theYipster

Two weeks ago I purchased an M5 Max MacBook Pro 16 inch with 128GB RAM and a 4TB SSD from Microcenter for $5100. (They had a $900 discount on the machine.) Not sure if that deal is still around, or if Microcenter still has any stock, but if you're in the market, I'd make a run for it! $5100 is now $8000 on Apple (and if ordered via Apple, it won't ship until August.)

You don't need all of the model in VRAM. 1 or 2 RTX Pro 6000s will do. $50K will get you there very nicely, and on a 1600 watt PSU if you go for the MAX-Q versions. (The same wattage PSU I'm typing this on, and have been using over the last 5 years.)

Mine, for one. M5 Max MacBook Pro 128GB with a 4TB SSD. $5100 after a $1000 discount at Microcenter. Great deal if you can find it in stock.

Not the new ones. Only the M1 and M2 have good support for Asahi. But you really don't need it. If you need Linux, use a VM (UTM is free and is equivalent to KVM/QEMU in speed, despite being a Type-2 Hypervisor.)

All for $670 :)

In all seriousness, I would probably throw $10 at a project to design and implement a modern turbofan FADEC + all of the certification artifacts.

I'm amazed at how many folks on this forum see the Web UI as intrinsically tied to the service. As someone else rightly said, e-mail = IMAP + SMTP. That is true of Gmail as well.

Frankly, I've always hated the Gmail web UI, so I never use it. Not in the 22 years I've had a Gmail account.

IMHO, Superhuman gets a ton right... A Superhuman clone (maybe in VIM or Emacs) would be ideal if you don't want the AI features or the $40/month fee. Don't even need to change your mail address, since it connects to Gmail.

Very nice. Primary use case is application development, where the applications leverage a mixture of cloud based and local models. Modelling complex architectures. My work is primarily in the aerospace and defense arena, so hybrid and on-prem are important, as are ITAR and CMMC compliance. The idea is to have the local rig to build and validate architectural deployments that can sit on prem on customer hardware, in cloud, in gov cloud, or in a mix.

Not really looking at colocation, as this machine would double as a heavy duty gaming and flight sim rig. That means at least one regular RTX 6000 Pro. Not sure if I can mix and match with the Max-Q version, or if I even want blower fans in a desktop case (last time I did that was about 16-18 years ago with an ATI card... wasn't a fan--pun intended.)

That's about what my OC'd and watercooled 4090 runs at. The cards are designed for it. Only problem I have is when sitting next to the computer under load -- I either have to open windows or blast the AC. Too bad I don't live in a cold climate -- that 60c heat output would come in handy :)

Great article. I'm about to embark on a similar journey.... Doing a ton of AI development right now. Don't need a server, but a very, very high end workstation is super appealing to me right now. Looking at $50-$80k. 1TB RAM. 2x RTX Pro 6000s. 64 core Threadripper Pro. As many 4tb or 8tb nvme drives as I can stuff.

I envision NixOS at the core... then everything I need virtualized on top with KVM/QEMU. Maybe a dual boot setup with Windows for gaming and Flight Simulator (but I could virtualize that too with easy GPU passthrough.)

Lingering questions I'm working to figure out:

- Will 2 RTX Pro 6000s run on a 1600 watt PSU? Not sure how much higher I can go without calling an electrician. (standard US home.)

- Assuming I plop this into my home office, should I expect the PC to run significantly hotter than my current rig? (3960x threadripper, 128GB RAM, 1600watt psu, overclocked and watercooled 4090.) My water temp, measured at radiator, is about 60c at peak load. (This is the only number I care about, as this is what I have to consider to be comfortable sitting next to it.)

This is awesome.

Agree 100%.

In the early days (before Claude Code mastered Rust,) I would get into this annoying pattern where Claude used different names for variables between tests and implementation, get confused, and then more times than not, would change the implementation to match the test (which was not written first--was not doing TDD and thus not the behavior I wanted.)

Static languages prevent that. I've had great success with Claude writing Rust, and I think it's an excellent language for LLMs not just for low level work, but for production-grade code of all types (I see rust as better aligned to compete with C++, Java, and C#.)

I've also had great success with Claude writing C#. Using Claude, I've built C#/.Net in Linux, deployed in Windows (via Visual Studio) with Claude Code running in WSL, and it's been a great experience all around.

Yep... Did this 3 years ago in my home and came to the same conclusion. If I were to do it again, I'd run fiber through the walls instead of Cat-6a. It took forever to find SFP+ modules that would work with my Unifi setup... (not wanting to pay for more than one router or switch with native 10g RJ-45 ports, which are still very expensive.) I loose POE but, on the whole, it would've been much easier and much less costly--I think--to have just run fiber.

I remember visiting GitHub's downtown SF HQ sometime around 2014 or so... it was soon after they closed their first significant funding round, and years before they were purchased by MS. I had a friend who worked there as a very early employee. I was at IBM at the time doing AI stuff.

I remember saying to myself, "every single meeting room and common area in this building is designed around the consumption of alcohol--the long bar downstairs, the meeting room modeled after an airport lounge, the meeting room modeled after a smoking club, the meeting room / roof deck...

A year or two later they had that public "me-too" snafu (years before me-too) that led to a founder's resignation, a whole bunch of other people leaving, and then Microsoft acquiring the company. I wondered back then, is this the end of the company?

Perhaps so, but perhaps not... Here we are, 8 years the acquisition, only now lamenting a slow demise. That's a nice run for a startup acquired by a behemoth enterprise software company. With the exception of Redhat (which is debatable,) IBM had no ability to keep a software acquisition's culture, verve, or ability alive past a year or two.

It was the best of times. It was the blurst of times.

Also of Xenosaga, and for the same reasons. Although, that's perhaps a more obscure reference (and is a PS2, rather than PC game series.)

