smat

Check out Asahi Linux, they run on Apple Silicon and have translation for 32 and 64 bit x86, so they even go further than what Rosetta achieved. Open Source as well.

A very interesting read.

The article states that they added a computational correction step which is run whenever a result might be incorrect but skip it most of the time.

I am wondering whether this might cause problems for small real-time systems that do not have an FPU. If the runtime cost of an operation depends on the input, it is very hard to figure out whether deadlines can be met in all operation conditions.

Great post, the fusion of scale with unit is a mess.

When using it as a factor, for example when describing attenuation or amplification it is fine and can be used similar to percent. Though the author is right - it would be even more elegant to use scientific notation like 1e-4 in this case.

For using it as a unit it would really help to have a common notation for the reference quantity (e.g. 1mW).

But I guess there is no way to change it now that they are established since decades in the way the author describes.

While this gives an impressive boost in performance, it also means that frametimes are around 10% longer when the status bar has to be updated.

Overall this can mean that in some situations the game feels not as smooth as before due to these variations.

Essentially when considering real time rendering the slowest path is the most critical to optimize.

Nordic semiconductor also has microcontrollers with BT and Wifi. They are much less common than espressifs solution though.

Though this is an artificial constraint imposed by the platform holders (i.e. Apple and Google).

Once they figured out how their new AI assistant needs to look like their own rules for 3rd party apps won’t matter anymore.

I could see that in the coming years the value of Waymo for Google is not actually in collecting revenue from transportation fees but to collect multi modal data to feed into its models.

The amount of data that is collected by these cars is massive.

You are right it feels off.

The position of the guitar in stereo is all over the place, higher frequency elements appear to come from the left while other parts are more centered.

At the same time France had to buy loads of electricity from Germany in the last year because several nuclear plants had to shut down due to a) maintenance problems and b) low rivers that could not provide sufficient cooling.

You also pay for quality control. Cheaper brands often have decent audio quality on average but high variation in between individual pairs of headphones. They also tend to fall apart after 1-2 years of use in my experience.

You can modulate the brightness without flicker, you can try it yourself with a lab power supply.

It is just that filtering the output of the regulator is “expensive” as you need a relatively large capacitor. If this is omitted in combination with a low frequency PWM, you will experience flicker.

I assume the author of the library works under a hard real-time constraint. Under such circumstances (an example would be low latency audio) you can not tolerate the latency impact of a sporadic syscall.

According to an article at Ars [0], longer sessions are more likely to output undesired responses from bing chat. So the limits might simply be a workaround to avoid articles about strange behaviour.

[0] https://arstechnica.com/information-technology/2023/02/micro...

I would argue that the people who use AI generated assets and combine them for a bigger thing such as a video game are still artists. They are simply realizing their vision on a higher level.

Boston dynamics also published a making of video. There it is explained why the luggage scenario is still far away: The robot has perfect knowledge of the object it is transporting (size and weight distribution). In order to be able to transport any object, the control algorithms need to be able to estimate themselves where to grip an object and what the objects weight distribution is.

Still, the results are very impressive.

They are using a combination of lidar with HD maps, radar and cameras.

Is there a list of functions for newlib?

Very interesting read. The author notes that double pumping the 512 bit instructions to 256 bit execution units appears to be a good trade-off.

As far as I understood ARMs new SIMD instruction set is able to map to execution units of arbitrary width. So it sounds to me like ARM is ahead of x86 in flexibility here and might be able to profit in the future.

Maybe somebody with more in-depth knowledge could respond whether my understanding is correct.

I am also curious how this period in movies will hold over time. Older movies with more practical effects (such as Star Wars) still have a very crisp look to them while newer movies look blurry and bloomy due to the heavy post processing.

Appears to be common industry practice at least for entry level devices. My old HP has the same behavior. Cannot scan with a defective print head.

While the ideal filters have an infinite long inpulse response and are therefore „non realizable“, shorter filters simply increase the noise level. With filters of 100ms length you are already at SNRs that is far from what any human can handle (think of hearing rustling leaves while standing next to a jet engine).

You will however need to accept delay from the lowpass filter, which is fine for music reproduction.

So while these effects are measurable with very expensive equipment, they are in all metrics orders of magnitude better than any analog equipment.

Fraunhofer is an institution in Germany positioned between universities and the industry. They provide applied research and license it to companies.

For example mp3 was mainly developed there. They hold many patents in the field of audio and video coding which obviously fund parts of the organization.

More recently, they established a new audio codec for Bluetooth LE under proprietary license which in my opinion does not have any advantages over OPUS.

At the same time not all of the criticism is fair because they are not a patent troll and anticompetitive in their license agreements.

I did not mean to discourage people from making their own evaluations. Maybe I did not word it correctly.

The article starts with questioning best practices for implementing real-time processing threads. The statistical analysis presented in the article is based on a single workload and a single machine, which in my opinion does not help to answer this question.

Not sure what this empirical analysis adds to answer the initial question. Sure, modern computers are fast and modern operating systems are designed to reduce latency.

At the same time the “best practices” of not using mutexes and malloc on real-time threads are there for a reason: They potentially trigger a system call, which will add quite some latency (as the measurements show). Because real-time is all about deterministic latency, this is undesired.

Signal offers to validate public keys of contacts via a separate channel (QR codes) to detect MITM attacks.

The second claim is that the ciphers used are outdated. A counter-claim that can be made here is that signal relies on well established and widely used methods which ensures that any significant weaknesses are more likely to be found. So far there are none that result in practical attacks.

At the same time the investment in epic is only a very tiny fraction of tencents market cap. So it does not make sense to invest in tencent if you are interested in epic.

Wifi6 adds QAM1024 as a modulation, so under excellent condition its downlink will be faster.

Many of the changes aim at high density environments, such as trade fairs and conference centers. Once the majority of devices are upgraded, these situations will be handled much more effectively.

Lastly, OFDMA is capable of significantly lowering latency if all devices in a network support it.

If you are using Wifi6 at home with few devices connected, do not expect noticeable improvements.

Germany is even worse than you describe. The pension system managed by the government is not backed by any assets at all, but works by taking from the working population to the retired population. Given the age distribution in Germany this means young people are increasingly paying more to this system, while retired people receive less and less per person.

This was of course obvious already a while back, so the government decided to introduce additional ways to encourage saving for retirement (by giving tax discounts). However, they also managed to screw this up, because only contracts from certain insurance companies apply for these tax discounts. And these contracts have such a high management fee, that the real return of those constructs is negative.

Not sure which aspect the earlier post meant but the rays are only refracting on the exit surface, when of course they also refract on the entry surface.

I think he does though. He mentions that many assets are priced on a level that is only realistic if interest rates remain on the current level for decades.