Loading video...

Video Failed to Load

Go Home

Andrej Karpathy on Tensor Cores and TF32 precision

112,983 views • 1 year ago •via X (Twitter)

9 Comments

Tommy Falkowski's profile picture
Tommy Falkowski1 year ago

He's an awesome teacher! I just love his clean style of explaining things!

ℏεsam's profile picture
ℏεsam1 year ago

he really has it in him

Konstantin's profile picture
Konstantin1 year ago

Can you share a full source please?

rogue node's profile picture
rogue node1 year ago

So basically TF32 is a lie.

Jacob Posel's profile picture
Jacob Posel1 year ago

Where does he post this?

john's profile picture
john1 year ago

Lmao what are the odds that the moment I get to this part of the video on my first time watch, i see this on twitter right after

ℏεsam's profile picture
ℏεsam1 year ago

i'm watching you :)

Felix Chin MD's profile picture
Felix Chin MD1 year ago

TF32 just means ur neural net’s matmuls are running on 18-bit numerics (FP 18 precision) apple (mlx) easily lets u control this (by setting intermediate weights and values to FP 16) don’t see how TF32 is anything special. seems like a marketing gimmick

MarcoDagna e/acc 🇪🇸 ⌛️🌍's profile picture
MarcoDagna e/acc 🇪🇸 ⌛️🌍1 year ago

@HeyGenLabs english spanish

Related Videos