares with Nautilus and watch in mpv on Linux… But JellyFin keeps track of where I was up to and what I was watching… So I can move from device to device with continuity…

3 Likes
42

Oh, I was just pulling you leg (teasing) you about granting access.
But to answer your question. I thought I have a static IP address thru my ISP. (?)

3 Likes
43

I also have a Static IP from my ISP - I don’t really need it - but it was the only way I could have port forwarding on my router… i.e. the ISP changed something at their end and my port forwarding stopped working (with Dynamic DNS through NoIP)…

Not that it matters - my router doesn’t support restriction to source IP / network… All the NAT rules are basically any/any…

Correction - not “any/any”

0.0.0.0/0 and specific TCP/UDP ports

2 Likes
44

You could experiment with Tailscale. You can have up to 100 devices and 3 users for free. Using the client on all my computers they can all talk to each other if I’m at home, at work, at Burger King, even my phone has the client installed.

2 Likes
45

I looked. Their website is mostly hype, but on github there is some information

2 Likes
46

Not sure what use I would have for that…

There are several things I have that implement UPNP stuff on my router (universal plug and play)…

ResiioSync is one - i.e. I didn’t do anything on my router - but the rules are there… So my Sync works when I’m out and about… Not just at home on my LAN / WiFi…

And my IP cameras - somehow “work” from outside - i.e. with the app on my Android phone - when I’m kilometres away - I can see both feeds… I don’t know how that works even… I refuse to pay for the vendor’s “cloud solution” - however the IP cameras are the same brand as my router…
– edit –
I guess I could implement a VPN… My router supports OpenVPN… That would probably be easier than trying to use a 3rd party solution (like TailScale)… I’ve used OpenVPN in the past - I’ve supported a "Pro’’ platform running OpenVPN (Open VPN Access Server - a licensed product) for one of my customers… So I know OpenVPN already…
– edit 2 –
NBN (National Broadband Network) have “promised” me fibre this year (not holding my breath) - if that does ever happen - I’ll probably setup OpenVPN so I can use my JellyFin from outside my LAN / WiFi - and - will probably share with some family members over east - but - NOT without a VPN… Probably mostly my brother in Melbourne… it would be a major head-f–k trying to explain to my pushing 90 mother (born in 1937 - just had a pacemaker fitted) - how to VPN - she’s actually pretty good, and mostly on-line - but she has trouble figuring out the difference between an SMS and an e-mail - and that some links she shares are behind paywalls she’s subscribed to…

3 Likes
47

Just tried my ffmpeg scripts using hwaccel “Intel QSV” - wow! This is on my “new” (2nd hand) Dell Optiplex i5 (gen10) with an Intel Arc A310 GPU running Debian 13…

Probably 10x faster than using my AMD CPU on my main desktop… But CPU rendering was quicker on the AMD (Ryzen 7 3700x) than the i5 (gen 10)…

When rendering with CPU - using the “ffpb” CLI front end for ffmpeg - the progress bar doesn’t seem to move… using Intel QSV - I can see it move along in real time!

3 Likes
48

I am interested in GPU’s for my statistical models work.
Would an Ark be any better than my RX6400 amd card? … they both have only 4Gb of ram
I need to multiply large matrices which is a similar compute intensive job to your ffmpeg conversions.

1 Like
49

No idea dude - can you GPU / hwaccel your jobs?

There’s quite a few compute loads that can be offloaded to GPU - mostly using the CUDA API - which I believe is NVidia only…

I just like the idea of buying a bargain basement GPU - and it can 10x speed up compute jobs by offloading… BTW - IMHO (from what I’ve read) the Intel Arc GPU chipset is an order of magnitude more power efficient than NVidia or AMD…

I’ve only used CUDA a few times - I ran “Folding at Home” during Covid (i.e. mostly 2020 - but also into 2021) on two Linux boxes I had with NVidia GPUs …

That was so long ago (even if it seems like yesterday) - I don’t remember what I had to do to get CUDA working with “Folding@Home”… it wasn’t straightforward on Linux (apparently much easier on Windows)…

CUDA is the reason for RAM and GPU shortages worldwide - AI likes to offload thinking to GPU server farms…

Me? I think it’s a VAST waste of resources and power… Just like crypto-currency… We should be figuring out ways to save power - not waste it on frivolities… Wouldn’t surprise me in the least if morons have asked AI about how to save power not appreciating the irony or the oxymoron nature of their enquiry…

1 Like
50

Think the only way to measure would be clock speed of the two. Memory plays and important part but the motherboard speed also effects these things great processor slowed by other bottle knecks. Plus disk access for next data set.

I used to do my big data analytics over lunch when systems were quiet.

Set process running. Go to gym or pool have lunch return to see if had finished.

But some of the girls who I worked with complained to the boss about me disappearing for a few hours. So stopped doing that. Had lunch then set job running and watched everything slowdown for the afternoon.

