Can someone post the full article. Its behind a paywall.

Apple, as we all know, does "value" based pricing, not cost-up. And value = brand, apps, ease of use, etc. They are not optimizing for folks who want upgradable RAM, etc. From apple's perspective, those folks can buy the high-end MacBooks ... And $40B in revenue per year suggests their product strategy is working.

Here is a cost based analysis of building ARM-chips themselves versus buying from Intel. Net-net, they save $2B + (and this analysis does not include the RAM costs) https://medium.com/@sumitg_168...

They trained on ~5000 people and inferenced on ~2500 people. The sample size is too small and not clear if they had enough clarity on how much corner cases there are. Full technical paper is here: https://ieeexplore.ieee.org/st... (have not read it fully yet)

on the contrary, this is all about focusing on a few things (software, systems, solutions) and spinning off the data center / infra management business

Since most passenger cars are used for short distances per day (a very small percentage of people drive more than 100-150 miles a day), this kind of breakthrough will be most useful for long haul vehicles like trucks and buses. And yes, if you are doing a long distance trip. But likely you will take a ~45 min break, in which you can recharge your car (even partially). The way to use this tech would be if you need fewer batteries (less weight) to get a 250 mile range in a car, because of improved efficiency.

Wonder why there is a link to Cringely's predictions at the bottom of the post? Cringely's predictions are so wrong, that they do not even get the business units at IBM right. That person didn't even bother to read anything about IBM. And all of Cringely's predictions about IBM and many other things have come out to be completely wrong. Clearly, just a writing negative IBM articles as click-bait.

Just like the movie Men in Black showed in the end, there are monster size aliens playing with planets as if they are rocks or marbles

Selling chips to phone makers is different from selling phones to consumers via telcos. This article highlights the stack fallacy https://techcrunch.com/2016/01...

Cringely hates IBM and constantly berates the company. His past predictions have turned out to be incorrect and wishful thinking that somehow IBM will die. Examples: Oct 2018: https://www.cringely.com/2018/... "Look for a huge shuffle in organization structures, reporting, departments, experience, and the current seven or eight tiers of management to be replaced with three or four. Lots of old hands at IBM — the bulk of the current organization — will go." Feb 2018: https://www.cringely.com/2018/... "Ginni will be replaced as CEO this year" Jan 2017: https://www.cringely.com/2017/... "The most important 2017 event for IBM will be the retirement at 60 of CEO Ginni Rometty. " March 2016: https://www.cringely.com/2016/... "IBM is dying" (revenue has been flat and the business is transferring from old businesses to new businesses - takes a lot of time to do this)

quick read -- this is not really an all-reduce communication / distributed deep learning innovation. its a clever way to do classify when you have tons of classes (good innovation)

btw, if your chip is small, you will get 250 net good die, so cost will be $20 per chip if your chip is complex, you need a large verification team + costs

C to HDL compilers are inefficient. Worth trying on something called a FPGA (its a flexible hardware logic that can be molded to pretty much any algorithm). The way to think about costs is: - Team to convert C to HDL manually: ~12 people for something reasonably complex - Optimize & map to a hardware architecture: ~6 people - Design tools: ~$1M - Chip tapeout costs: $1-5M depending on if you want to use the latest or old tech - If there is any software work that goes on top of your chip (because you have a CPU), then add that team too Per chip costs: - Each wafer costs ~$5000 - Lets say you get ~50 net good die, then its $100 per chip I suggest talking to a chip design consulting company & getting a quote

The Terminator storyline has an amazing amount of scope for innovation. In particular, how they have now introduced alternative timelines. Can't wait

The $1000 price tag is so ridiculous, that naturally everyone will hesitate to upgrade. Battery life is good enough Camera is good enough Now its all about damage - still $800 cheaper to fix broken screen than buy new phone.