buildbot

46 ms·

781.
by buildbot 3y ago
I recall reading a great intro to ZFS article that did this! They had like 16 USB drives!
782.
by buildbot 3y ago
What’s the weirdest thing one can raid together these days? Maybe a bunch of raspberry pi picos?
783.
by buildbot 3y ago
I feel like there has been kind of a weird concerted effort to push that zfs is bad due to this bug and how trust has been lost etcetera etcetera - super annoying when most other filesystems just corrupt your data and nobody will ever know
784.
by buildbot 3y ago
It does, technically.
785.
by buildbot 3y ago
Probably. There are several companies going with this approach. The real trick is the software and ecosystem around it. It does not matter if you can outperform an H100 if you spend weeks trying to get software to work or debugging if a NaN
786.
by buildbot 3y ago
The Microscaling paper and the following MX OCP spec is along the lines of what you call batch floating point, though I believe the original 1963(?) work on it called it "blocked" floating point. Datatypes are really tricky. Hardw
787.
by buildbot 3y ago
Even below FP8 works: https://arxiv.org/abs/2310.10537 (6 bit and lower)
788.
Introduction to MicroScaling (MX) for AI Tensors(fpga.org)
4 points by buildbot 3y ago|0 comments
789.
by buildbot 3y ago
There are also fully optane based SSDs such as the 905p or the 5800x
790.
by buildbot 3y ago
Optane SSDs are really special and still have a place. SLC is 30 microseconds at best, FIO puts my 905p optane drive at 5-11 microseconds. That means for a single threaded latency bound task, a 905p may finish it 6 times faster than an SLC
791.
by buildbot 3y ago
That was probably a hybrid optane setup, like the h10 ssd. It was a tiny optane ssd and a normal ssd on the same m.2 card. Not very good compared to real optane. A really noticeable improvement when running apt, or dnf, or any windows progr
792.
by buildbot 3y ago
Oops sorry, I missed this!
793.
by buildbot 3y ago
51 million isn’t actually all that much for a chip startup, in my opinion.
794.
OpenAI is paying out $51M to a chip startup that Sam Altman invested in(businessinsider.com)
15 points by buildbot 3y ago|9 comments
795.
by buildbot 3y ago
> Not paying will be even cheaper. They don't need this to obtain the de facto monopoly given the massive costs in training. Even if other companies offer their own AI, it will be for niche scenarios (like character creation for gam
796.
by buildbot 3y ago
Big tech would _love_ to "loose" these cases so they can gain a complete monopoly on it. It would take massive capital to pay the aggregate license holders, working in Capital's favor.
797.
by buildbot 3y ago
Yep exactly - large corporations are like, 25-75% compliance keeping machines before they do anything else. They have entire departments for this kind of thing. All these companies would love to lock the best training data behind a multi-mi
798.
by buildbot 3y ago
Oh I didn’t know that! Thanks for the explanation.
799.
by buildbot 3y ago
Exactly. The sad reality is in any (mostly) closed system, say, the USA or even earth, there are resource limits and you might be able to do more good by treating 4 patients with more simple cancer's vs. 1 patient with a very complex c
800.
by buildbot 3y ago
That's not a particularly good example, their latest self-promo post has a warning from the mod on it...
801.
by buildbot 3y ago
Medicare Advantage is incredibly predatory and a misalignment of incentives. It pushes what should be a socialized program to the free market, which abuses it and rejects care in order to turn a profit. It's very funny people were worr
802.
by buildbot 3y ago
This is basically only valid for software engineering FYI - a Data Science/Applied science interview is going to be very different. And probably differ from team to team with much higher variance!
803.
by buildbot 3y ago
Aww, sadly nothing for my own websites! This is such a great idea, often when I find a small blog or site I want more of it! This is the perfect tool to discover that. It’s a clear and straightforward idea in retrospect, as all really great
804.
by buildbot 3y ago
I too switched to OPNsense., which has been awesome. Partly due to weird dns failure issues, but mostly due to what turned out to be the bullshit claims about security issues :/ krebs did real damage to them by pushing that story.
805.
by buildbot 3y ago
That seems needlessly rude to those people. Okay, refresh probably won’t help, but you can build logic gates in css, and css has for example, the ability to animate things forever: animation-iteration-count: infinite; Hypothesis - you can f
806.
by buildbot 3y ago
Can you explain how the blog post/other results people have on CSS being turing complete are wrong? Edit - would it be considered cheating to use: <meta http-equiv="refresh" content="10">
807.
by buildbot 3y ago
Don’t oled displays support pretty high refresh rates? And also have way lower GtG latency?
808.
by buildbot 3y ago
Aren’t they still violating linux’s GPL license?
809.
by buildbot 3y ago
Yeah I totally agree with this as well.
810.
by buildbot 3y ago
Also either network people get it wrong, or the optics modules really are DACs? Now I am confused too haha : https://serverlabs.blogspot.com/2019/10/a-study-about-sfp-mo... Tangentially… you could build a SFP modu

More ›