2 Likes
51

Clock speed of a GPU makes no difference sometimes - it all depends whether or not the vendor has released an API that developers can use… and even then - if the developers have actually used those APIs anyway…

Strangely - AMD have nearly always been more open than Intel or NVidia… but - Intel drivers and API’s have always been pretty good…

From the little bit I’ve read - ffmpeg works way better with NVidia and Intel GPU - than it does with AMD - which is a shame…

As for Clock Speed itself - I’ve already mentioned - my 2021 AMD Ryzen 7 3700x is considerably faster at doing CPU transcoding in ffmpeg than the intel i5 (gen10 - I don’t know what year - just googled it - 2020)… I bought the Ryzen in very early 2021 - so it’s probably also 2020 era…

When’s our resident Carpathian video transcoding guru @kovacslt going to weigh in again - his opinion is always valuable (and sorry if I got the wrong Central European mountain range - is it Transylvania not Carpathia :smiley: ?)…

3 Likes
52

That is right… CUDA is for nvidia only
There is an AMD equivalent … OpenCL… which I believe also works for Intel GPU’s
There us an R package gpuR which defines gpu matrix classes, so a matrix object can stay in gpu-ram… but 4Gb is not big enough for large matrices.
I will give it a try someday, but I think I really need a gpu with more ram. I think the faster clock speed of the gpu would be negated by having to shuffle matrix elements between ram and gpu ram

2 Likes
53

We had one of those ‘bean counter’ types. I used to come late/stay late to run overnight jobs … that was frowned on.

2 Likes
54

As I was quite busy (still I am), did not have time to keep up with the forum.

But I was mentioned and that triggered an email notification which I can’t whithstand. :smiley:

You got it almost right, however that rises nowadays a (geo?)political issue, as it makes remember the shape of Hungary in the ancient times and before 1920…

My area is called today simply Hungary, and this area does not match the Carpathian Basin as it was the case before the Trianon peace treaty… It’s not just Transylvania, but also west part of Ukraine, south part of Slovakia, and so on around all the borders.

I’m not hurt anyway, especially reading your kind words

Thank you, I really feel honored.
But I’m definitely not a guru, I just use the trancodings on a daily basis.
There are some input formats I get from clients recorded with their phones maybe: Davinci refuses to work with those, mostly totally, sometimes only the audio track is missing.
So I need to convert…
My camreas record in a format today compatible with the studio version, but I used the free version for years, and I had to convert them too…
When I render out a timeline, it ends up in a prores or similar “high-format” file.
So I need to convert it to .h264 or rarely .h265.

I tried nvenc on nVidia, my laptop has GTX1050Ti, my desktop had GTX1060, later GTX1070, the speed was similar, around 170 Fps. With VAAPI it was slower.
Using the intel GPU with VAAPI was slow again, but I don’t remember the number.
Using the intel GPU via Quicksync is unbeatable, it’s always above 250 Fps, and sometimes near 400 Fps, I guess this depends on input file format as well.
This is with 8th gen CPU and for h624 at fullHD resolution.
So I stick to QSV for transcoding (ffmpeg), nvenc for encoding on laptop from OBS (very easy to set up, and fast enough to efficiently decrease CPU load during a broadcast).

Nvidia supports OpenCL as well.
Yes, intel has OpenCL support too.

There are various implementations though, so different apps work with different OpenCL versions. That’s for me a pain, as Davinci Resolve is quite picky, it doesn’t lik intel’s OpenCL currently available in Linux, and requires the version from AMD in the “pro” drivers.
The Mesa OpenCL does not work :frowning:

3 Likes
56

I was always taught, then taught my students that speed of a computer dépendent on

Memory available (more the better)

Clock speed of the motherboard

Clockspeed of the processor

Memory availability on processor

Then things like disk access to get data

Graphics cards to display things like games or image manipulation again Memory and speed

Failing any of these cause bottleknecks and slowed the system down.

Perhaps this was more the case in my teaching times as we had expanded and extended memory issues moving from DOS and 286, 386, 486 processors but with pentium and other that followed this has changed ?

I am not up to date with processors any more and usually avoid answering questions of one computer against another especially when asked about specs, tend to look at price first, memory second, upgrade availability. But also now most of my clients are like me retired soon not as urgent only for web surfing not calculations or limited photo manipulation.

Perhaps another discussion point ?

1 Like
57

Can I add software to your list

2 Likes
58

I was avoiding that, similar with operating system.

As we get into games vs excel type conversations or linux vs windows and drivers, graphics cards.

Was thinking general off the Shelf systems. But not dell vs lenovo vs samsung etc.

1 Like
59

An excellent break down of how fast a PC can process information.

By clock speed of the motherboard, I hear that called the Bus or Bus speed. So yes,
you can have a fast CPU and fast memory, but the connector between the two known as the
Bus can slow the overall speed of your PC.

A very good tool to see some of the information of your PC that Paul described can be seen with CPU-Z (Windows) and CPU-X for Linux.

2 Likes
60

BIZARRO!

Just relocated the JellyFin server to another location in my office and disconnected HDMI and set LightDM to stopped and disabled - i.e. I want to use it as a headless server…

But - just tried to use ffmpeg - and it won’t use QSV or hwaccel or /dev/dri/renderD29 while there’s no GUI running! WTF?

So - to use hwaccel I have to at least be running lightdm… started lightdm and ffmpeg / ffpb working with QSV again…

Oh well - lightdm is fairly “light” (pun intended?) so I’ll just leave it running anyway…

CR@P!

Not just lightdm - I need to have autologin enabled in /etc/lightdm/lightdm.conf - and it has to load my MATE desktop in order to enable the discrete GPU… This is kinda cr@p…

I’ll be looking into ways of bypassing this… Even though MATE isn’t has heavy as Gnome 4 - I want this to be a headless server - NOT a desktop machine! The discrete GPU is for processing - not displaying!

1 Like
61

That is nonsense… there is no way a gpu has to have a whole DE running , just to function.
IceWM, or JWM, or Fluxbox are lighter than Mate.

1 Like