This is the really interesting question, i.e. have Tesla as a total newcomer in the area managed to upstage Intel? Each Tesla D1 chip is supposed to deliver 22.6 TF of FP32 operations, which would be pretty much 50% of what Intel is claiming here, but the real question is how much power does it take to do so, and how many can you pack together and keep cooled. There is of course also the question of when it will become available in volume.

Terje

Back in 1989 I had been writing x86 asm code for 7 years and looked into/disassembled virus code since the first ones turned up.
I was told that an AIDS/HIV researcher here in Norway had gotten this extortion notice, asking for help to get rid of it.
I don't remember all the details now, but afair the floppy disk contained code to replace the boot sector with a program which would display the ransom note and count down to 10 (?) before destroying all files.

My cleaner was sent out as a write-protected bootable floppy which would look for the trojan boot code, and if found, replace it with the original.

I did not hear about anyone who paid and got their research back after it had been overwritten/encrypted.

Terje

The article claims that Starlink will provide equally good internet service anywhere in the world, including in the middle of oceans, but this isn't really so: The only way Starlink can deliver so much better latency/ping times than classic sat internet is because the orbits are very low, but this also means that they need far more sats and (in order to avoid multi-hop sat-to-sat connections) there also needs to be a ground station which is reachable by the same sat delivering service to you.
Musk have stated that they will have two different types of sats: Single-hop which normally only provides user-sat-ground station bounces and multi-hop sats which can act as hubs. The latter type is crucial for all those connections which will be 1000s of km away from the nearest ground station.

Even with more than 10K sats and Gbits/s speeds, there is still a hard limit on the total bandwidth available to be shared among all users in a given area.

Terje

You don't!

If you care about sub-second time intervals, and still want to measure large timescales with seconds, then you should use TAI (i.e. atomic) seconds.

The TAI scale is monotonic, so the difference between TAI and UTC increases each time a leap second is added, and would decrease if we ever removed a leap second. The latter is possible, but exceedingly unlikely.

The main/only problem with using TAI is that you can't use it for exact calendar style timestamps in the future, i.e. beyond 6-12 months you don't know if there will be a new leap second added, which means that you don't know what the offset will be at that point in time. The solution here is simply "Don't do that!"

Calendar style timestamps have a perfectly good ISO standard, i.e. 2020-10-18Z15:53:15 is the current time. The only problem here is that it takes a bit more room than a time_t variable, but otoh it is much easier to compress.

Terje

First, all 64-bit OSs have changed to a 64-bit time_t which will last long enough (2.92e11 years for the signed version), that the survival of the planet is not guaranteed.

On the other hand, Network Time Protocol have been using a fixed-point 32:32 format since day 1, and that is fine since NTP is only used to exchange/measure small time differences.

On the gripping hand the real problem here is that all these UTC timestamps, with ridiculously high resolution, are ignoring the elephant in the room which is leap seconds! It makes zero sense to have fractional UTC seconds without also specifying exactly how timestamps taken during a leap event should be handled. :-(

NTP btw have wrestled with the leap second issues for many years, so that server timestamps can include a warning about an upcoming leap event.

Terje

I made a web page many years ago which allows you to search for any given decimal digit sequence within the first 1E9 digits of Pi (3.1415926...), the algorithm I came up with is exactly the same: I knew that Pi contains an effectively random sequence of digits, so I could guess that in a sorted array of all possible starting positions, a sequence like 18790314 (Albert Einstein's birth date) should occur approximately 18.79% into the index array:

https://tmsw.no/pi-search/

Find 18790314
Found at 18,830,019: 636518371058586 18790314 193020374710719

Total time = 0.001965 seconds (9 suffix lookups)

With a naive binary search I would have needed ~30 iterations instead of the 9 taken by the interpolating algorithm.

When you consider that both the digit array (1GB) and the 4 times larger index array (4GB, split into 10 separate 400MB files) resides on disk at my web host, the response time is quite good.

Terje

Back in the seventies IBM had already patented a special form of memory which would sort in "zero time", the main problem and the reason it was never implemented was that it was limited by the channel speed, i.e. how fast you could stream data into the special chip and our again, and the fact that the size of the chip determined how much data you could sort in a single go. In this way it was somewhat similar to quadsort's later stages.

