buildbot

48 ms·

751.
by buildbot 3y ago
The HICANN chip is really interesting, never seen or heard of it before!
752.
by buildbot 3y ago
Dell Windows laptops will also cook themselves. (For example: https://www.dell.com/community/en/conversations/xps/dell-957... ) Macbooks are the only device I would trust to not light a bag on fire…
753.
by buildbot 3y ago
Disagree there. Humans have massive compute, dual optics, and amazing filters. Computer vision has 1-2 of those three, and I don't think we are near an AGI for self driving yet. Driving is IMO, an AGI level task. Does you dataset have
754.
by buildbot 3y ago
Very interesting work! More details here: https://depth-anything.github.io/ It seems better overall and per parameter than current work, with relative and absolute measurement. Is there any research people are aware of that
755.
by buildbot 3y ago
It's always felt like a relic of the past to me. It feels oriented to someone of more advanced age who wants to spend a medium amount for a medium product and not think about it anymore. So they come up with golf ball tests for toilets
756.
by buildbot 3y ago
That’s a relatively inexpensive choice, nicely suited to the job. Could have gone with a h6d-400ms for a bit more resolution ;)
757.
by buildbot 3y ago
Pretty straightforward these days! Though I've been doing it for awhile. There a lot of little gotcha's like tubing choice and fittings but once you get the hang of the options it's pretty fun. Though cooling 4090s even with
758.
by buildbot 3y ago
A self built machine with dual 4090s, soon to be 3x. Watercooled for quieter operation. Did the math on how much using runpod per day would be, and bought this setup instead. Using Fully sharded data parallel and bfloat16, I can train a 7b
759.
by buildbot 3y ago
Mamiya did as well, for the RZ67. Or at least, their charts claim they are corrected well into 700nm+: https://ianbfoto.com/downloads/Mamiya%20RZ67/Mamiya%20RZ67%2... They say no focus adjust is necessary for the
760.
by buildbot 3y ago
Anecdotally, a lot. Modeling of Aspherical elements has gotten better and it is much easier to make them as well. Mamiya made some of the first APO computer corrected lenses back in 1988 - https://lens-db.com/mamiya-apo-seko
761.
by buildbot 3y ago
I found the article explained exactly why using fluorite helped chromatic aberration. It adds another degree of freedom, due to the partial dispersion.
762.
Neural Network Quantization and Number Formats from First Principles(semianalysis.com)
1 points by buildbot 3y ago|0 comments
763.
by buildbot 3y ago
I hope AMD sees the light and helps F4FPGA develop a more complete open source toolchain for their FGPAs ( https://f4pga.org ). With this subsystem and an open source compilation flow, FGPA experiments would be way easier.
764.
by buildbot 3y ago
There is something kinda like this already - a much less integrated, but works via v4l2: https://developer.ridgerun.com/wiki/index.php/V4L2_FPGA
765.
by buildbot 3y ago
I am - literally - heating my condo with my ML training workstation right now. Works great! I have electric wall heaters, so I might as well get something out of burning 1000W of electricity.
766.
by buildbot 3y ago
Fun tiny fact I found a while ago - The reverse of this also works! You can init deeper layers by copying the weights from previous layers. Not a big improvement in training time, but it does reduce the initial loss!
767.
by buildbot 3y ago
Yep. Pushes the industry to research lighter composites too, let’s see more carbon fiber chassis like the BMW i3! (A really underrated great little vehicle, my sibling has one and it rocks!)
768.
by buildbot 3y ago
As another commenter said, it's CUDA. Intel and AMD and whoever can turn out chips reasonably fast, but nobody gets that it's the software and ecosystem. You have to out-compete the ecosystem. You can pick up a used Mi100 that per
769.
by buildbot 3y ago
Huh, I wouldn’t have expected rust to work on solaris+sparc but I’m oddly glad it does! I keep getting tempted by cheap eBay sparc systems…delicious strange risc’y’ness..
770.
by buildbot 3y ago
Interesting! Thank you!
771.
by buildbot 3y ago
Is there any research out there that can be used to reconstruct a fairly accurate model for CAD purposes? Or is classic photogrammetry still the way to go there?
772.
by buildbot 3y ago
I worked at a company that shipped petabytes of ZFS storage for years and years. Think 42u racks with 60 drive JBODs. Hundreds of these. Hundreds. And they were a very small company. Can you imagine the scale Sun and Oracle shipped hardware
773.
by buildbot 3y ago
YES! You jogged my memory - it absolutely was a lecroy with a floppy disk as well! What a cool piece of equipment. Though I won’t say I was too sad when we got to use a 5GHz touch screen Tektronix instead! It ran Windows XP and was generall
774.
by buildbot 3y ago
The lab I worked in during undergrad had a massive pile of old scopes and such. One of them was this monstrosity of a scope from the 80s (?) that had a built in thermal printer to capture the output, and was I think in the 1-2ghz range! I’m
775.
by buildbot 3y ago
Did not know this, I don’t mind learning something. You’s think though that autocorrect would pick up on that…
776.
by buildbot 3y ago
No worries, I completely understand!
777.
by buildbot 3y ago
But it's totally different than corruption in other filesystems. People are acting like upgrading to ZFS 2.2.0 ate all their data like XFS used to back in the bad old days. I remember once the power went out at my house, and the entire
778.
by buildbot 3y ago
Perfect for me is not perfect for others of course. I have not run into the issue you linked, that sounds frustrating to deal with. Backups is the only (near) 100% way to insure you data I agree.
779.
by buildbot 3y ago
I wonder if you did that same calculation today, what percent of the speed of light would the edge of that disk be moving...I miss that kind of humor.
780.
by buildbot 3y ago
Oh that's a good one

More ›