2020-08-25
6 indexed posts · All posts for this date, newest first.
@waitbutwhy Now I want candy
View post →Captured source text.
@dauqhx @universal_sci There’s a lot of frozen CO2 & H2O on Mars. Heating the planet will densify the atmosphere. It’s solvable.
View post →Captured source text.
@universal_sci Perhaps one day there will be an ocean on the northern part of Mars
View post →Captured source text.
@jchybow @jwangARK @wintonARK It’s almost working reliably
View post →Captured source text.
@jwangARK @wintonARK Very different computers are needed for training vs inference. FSD is mostly Int8, but Dojo is de facto FP32 (feels like FP32, but actually FP16 with boring bits truncated) so we don’t need to rewrite all the training software. It should work almost immediately.
View post →Captured source text.
@jwangARK @wintonARK Not exactly GPUs for NN. Lots of cores optimized for dot products with fast local memory & super high bandwidth between cores & efficient liquid cooling. Compiler needs to be optimized too.
View post →Captured source text.