exsf0859

That's simply not true. SV was built in the 60's thru the 80's long before the H1-B visa program started in 1990.

The problem is that with most GC algorithms 1% of the frames take 2x (or even 20x) as long, not that individual frame times vary by 1%.

Switch and PS 4 are easier to support than Linux or MacOS because there's only a few hardware variations for Switch and PS 4.

This was canceled at the point where Apple needed to commit to spending a lot of money to implement the design in silicon. I heard it was canceled because upcoming COTS processors looked like they are going to be faster and cheaper. (Source: I worked in a different part of Apple ATG and was talking with some of the people on the project.)

Ouch, you're right! I misunderstood.

I read this article as saying, "hey we can emulate bfloat16 pretty well in software on top of our existing hardware features". That's what "graft" and "minimal impact" mean to me.

Intel (for better or worse) takes a very experiment-results-driven approach to choosing which features to implement in hardware. So this result -- that software emulation of a feature works almost as well as a hardware implementation would -- probably makes Intel less likely to implement the feature in hardware.

ARM, AMD, RISCV etc, will probably come to similar conclusions.

It's called brain floating point because it was developed by Google's "Google Brain" machine learning project.

bfloat16 is "best" primarily because it has the same exponent range as float32. That makes it easy to port models that have been developed using float32 to bfloat16. (As opposed to using int8 or float16, both of which have a smaller exponent range.)

It's possible that some other custom format is better in absolute terms for models trained specifically for the custom format. But for the current ecosystem, where models are trained primarily using float32, bfloat16 is a very good choice.

If I recall correctly, the 2001 ICFP contest was won by someone who happened to have done their thesis on the contest problem. :-)

If you like Gassee's writing about Be, there's plenty more available in his articles for the Be Newsletter:

https://www.haiku-os.org/legacy-docs/benewsletter/index.html

I think the main reason for the QR code is to discourage third-party juice packs.

Two Minute Papers https://www.youtube.com/user/keeroyz - enthusiastic summaries of machine learning papers.

Pachter Factor on SIFTD Games channel https://www.youtube.com/channel/UCFaBUWXO8o4jSfF6PhJOyZA Discusses the game business.