46 half precision (16 bit) TFLOPS is nice for neural networks but I doubt you could run AGI with only 16 or 24 GB of RAM. If you could upgrade each board with 1 TB of SSD and put together 500 or so boards then maybe you could run a human brain sized neural network for about $1 million and 200 KW of electricity.
That price is almost competitive with human labor, but I doubt that Moore's law will help much more. They are already pushing the physical limits of transistor sizes. 7 nm is 63 silicon atoms. Clock speeds are no faster than in 2010. On Wed, Oct 28, 2020, 6:23 PM <[email protected]> wrote: > ...so.... > How faster can it run GPT-2? If you can't show us, we'll never know, these > numbers can mean something different than when used in practice no? We need > a test. You can say 600 or 99999 all you want, but it doesn't tell me much. > How faster is todays computers? > *Artificial General Intelligence List <https://agi.topicbox.com/latest>* > / AGI / see discussions <https://agi.topicbox.com/groups/agi> + > participants <https://agi.topicbox.com/groups/agi/members> + delivery > options <https://agi.topicbox.com/groups/agi/subscription> Permalink > <https://agi.topicbox.com/groups/agi/T079f29fb2fb0c97c-Mb67871ee394810856a3026ba> > ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T079f29fb2fb0c97c-M63d4e97bc5c41fd50338a791 Delivery options: https://agi.topicbox.com/groups/agi/subscription
