Performance (tok/s and PP) or quality (model size)? Pick one.
In terms of GPU memory bandwidth (models fitting in the ~48GB of RTX 5000 Pro card), the RTX card I described above has over 2x the bandwidth of an M5 Max.
If leveraging system RAM (the 128GB-256GB outside the GPU) to run larger models, then the memory bandwidth is ~6x slower than M5 Max.
For models fitting in the ~48GB RTX memory, like dense Qwen3.5 27B models, the RTX will be 2-4x faster than M5 Max. For models that don't fit in the 48GB RTX memory, the M5 Max will be 5-20x faster.
Also worth considering future upgrades: Do you plan to throw away the machine in a few years, or pick up multiple used RTX 6000 Pro cards when people start ditching them?
Side note, I read that GrapheneOS project is having some challenges recently.. between [0]the Android kernel drivers no longer having their Git history of changes being released (only a code dump with no history) - and [1]one of Graphene's two core contributors being detained/conscripted into a war.
it doesn't matter who the war is with/against, if the lead developer of a software project gets drafted that will likely affect the users by virtue of them having less time to work on it (or none, if they get killed, which is not, y'know, unheard of in wars)
If you want consistency, you only need to convince everyone to switch to a single font renderer (e.g. freetype). That won't happen, though, because OS font renderers have quirks that cause them to render the same things in subtly different ways, and users have come to unintentionally expect those quirks. Even if rendering is 'better' in one app.. if it doesn't match the others or what the user is used to.. then it won't 'feel native'.
Maybe if what freetype is pushing for (fonts are WASM binaries) continues to take hold, and encompass more of fonts, we'll find more consistency over time though
I think a better UX for average consumer would be more a side-swiping filter menu similar to that of social media mobile apps, with different non-math style names "default blur", "circle blur", etc. Especially as more people do not use desktop computers today.
Also maybe LLM integration so you can just explain what you want done, then it does it, instead of needing to follow some tutorial to learn the software
> Also maybe LLM integration so you can just explain what you want done, then it does it, instead of needing to follow some tutorial to learn the software
I like how this counts as a reasonable side remark today but would have been utterly delusional just a few years ago.
He wasn’t stupid. He’d observe their own developers and see how they rely on the terminal and command line for their work. He’d ask pointed questions and demand thoughtful answers.
Then he’d find a way to make it the #1 AI developer platform or distort reality until it is.
FPV today is largely using ExpressLRS[0], an open protocol for running FPV quads' control links (also using esp32 usually)
Video for FPV quads on the other hand is divided: some use analog (mostly because when you lose signal, it's a gradual loss rather than complete loss + a reconnection process), while others use 802.11-based communication which is actually quite competitive. Walksnail Goggles for example use 802.11 for their video links, although not super open/documented people do dig into how it works (see Chris Rosser's work.) Latency is good enough with such setups for fast-paced competitive quad racing, at least
I will add for those who haven't tried these, as it usually doesn't come across in verbal comparisons:
Digital FPV (like DJI's system) looks vastly better than analog. I think the best comparison is: Digital looks like a modern movie, youtube video etc, and analog looks like CCTV footage from the 80s.
I will also clarify on ELRS: It uses ESP32, but it is a thin wrapper around Semtech LoRa (~915Mhz and 2.4Ghz) SX128x and SX126x chips, which are doing the heavy lifting.
I heard people flying analog also use quite a bit more transmitter power than allowed for non-regulated devices to keep things manageable. Do you have any reference what's possible with digital when staying in the allowed regime?
The "allowed regime" also differs based on where you live. Iirc 25 mW is the maximum permitted power in Europe. Have to double check. That gives you about 100-500 meter range in an open area.
A big advantage with digital video in the US is that some systems, like DJI’s, have FCC approval and can go up to 1 or 2W EIRP legally, so the allowed regime is bigger (there are basically no FCC approved analog video systems, so really you need an amateur radio license and to operate under amateur radio rules for analog video in the US).
I think graceless degradation isn't an inherent property of digital radio. You definitely could have some kind of digital modulation that takes advantage of FEC etc. but gracefully degrades.
I'm sure it exists, there probably just not much market for it.
Almost all existing digital FPV systems gracefully degrade as much as they possibly can, both by using FEC and walking down a large number of available MCS as link quality gets worse (and reducing bitrate to match). But when they do eventually fail, re-sync is a much harder problem for them than it is for analog.
To a great extent you trade statefulness (inter-frame compression, FEC, HARQ, sounding parameters, overhead) for efficiency, so improving link recoverability comes at a cost. There is an existing solution which makes this trade - HDZero. It works amazingly at close range in clean RF environments and has good analog-esque falloff characteristics, but in terms of absolute efficiency it is very bad and therefore does not have the same quality or range as other solutions. Certainly a better hybrid approach is possible but to your point, niche.
(1) Doesn't match the situation at all, because the law didn't require the paper to shutdown - it required a foreign company to divest so that it is US-owned, and the paper could continue operations as normal.
That's a pretty substantial difference.
(2) Also doesn't match the situation, there is no requirement that TikTok restrict the reach or audience of their content in any way AFAIK.
(3) The situation is more akin to "foreign government owns the local library, and can decide based on the identity of the person walking in which books the person is allowed to see and check out" - seems obviously problematic at least /if they do that/
reply