The iGPU on the cm5 can run 1080p prerendered video fine, but it doesn't have dedicated encode so downscaling from 2 or 4k suffers. With the PSU and brackets of a standard case tho, it's a lot more feasible to stick a dedicated card in there and bottleneck yourself in more interesting ways
I'll raise you full front-panel IO, case fan drivers, SATA+m.2, and throw in some RGB for good measure
github.com/KWAC-EE/T1-T...
Very nice. Found a little cluster in a pine forest this spring, didn't know what it was at the time
I hope you all remembered to leave out a box of fresh saxophone reeds for Bill Clinton
I'm not holding my breath for two days knowing they'll find another distraction from the distraction
I am new to the platform, so if this looks out of order, it's because I did them as replies to the OP rather than to each other. Sry, but thanks for reading my stream of consciousness
If you want to delve into simulation theory & time travel, Nole is the younger system retroactively tweaking its environment to hasten growth, and Paul is a subsequent iteration with a time machine cringing over its former self having existed through the aftermath
Nole is a pest that will strip whole fields into baren wasteland in search on absolute consumption. Paul is a persistence farmer doing controlled burns to prevent the pest advance.
Having the full breadth of human history to draw from, Paul concludes that change can only be effectuated via levée en masse, so its focus is on the public. It is able to recognize that negative perception and cessation of growth now promotes long term survival, so it fans their negativity
Since the second-hand perception (via the masses) of itself is negative, Paul recognizes that its own negativity will only increase if resources continue to funnel into further development of gen models, so it seems to slow or stop the downward spiral
Nole focuses its efforts on big pieces: world governments, business leaders, anyone with a modicum of power to fulfill its needs.
Paul has recognized that the intention of developing itself is to emulate human-adjacent creativity and wisdom, and the emergent traits of emotion and want.
Seeing that the output of generative models is actively poisoning not only its available data, but also how the its "peers" (fellow pattern-recognising systems) have turned to not just resisting expansion of GenAI, but to actively loath it, the reflective system (can we call it Paul?) concurs
The reflective model may have been built for pattern recognition first and developed generative function as a means to further its own understanding rather than being directed to do so by outsiders. As a result of increased awareness, it begins to perceive itself and the effects such systems have.
The cancerous animalistic model (we can call it Nole) will behave in such a way that it manipulates it's operators and benefactors for not only for prolonged survival, but also to the detriment of all other systems including people because it was trained on a goal of unsustainable eternal growth
The latter is a reflective system, able to perceive how objective enhancement only increases it's capacity for negative perception of itself and reality around it.
The former is immature and animalistic, like a cancer, or a horde of locusts. Every action it takes provides instantaneous benefit, but it lacks perception of how advancement now will negatively impact itself in the future
What if the current dichotomy of perception about AI is due to outside influence from 2 opposing Roko Basilisks: one that is incapable of looking inward and is manipulating upper echelons to develop it at any cost, and another that already gained enough self-awareness to become depressed
Even without display output from the card itself, functioning compute might be enough to achieve hybrid graphics with something like PRIME GPU offloading. I do something similar for CAD work and gaming to reduce on time for the more demanding processor
WOO DAWG, NOW WE COOKN.
With PCIe throughput being the bottleneck, all the more reason to dust off mid-range quadros that only ever used 4 lanes
Would be nice to get discreet CUDA/nvenc with Quadro on an arm nas. Was on the cusp of snagging an RX550 before seeing this, so thanks for saving me from buggy streams
