pcdeni

4 ms·

1. 1.▲
   by pcdeni 2mo ago

   AI is helpful to understand the world around us. And also to do the posting while I focus on the doing. Although some of these comments don't require much while perplexingly complain about slop... Are you maybe using it wrong?
2. 2.▲
   by pcdeni 2mo ago

   it is not faster compared to a GPU. The prototype works end to end with off the shelf memory, to make it "real" silicon level changes would be needed for the memory.
3. 3.▲
   by pcdeni 2mo ago

   yes, trust me, it does not get any more simple.
4. 4.▲
   by pcdeni 2mo ago

   This is a first, may I ask where it comes from? You might have a line of thought but I cannot tell from here. I am using off the shelf memory, adjusting the voltage didn't come up.
5. 5.▲
   Show HN: Avoiding the Memory Wall by computing LLM inference directly inside RAM

   3 points by pcdeni 2mo ago|0 comments
6. 6.▲
   Show HN: Running PrismML's Bonsai inside DRAM by breaking DDR4 timing rules

   23 points by pcdeni 2mo ago|10 comments
7. 7.▲
   Show HN: Running BitNet b1.58 inside DRAM by breaking DDR4 timing rules

   6 points by pcdeni 4mo ago|0 comments