Advertisement · 728 × 90

Posts by sophia (reality toucher)

i am nigerian clairo

1 year ago 0 0 0 0
Post image
1 year ago 0 0 1 0

wtf is up with the kids these days

1 year ago 1 0 0 0

totally insane

1 year ago 2 0 0 0

have to agree with that princess!

1 year ago 4 1 0 0

major blackpill on me is that i say "ilya" far more often than "ily"

1 year ago 2 0 0 0
Preview
Albert Stevens - Wikipedia

they injected 18 people with plutonium in the 40s, but none of them died from it en.wikipedia.org/wiki/Albert_...

1 year ago 0 0 0 0

I feel it in my fingers, I feel it in my toes.
Truth is all around me, and so the feeling grows.
It's written in the wind, it's everywhere I go.
So if you want to know, come on and let it show.

1 year ago 0 0 0 0

you need love so badly you would betray your own people

1 year ago 0 0 0 0
Post image

forced reconstruction erotica -- "we'll wash your dirty musket into a satin sheath" "let's take that mud-stained shell of yours and reconstruct something beautiful" "we'll watch your rebel yells turn into whispered pleading"

1 year ago 1 0 1 0
Advertisement

approximately 0.5% of the GDP of the Dominican Republic is MLB player salaries

1 year ago 1 0 0 0

racist guy who loves yiddish words because "it's a return to our anglo-saxon roots... none of this effete latin"

1 year ago 0 0 0 0

looks great!!

2 years ago 1 0 0 0

i think chrome tried something like this, privacy sandbox maybe?

2 years ago 3 0 1 0

wonder what yelp’s compute spend is. maybe there’s some “law of data analysis” that its worth spending 5% of revenue on compute for data analysis for any revenue

2 years ago 1 0 1 0

the problem being that with the cuda matmul api you can’t access the fragments without reverse engineering

2 years ago 1 0 0 0

also is there a better development environment than godbolt for looking at a bunch of CUDA and corresponding assembly at the same time?

2 years ago 2 0 0 0

it’s unfortunate that the only API for doing matmuls on CUDA capable of implementing eg FlashAttention or Swiglu efficiently is just emitting raw assembly

2 years ago 3 0 1 0

i wish cuda wasn’t totally broken

2 years ago 2 0 1 0

thinking about layouts in my head over and over again has made me appreciate triton a thousand times more

2 years ago 3 0 0 0
Advertisement

common occurrence

2 years ago 1 0 0 0

somehow i didnt realize cuda had all the good old weird integer things like clz

2 years ago 3 0 0 0

not sure what i would do — probably semi-open nicely decorated space to work and read and talk in. dont have a ton of great ideas. what would you do?

3 years ago 1 0 1 0

yes. i could do it now, but im too resistant to spending money atm and dont feel like i have the energy.

3 years ago 1 0 1 0

my only meaningful thought is something something tax loss harvesting

3 years ago 1 0 0 0

oh mostly thinking about bluesky as a platform

3 years ago 2 0 0 0

legally restricted from re-jekking on new platform

3 years ago 0 0 0 0
Advertisement

to be clear i am aware that dram uses fewer transistors per bit than sram but logic transistors are significantly more expensive than dram transistors i think more or less for the same reasons

3 years ago 3 0 0 0

“network wires are cheaper than membw wires for the same reason dram transistors is cheaper than sram: they’re off the logic die” is this true?

3 years ago 5 0 1 0

@mcint.io

3 years ago 5 0 0 0