Toshiba Touts Algorithm That’s Faster Than a Supercomputer

My understanding is that it's an ASIC for a modified simulated annealing algorithm, not a supercomputer. OTOH simulated annealing doesn't hugely benefit from parallelism, so the comparison is a bit of a snore.

Edit: my mistake, that's a different Toshiba offering. Sibling comment has a better description and a citation.

It's more a QUBO solver (ala D-Wave) that can be efficiently parallelized on GPUs, they were able to "solve" a QUBO with 100K variables on 8 GPUs [1], where before people had solved at most problems with 10K variables.

[1] https://phys.org/news/2019-04-toshiba-breakthrough-algorithm...

Yes, the title makes no sense.

Like "this recipe is better than a gas stove", "this movie is more enjoyable than a TV" or "this key is more secure than a safe". We are not comparing the same thing.

Technically you could say that about the entire field of machine-learning. That doesn't mean it isn't hugely useful.

I think you're imputing more meaning to their article than it possesses.

Just let the nice word-noise wash over you, and acknowledge that Bloomberg publishes lightly-edited press releases.

"Category error" is the term for this, I believe.