bippingchip

Intel changing up its GPU strategy with a pretty big reorg, which to me feels a bit like backing rather than doubling down, as they try to spin it.

Having worked closely on the circuit side of things with folks doing research on spintronics and STT-MRAM devices: it’s going to be a very long time still before this can replace embedded SRAM. The MRAM itself can be made quite small, but you still need an access transistor and wrote currents are still too high, which means very big transistors, which then dominate memory area… Also, writing is much slower compared to SRAM.

It’s a very nice embedded flash replacement though…

This is impressive!

I’ll admit: My first reaction was: cool but why not just use qemu to get a more full fledged system, add rv64 etc?

But the readme explains it very well, I admire the dedication and consistency it takes to stick to rv32 no MMU only to be able yo fit it into very minimal environments in the end.

(I myself tend to be much more prone to allowing feature creep :)

I can confirm. Set and forget all the way. Works very well on iOs and also on Samsung smartTV (those are evil without some form of Adblock!)

Hardest part for me in the set up was that ad guard should query my router for local domain, because that one keeps track of which dhcp IP address is owned by which host. (This way I can always use host names on my internal network even for devices that get an address via dhcp. Very convenient if you play with Pi Zero and other toys)

Yes! I never got the hang of the read line short cuts. Only one I remember and regularly use is reverse history search: ctrl-R.

The downside is that while vim mode works nicely on bash, other commands like gdb etc that also use read line aren’t as easy to get into vim mode, if they support it at all…

You can edit the current prompt in a full vim by pressing ‘v’. On :wq the content of the file will be run as the command. In that case you can use the macros with q (and any fancy vim plugin you might like)

Not being familiar with recommendation engines at all, this article does a decent job showing some of the basics in a bird’s eye view, at least to my untrained eye.

That said, it’s still a promo piece for their graph database.

For me this is a good example of the old adage that you should not confuse vision with strategy.

The vision is you realize the universal truth (in Amazon’s case people want things cheap and fast).

The strategy is how you deliver on making that visions reality. This is what you call (rightly so) your core business model.

Only experience people tend to get all caught up in the big ideas of the vision (especially middle managers) but spend far to little time on making the hard choices that come with deciding in strategy: it’s as much, if not more, about deciding what not to do as it is commitment to what you will do to realize your vision.

Thank you for the clarification (it’s not nitpicking). I hadn’t read any of your posts recently - I’ll have some reading up to do I see :)

(This should probably have a 2016 tag)

I’ve been a long time fan Anders Hoff’s work - it’s well worth it to check the other posts on incovergent.net - quite a few have been covered here on hacker news.

Interest also to see how he has moved to Lisp over time, functional programming seems to work well for him for this type of generative art.

Lastly - I think it’s refreshing to still see this on the front page : beautiful generative art without DNNs…

One thing the article does not mention is that the darker (and softer) the pencil, the more easily it smudges. Even if it is not charcoal, you’d do best to fixate if you use really soft pencils like the 12B they tested.

As a side note: I love a thick 5.6mm lead clutch pencil for quick notes and sketches in my notebook at work. It writes like a pencil and you can erase it, but it has the feel of a sharpie (thick lines, not very accurate). It forces me to be concise in what I write and draw.

But after experimenting there I did settle on #2 lead for most uses. My notebook simply becomes one grey smudge much too quickly with softer leads.

This is a very important distinction. With deliberate attention, you can indeed catch many fakes in this kind of scenario (to me it seems the background is often a giveaway, but you do need to focus on it)

But in passing, accompanying a news article, a tweet or an instagram post, are you paying as much attention? Those are the scenarios where the potential for harm is much bigger.

the problem is the material properties link power and variability. This means that you either get good variability or higher power consumption, but never good variability at low power at the same time

Having worked on this for quite a few years (compute in memory with a variety of emerging memories: RRAM, MRAM others) with a substantial research team, our conclusion ended up being you are better off with SRAM based solutions. And most likely tightly coupled memory with digital compute is better than doing compute in memory.

