Huh, consumer GPUs are doing Petaflops of floating point. FP64 isn't a useful co... | Hacker News

Hacker Newsnew | past | comments | ask | show | jobs | submit

		layla5alive 6 months ago \| parent \| context \| favorite \| on: Top researchers leave Intel to build startup with ... Huh, consumer GPUs are doing Petaflops of floating point. FP64 isn't a useful comparison because FP64 is nerfed on consumer GPUs.

kvemkon 6 months ago [–]

Even recent nVidia 5090 has 104.75 TFLOPS FP32.

It's useful comparison in terms of achievable performance per transistor count.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact