3,232 karma · joined June 27, 2014
kevmo314@gmail.com
blog.kevmo314.com
[ my public key: https://keybase.io/kevmo314; my proof: https://keybase.io/kevmo314/sigs/jKuWflZVNiTzhIe0ttNcauvsm_uNCKaCk5U7Ni6SuFk ]
I bought a 2019 Intel MBP and that was by far the worst laptop I've ever had. After just a year of use it was constantly overheating and running out of memory and disk space, barely able to open a terminal. It was so bad that I hesitated to buy the Apple silicon versions, but the good reviews convinced me and it has been going strong ever since.
That matches my personal experience too, writing naive cuda code that doesn’t take advantage of parallelism is roughly half the speed of running it on cpu.
Sure, if you have that then of course it would be fast. But that’s not what this library is proposing.
I use this to access UVC devices correspondingly without cgo: https://github.com/kevmo314/go-uvc
> The AXOL1TL V5 architecture comprises a VICReg-trained feature extractor stacked on top of a VAE.
> It's wild to think that 4096 bytes are sufficient to play chess on a level beyond anything humans ever achieved.
According to TCEC the time control is 30 mins + 3 sec, that’s a lot of compute!