The newer memories like RRAM are simply not stable enough: too much variations, drift with temperature, reliability etc. On some cases you can try and re engineer the devices to be better suited but they invariably end up being larger, or more power hungry, and often both. See for example https://ieeexplore.ieee.org/document/9405305 (sorry for the paywall - no open access available)

Adding insult to injury, none of these emerging memories can be integrated with highly scaled digital CMOS. (22nm is about a low as you can go for eg MRAM and RRAM - where they are offered as embedded flash alternatives) But you will always need flexible, programmable digital compute in order to have an AI accelerator that can do more than 1 flavor of resnets.

SRAM, in the meantime does scale relatively nicely and co integrates well with digital logic across the whole spectrum down to 5nm FinFETs and below.

+1 for pushover

I’ve been using for quite a fee years. The app is not free but once you pay the one-time fee for the app, they have a generous free tier for personal use, and it has worked quite flawlessly for me so far.

It’s always meter specific, I use the plugin and dongle as described here: https://www.home-assistant.io/integrations/dsmr/ this works with a bunch but by far not all of the meter types in EU.

While I haven’t gotten my lights set up like this, I do use home assistant, just like the author and it is simply amazing. For things like this, the setup can get quite complex, but the main benefit is that it provides a unified interface to any connected device in your home.

My main use case is energy tracking: it does a marvelous job collecting and graphing data across a bunch of different sources: from the solar panels from one vendor, the semi proprietary smart meter protocol my electricity company has, weather data etc. Having all that data together in 1 place is an invaluable tool to make any investment decisions on energy improvements to our house.

The form factor is a nice gimmick and fuels nostalgia for people who played these games as kids.

I have a SNES mini, but it’s mostly collecting dust by now. I simply don’t have the time with 2 young kids, and to be honest: I underestimated how limiting it is to have to use the family tv for this. Much more convenient to do some handheld gaming in the sofa on a phone or tablet…

The collection of games is curated so yes it’s a nice selection but in my case I mostly just played Mario and Zelda.

They have had decades of close collaboration an customer feedback from their internal design teams, they have process technology teams using some of those designs as reference points to improve upon for the next node.

Don’t underestimate the kind of gravity well that creates, especially in a large organization like intel. Their process is well tuned for high performance Xeon chips. Its not impossible to change that, but extremely hard, and it will take tim.

At the same time they currently have near 0 external customers trying to use their process, they are having a hard time figuring out what their customers need. (I have no doubt they are working hard to change that) But is there a way for them to improve a process node for an external customer? What if it creates an ever so slight downside for x86 products? These are the difficult decisions they will need to make at some point.

TSMC, on the other hand, has decades of experience of doing exactly that: designing a process for everyone with every fabless company in the world, for type of chip, whether it’s server or tiny ultra low power class. And they have a very broad range of flavors of every node to show for it.

This is probably one of the biggest challenges ahead, likely more so than making hardware that still somehow scales despite Moore’s law slowing down and tapering out entirely. The increased scale of integration with advanced 3D packaging and chiplets is a nice way forward, but the heterogeneity of those solutions need innovation at the programming model as much as anything else: how do you make such a complex system accessible in a way that the programmer can actually use it efficiently?

As much as I like semiengineering and their articles, they are very hardware and process technology focused and tend to miss the software side of things.

I was looking for a Christmas present robot kit for my 8 year old and still on the fence between mindstorms or the makeblock robots. (The Codey Rockey in particular, but the mbot looks really nice too)

I have colleagues that organize coserdojo sessions for young kids fully centered on the makeblock robots. You can program them in scratch to get started, but a push of the button gives you the equivalent Python code once the kids graduate from scratch.

I am surprised to see Lego cancel mindstorms, but it does make the decision easier…

That’s because they skip quite a few steps. You would first deposit a photosensitive layer, then expose it wit a pattern of light, which will allow you to selectively open up the parts you want to toch next.

For step 7 for example, you create a masking layer that way, after which you etch away only the exposed parts where source and drain contacts will end up.

For step 8, you do something similar. The openings you’ve etched in step 8 will allow only to dope the contacts, but you need to mask the nmos and pmos separately, otherwise you get the same doping in both transistors.

A lot of the ‘magic’ to make this work is in the materials and chemistry: you need selectivity: etching that only interacts with the exposed parts. In highly scaled finfet nodes, the material combinations are very complex as you sometimes want 2, or 3 different material combinations to interact in the way you want, reliably, at nanometer scale.

(Edit finfet not finger)