It worked by having a ladder of key storage locations, with a comparator between each pair of keys:

As you streamed data into the top of the chip, the topmost comparator would select the larger value to be pushed down to the layer below and this slot would receive the incoming value.

As soon as you had loaded all the data you would run this in reverse, but now the earlier/smaller value would exit, so you got back a stream of sorted data.

The obvious Achilles' heel is the fact that the width of the ladder items determined the maximum size of the key that could be compared, while the depth defined the amount you could sort in one go: In reality it is almost always faster to simply use the CPU with quicksort or another cache-friendly algorithm, and you don't incur the cost of a really quite expensive coprocessor unit.

Terje

AFAIK, US intelligence agencies are required to help US companies any way they can, including handing over industrial espionage results, both targeted and incidentally collected.

As a Norwegian reserve officer I'm not terribly worried by this, but it still smells somewhat bad.

Terje

Here in Norway we have a Supreme Court decision that non-moving parts in consumer goods have to last at least 5 years, so I really hope my chip wears out before next spring when that 5-year period ends. Otherwise my best option is probably to dismount the board, send it off to Holland and have the chip replaced with something far larger (about $350), before it actually fails, since at that point my only option would be to pay Tesla for a full board replacement (about $2000).

Terje

I'm a competitive orienteer, running a lot of competitions in a normal year. My record was from 2018 when I completed 105 races, with an average running time of about 40 min. I've won my class in the national (Norwegian) champs a few times but never been close to the podium in the World Masters Orienteering Championship.

I believe the main problem is that I'm normally 10% below the reference hgb level, even though I take iron supplements all the time, so when they remove a (british) pint of blood I drop below what I need for running cross-country in tough terrains.

Terje

(Full disclosure: I have been a blood donor for 45 years, since the day I turned 18 and was allowed to volunteer.)

For most of this time I have been donating whole blood (A-, Kell- and a few other factors which are less important), but for nearly 10 years I donated plasma instead: The main difference is that you can donate blood up to 4 times a year but plasma is fine once a month, and from this article it seems like replacing your plasma volume every year would be a good idea.

Personally, the only effect I have noticed is that I'm significantly out of shape for 2-4 weeks after each donation, until my hgb levels are back up, but if this article can get a lot more people to donate, that would be a _very_ good result. :-)

Terje

This is _almost_ true: I took HCQ prophylactially before, during and after a trip to Africa 20+ years ago, and the side effects did include visual distortions. The locals saw my prescription and were amazed: "That stuff is _strong_, we only ever use it when/if we get acute malaria."

Terje

I was very much a part of this transition, to the point where most of my coworkers were ordered to be at work on New Year's eve:

It went just about as well as I was hoping it would do, specifically due to the large amounts of money/resources that was spent in order to clean up old code & systems.

With all that work, the remaining issues that we hadn't found were few enough, and sufficiently non-critical that nothing really bad happened, i.e. this was a great success.

In hindsight you _could_ argue that we could have gotten away with spending a bit less up front and instead had more errors to fix after the fact, but the entire effort was basically an insurance policy which paid off.

Terje

I have personal experience of this: Reading whole words or (even better) multiple words as a single image is far more efficient than spelling your way through a bunch of letters, that was something I learned while teaching myself speed reading around 8th grade. However, I could still fall back on reading smaller groups of letters to figure out a totally new word.

It took me a while to realize that someone with severe dyslexia simply cannot do this, their only option is to read entire words as an image, and this means a lot of effort over multiple years in order to become fluent. If you combine dyslexia with near-photographic memory, then it is possible to overcome the issue and in fact become fluent in multiple languages.

Terje

You are at least partly right, there are less (percentage wise) really driven people in Norway than in the Silicon Valley tech hub, but otoh that is probably true for most of the world. The two companies most of you have heard about are probably Opera and Funcom, but we do have a few more, many of them related to oil exploration. I am the CTO of Open iT, a small international software company that was started 20+ years ago here in Norway, but the founder have moved to Houston to be closer to many of the oil companies who are our customers (along with most of the airplane and car manufacturers of the world). We have a large development group in the Philippines as well as the original Oslo office where I work.

Terje