Folks in MEGACORP cloud env can spend > 5 digits a month and not get noticed.

Can they spend 7 digits?

Yes, but it would get noticed.

This has everything to do with OA using digital toxic waste shoveled into a GPU furnace and patched with the suffering of RLHF. Everything is in there.

It has downloaded every public key from github?

I would use this trick to add them to my test clusters.

This is awesome!

https://en.wikipedia.org/wiki/Multiseat_configuration

I am super curious about your setup. I played with MS years ago, but I lost the need. It is a super cool tech that I'd love to see its efficiencies embraced in some way.

Install an old GPU, Connected a monitor to the extra GPU, connect mouse and keyboard, Use the loginctl command to list available devices/usb ports and attach them to a seat.

I suggest using Arch linux although loginctl should be available in all distributions using SystemD now.

If you don't have enough USB ports you can use a USB hub, some monitors comes with USB hub. And some with built in sound, or you can use wireless headset.

My main issue was that driver support was dropped for my oldest GPU card. So one day when I upgraded the OS it just stopped working. So to be on the safe side get another GPU like the one you already have.

So for a GPU with say 4 display ports, I can run 4 sessions off the same GPU provided I can get enough USB ports.

Modern GPUs aren't constrained by the number of ports. You can use DisplayPort MST (hubs) to output multiple displays off a single port.

But most GPUs are limited to being able to run 4 displays. Except some AMD Eyefinity can do 6

It might be possible one way or another, although I used separate gfx cards. You might find a different x server that lets you do it using a single card. I suggest getting an extra hard drive, install grub so you can dual boot, install the experimental OS on the extra hard drive and start fiddling. I also dual booted different OS in order to experiment with machine learning when the kids was sleeping, as I got multiple gpus, although the old ones where crap it was a good learning exercise.

This is also a reason why I have 4 separate stations now, if I have to upgrade hardware only one station is down at a time. And while you can get a 3 GPU system at budget, a 4 GPU system will get expensive, at least last time I checked. It would be interesting to look into using old used ai servers for multi seat purposes.

The kids will get over it lol

Throw some big kites on it and sail it, use the jet propulsion just for vector control.

but people who take ferry rides want to know roughly when they'll reach the other side?

Just to get the ferry from Tasmania to South America.

Pressed CDs do not fail in this way.

I got into hydroponics first because I wanted to research closed loop food production.

A JSON parser is less than a thousand lines of code.

And where most of CPU time will be wasted in, if you care about profiling/improving responsiveness.

At that point you're just io bound, no? I can easily parse json at 100+GB/s on commodity hardware, but I'm gonna have a much harder time actually delivering that much data to parse.

What's a better way?

You know those puff pieces are to assuage our fears of job loss and provide an folksy aww-sucks patina to the plucky LLM who can't run a vending machine.

Or maybe that LLMs just aren't that good.

Dr Bronner's made it into the training data.