I don't even think they're driving hype, they just have a problem common in the hardware world, they are far too optimistic in their shipping timeline.
Versus a company like Bambu who seems to ramp up production now, in advance of selling the product.
> Formlabs will happily upsell the "Form Wash" for this for another 699
(A separate aside, since the main ones were addressed: Formlabs also launched a new "Wash+Cure" station that does wash and cure for $199, instead of their standalone (but better, of course) units that cost something like $1000 together.)
You need to buy a tank anyway, which is $50 or so — that works with their new Alpha resins. The slightly more expensive tank is necessary only for their engineering-grade resins (only a couple are certified for the Studio 1 so far) and 'OMM' (Open Material Mode), which as mentioned above, comes out of the box on the Studio 1.
(Not replying to you directly, but generally to this thread)
Scrappy builds aren't "affordable" but are more so than running Nvidia + optimized builds. You could put together a machine for $5000 or less that competes with frontier models from less than a year ago, albeit a little slower.
And cards like the $1600 (sheesh) R9700 have actually gotten more capable through community efforts at optimization, since their launch.
I think if the bubble can ever deflate/pop, and GPUs and RAM back off a bit, we'll see the dam burst for local AI.
Right now it's only accessible to people who can justify the costs.
Once you do get measurement down to nanoseconds, you quickly realize how 'bad' so many clocks you thought were good actually are. Lots of GPS modules with massive sawtooth behavior. Over a day they may be within tens of ns. But over a minute, a lot worse!
Yes, but over a decade they are ridiculously precise.
And that's precisely the problem: when you need very short term stability that nicely couples into longer term stability you have your work cut out for you.
Another downside of this kind of project is that to collect the data you are always looking at a minimum of 24 hours or so of sample time and at high enough rates that really adds up. Suddenly you need media that are never going to hiccup during long sequential writes and it turns out that's not a simple problem either. Before long you'll be messing around with the buffering deep inside of GNU Radio (which is amazing piece of software by the way).
But: when it finally works it is extremely satisfying. To see a clock driving a frequency counter and a non-derived clock that lists 9 zeros with a 100 second sampling window is pretty cool.
Yup. I’ve known the guys at Meinberg since around 2003, when I started helping out with the NTP Public Support Project for NTP.org.
They are very well aware of both the short-term and long-term effects that they have to compensate for. And getting that both smooth and accurate down to the nanosecond level is particularly hard.
They are not the only vendor of Stratum-1 NTP time servers (on the AWS EC2-Chronos team, we called them Primary Time Appliances), but I think they are one of the best. If not the best. And a friend and former co-worker of mine from EC2-Chronos is now working there.
Fascinating stuff all this. I just had a very practical problem to solve so my interest was purely accidental but I have deep respect for the people that make such an outwardly simple thing the focus of their career to the point that they can contribute in a meaningful way to the field itself. Half the world (or more) does not recognize the degree to which we casually rely on incredibly precise timekeeping and how much of an achievement it is. More so because it is so invisible. What I've also found is that everybody in this domain is incredibly friendly and willing to share advice and knowledge.
Even some smaller vulnerabilities are spread through hosting companies pretty quickly within a day or two prior to public release, to give teams time to make sure staff will be ready to patch immediately.
The wider the pre-release news spreads, the more chance you get someone who will spill the beans (intentionally or not).
It would not be easy (or reasonable) to take a public build like that private to build a 'secret' image version, and I'm not sure GitHub even has the capability to do that. I would not trust a build pushed to Docker Hub separate from the established / public / auditable process they have in place.
All dependent (official) images need to go through the same process, so that makes it fairly problematic, and I think Docker needs to take a look at it for the future.
One pattern of AI-assisted writing I hate is that technical articles now read as Atlantic journalistic tech-noir stories, which require scrolling through 20+ pages of content to get to the meat of a post.
Versus a company like Bambu who seems to ramp up production now, in advance of selling the product.
reply