I'm looking to replace my worn-out Sony XM-3s and am wondering if the AirPods Max 2 could be a better option than the XM-6s. I've always liked Sony's audio quality and ANC, but felt the headphones were not as comfortable to wear while sleeping on airplanes than older Bose models I had prior (like the QC-2s and QC-3s.)

Can anyone with experience with either the AirPods Max 1s or XM-6s tell me what they feel like to sleep with on an airplane (business class with a lie flat bed?) Plane travel is my primary use-case for these type of headphones.

There really is no going back when you go split and tented. I've was on a Kenesis Freestyle for years, then upgraded to a Dygma Raise (V1) about three years ago. It's served me well, but I've been eyeing a CyBoard Imprint, which is like a Dactyl or Charbydis, but has hot swap-able switches (a rarity for curved key-well boards.) Can't give up my holy panda switches. :)

Just because you’re a good programmer / software engineer doesn’t mean you’re a good architect, or a good UI designer, or a good product manager. Yet in my experience, using LLMs to successfully produce software really works those architect, designer, and manager muscles, and thus requires them to be strong.

With Claude, I built a TUI for managing QEMU/KVM VMs (Rust with ratatui.) Solves a lot of problems I had with virt-manager, so I made it a FOSSS project. https://www.vm-curator.org.

Superb!

Assuming you are using the flagship copilot that is a $30 / mo add on to a 365 subscription, and maybe, maybe if Microsoft replaced CoPilot’s “brain” with Opus 4.5. In my experience, while flagship CoPilot does deliver value if setup correctly, it’s no where near as capable an “agent” as Claude. (And even though Open Claw is now model agnostic, there is a reason for its association to Claude. Despite it’s expense, I find Opus 4.5 works best.)

Actually I just installed it over the weekend in a QEMU/KVM VM on my modern desktop. Took about 20 minutes, and works very nicely. You can give it up to 4GB of RAM. Now, I can't use it for much (aside from playing the welcome tour,) because the last thing I want to do is expose it to the internet... On the other hand, I hear there is a modern browser that is maintained for it.

I love the care and polish that went into the Microsoft Windows XP Welcome/Tour app that played after install. That was the peak of the summit—quality wise.

I’ve been trying out both DMS and Noctalia in separate VMs this week (both on Niri.) I like them both. Noctalia seems a bit more refined out of the box. DMS is more customizable. I foresee both taking over from .dotfile packs (and maybe even Omarchy) as better ways to bootstrap a Nir or Hyprland.

But that's the point. Choice and customization. It's the natural result of FOSS and the as-designed modularity of the Linux ecosystem.

Exploring popular options and finding what works for you is easier than it has ever been, and fun too. The difference between Linux today and the Linux of old is that for most setups, all the pieces you choose can fit together nicely and "just work." Despite all the different flavors and variations and distributions and desktop environments and window managers and the like, pretty much every popular distro uses a recent or near recent version of the monolithic Linux kernel + system-d, so all the important stuff is more or less the same (with tweaks here or there.)

Btw, this feature is now available in v0.2.x! vm-curator supports single-gpu passthrough (tested locally,) and multi-gpu pass-through via looking-glass (experimental: needs testing.)

single-gpu-passthrough relies on a script (run outside the app) to disconnect the GPU from the current X.org or Wayland session and then to attach it to the running VM. When the VM is shut down, the script runs this process in reverse. This means you can only run one VM at a time with your main display and peripherals, and while you're running that VM, you can't access your host with your display and peripherals (you can always SSH into it while the VM is running.)

This is the common process for getting single-GPU-passthrough to work. vm-curator helps prepare the system and generates the scripts automatically.

multi-gpu-passthrough is designed to run with looking-glass, but it can also support physical KVM switching if the user prefers.

It's really a (good IMHO) sign of the times that us old hats have to remind ourselves that most new comers to Linux today aren't necessarily adept at installing another OS, let alone using the command line. The first time I installed Arch was maybe four years ago, but the very first dual boot setup I made was between Win 3.1 and OS/2 2.1 in 1993 when I was 10, and I've been playing with Linux since the mid-late 90s. When I first installed Arch the "hard way" I said to myself--"I don't understand why it has this reputation... this is all stuff I've done before countless times." Frankly, I'm still trying to figure out the distribution graph of Linux knowledge and how to engage with different skill levels.

Hi! Yes, it absolutely will. The only caveat with single-gpu-passthrough is that when you are passing your gpu to the VM, you can only use the VM with your main display and peripherals. (The host is still accessible via SSH.) When you shut down your VM, the host returns automatically.

In the "i can only sort of use one OS at a time" sense, it is like dual booting, but the transition is much faster (since there is no reboot required) and, as mentioned, your host doesn't really go away (you can still SSH into it from another PC on your network.)

vm-curator, a FOSS Rust TUI for Linux QEMU-KVM VMs, now supports full GPU pass-through in its v0.2.0 release for near-native GPU performance suitable for high end gaming and demanding GPU-centric workflows. This works for both NVIDIA and AMD cards (and should work with Intel as well,) and it can create and run both multi-GPU and single-GPU setups!

Testers and feedback are wanted very much: I only have one hardware configuration I can test on, and I'd be grateful for anyone to try it and report their experience--especially for multi-gpu pass-through (as I can only test single at this time.)

Update: Released first bug-fixes (v 0.1.1). Notable fixes include:

- Defaulting to SGL display-mode when enabling 3D accel for Linux VMs for better graphics performance. - Changing display mode for existing VMs is now available in Manage - Adding multiple VMs for the same OS is now supported - Custom VM naming now persists. - Renaming VMs is now supported.

Next major feature: Full PCI passthrough support, including GPU passthrough via looking-glass.