Since nobody seemed to actually answer your question: the answer is that ram is actually really simple electrically. Modern DDR5 is very difficult and expensive to manufacture at scale, but is very simple to design.
If someone were to try and poison a memory package, it would be massively obvious by virtue of the package being larger, being very electrically noisy, or by sucking an order of magnitude more power to function.
DRAM sockets are not generic pci busses, and cannot be used on a typical motherboard to load arbitrary hardware the way USB or PCIE can.
Also, the way ram works, you really couldn't do much more than read contents and relay via an on-dir radio, which would have to be super short range. Even something as "simple" as Bluetooth or wifi would be too big, too slow and take too much power to still function as a memory die.
You should be way more scared of cloud services, appliances, or iot devices than a stick of DDR5.
Tl/DR: it'd be prohibitively expensive and itd have nowhere to go, if it could even work at all without Corsair noticing.
Basic output? probably. Gaming? Probably not. Again same problem of some cards missing hardware components that are required for a rendering pipeline, regardless of physical or virtual outputs
The 6000 Blackwell has all of this and would totally work, the H200 does not, etc
You could definitely do that for rasterization, but if any part of the graphics API required something the card didn't have (like ray tracing) you couldn't synchronize the work between it and another RT capable card fast enough.
See also: why they killed crossfire and sli in the gaming space, and why the derivative works of that technology only exists in the data center for non latency-intensive workloads.
Common misconception, display head =/= display socket.
It's not just missing the physical plug, it's missing the entire logic circuit dedicated to rendering and encoding a display output. Even if you wired an hdmi or dp to the board, it would have nothing to drive and time the signal.
That circuitry isn't a separate chip it's physically part of the GPU die architecture and you can't add it on later via a mod.
AI GPUs either just use the same standard driver packages that gaming GPUs use, or they use slightly cut down ones because they physically lack internal hardware like display heads or rt cores. With those it's not a matter of drivers, it's a physical lack of hardware that a game engine or display would expect.
If you had an rtx 6000 Blackwell in your PC right now you'd just install the same drivers as the 5090, 5050, or 3060 (its the same driver across all three of these)
People shit on it but there's a lot of good open-source tooling that supports it.
There are nist l1 profiles
Tutorials and guides for everything
etc
Part of being a good sysadmin is knowing when not to reinvent the wheel. Ubuntu has a lot of options for vetted, hardened, "other people's wheels."
Also, for posterity, the competent ones are running the headless, server version of Ubuntu. (As opposed to the bloated mess that is Ubuntu Desktop). The server version catches a lot of flack it doesn't deserve.
Microvms or containers could give you external control of the networking. Then you would put whatever you want behind warp inside the warp container/vm.
+1 to this observation. I run zfs arrays at both home and work and it's way more likely that your controller is flaking than you have that many simultaneous drive failures.
The unfortunate reality though is that you can't trust the current copy of this data, even the snapshots, unless the restore passes a scrub post-restore.
My OS uses about 700MB when I log in from a reboot, then I immediately load ~15GB of terribly optimized browser tabs, mostly for work (teams, google voice, discord, outlook, etc)
Less cynically, I believe the argument in scripture is the inverse. Man was created in god's image therefore we probably inherited a lot of properties of the devine.
Inference is dirt cheap in comparison. Hundreds to thousands of concurrent users can be served by hardware costing in the high-thousands to low-ten-thousands.
Training those same foundational models is weeks to months of time on tens to hundreds of millions worth of hardware.
P4 motherboards are generally modern enough to USB boot, but only if it's MBR formatted.