4 ms·
>> inference kernels produced incorrect results when used in certain GPU configurations. It seems reasonable to assume that GPT inference is done entirely on N
by noduerme 3y ago
>> inference kernels produced incorrect results when used in certain GPU configurations.
It seems reasonable to assume that GPT inference is done entirely on Nvidia GPUs. I wonder if this is a subtle clue that they're experimenting with getting it to run on competing hardware.
- ShamelessC 3y agoWhy would they subtly hint at anything they’re trying to keep secret?
- noduerme 3y agoI don't know. Maybe they'd like to drive down the price of Nvidia so they can buy it.
- Rastonbury 3y agoNow I know why Altman said he needs 7 trillion, 2 trillion is for Nvidia