• 0 Posts
  • 131 Comments
Joined 3 years ago
cake
Cake day: July 1st, 2023

help-circle









  • I like to subscribe to the “magic makes it’s users imbeciles” fan theory. (Though the truth is that JK just isn’t all that bright).

    It isn’t that the killing spell is unblockable. Harry and his mom managed to block it twice. But apparently magicians in HP universe are just completely dumb and unwilling or incapable of innovation. That was spelled out clearly in book 1 where an elementary logic puzzle was seen a good way to protect the greatest treasure on earth.

    Ron’s dad, for example, lived in England. He could wander the muggle streets freely if he wanted to. He had a deep fascination of basic muggle items, yet he didn’t just go to his local library to check out a book or log on to the internet to learn about things that were his passion.





  • Silly commenter.

    L1 cache shouldn’t be large. Increasing the size of the L1 cache increases the latency. Maybe if you shrink the size of the cloths you wear you can squeeze more into the chair, but the ideal L1 cache has to minimize it’s distance from processing. Oversizing adds latency.

    Your L2 cache is where you generally try and shove a much bigger cache into it, but it’s still got a size constraint for the latency you are after. Further, typically L1 and L2 only serve 1 CPU. To multi-process stuff you’ll typically need an even larger L3 cache which is shared among cores.

    So the cloths on your chair should be minimal for fast access (L1). You can put more cloths on your bed and dressers or in laundry baskets that can be promoted to the chair if you start needing them more often (L2). You can throw a bunch of cloths into a pile in the corner which sit there for a few years and serve many occasions (L3).

    The worst thing is going back to main memory (your closet) to search for specialty cloths you are ultimately going to need to send back to the closet. And heavy help you if you have to swap (do laundry).


  • Given CXMT doesn’t just up their prices after gaining market share too.

    They might eventually, but certainly not immediately. In the process they are going to force the other big players to lower their prices to compete.

    Also China can just ban the export, or tariff it.

    I don’t think China has ever banned an export. It’s pretty rare to tariff an export, basically only happens when it’s a limited good that the government wants to ensure a local supply of. I think the only country I’ve heard of doing that is Greenland due to some british wankery.


  • Maybe. I mean at least a major part of it is that AI pays a lot more money than the consumer does, and even if the consumer market pulls back, they are banking on the raised prices to stay around even if they can increase capacity.

    I think what scares them is that CXMT is rapidly catching up to the state of the art. If they dick around for too long, they run the real risk that China and CXMT will do what China does and sweep the market with really cheap memory they can’t compete with.

    These companies still care about non-ai servers, and that’s a big part of the market that could be obliterate pretty quickly.




  • Hey, can we stop calling everything with a computer “AI”? Order management systems have been a thing long before LLMs were invented (I’ve worked on one). This was perhaps one of the first applications of computing. Humans hand writing an order form in a major grocery store hasn’t been a thing since like the 80s.

    Also, I’m like 80% sure this article was barfed out by an LLM. The em-dashes be everywhere.