Daily Tech News 15 December 2022

• ChatGPT wrote a terrible Gizmodo article. (Gizmodo)

  Suffice it to say, multiple attempts brought less than satisfactory results. For one thing, ChatGPT kept including incorrect information in its explainer—sometimes mixing up basic facts about the history of its own technology (factual inaccuracy has been an ongoing problem for the program). In other instances, it skimped on the details in critical areas, writing text that read more like OpenAI ad copy than a thoughtful explainer. In at least one instance, it forgot its own name, repeatedly telling me that it didn't know anything about a "ChatGPT" and thus couldn't help with my request. In general, it struggled to find a good balance of factual information, story structure, and accessible, humorous language, when putting its stories together.

  The comments are exactly what you would expect:

  The joke is so obvious, I'm not even going to bother.

  
  
  
• Looks like all my Amazon packages that I was hoping would arrive by Christmas, will.
  
  One seems to have taken three days by plane to travel from the US to Australia, which is odd. But given that the expected delivery date was January 11, the fact that it has already arrived in Sydney and left again for regional NSW is promising.
  
  
  

• Samsung has used PIM to speed up AMD's Instinct MI100 GPUs by 150% while reducing power consumption by 60%. (Tom's Hardware)
  
  PIM stands for Processor in Memory - adding compute logic directly into the memory chips. That means data doesn't have to be read from those chips, processed, and written back; everything happens in one place.
  
  That does depend on all the data required for a given computation be located on the same chip, so it's not remotely a general-purpose solution. But in those cases where you can use it (like training chatbots to write terrible articles for terrible tech news sites) it can bring huge benefits.
  
  
  
• Meanwhile SRAM is dead. (Wikichip)
  
  Moving from 7nm to 5nm, logic sizes shrank by close to 50%, but SRAM sizes only shrank by 20%. Moving from 5nm to 3nm, it looks like SRAM sizes will only shrink by 5%.
  
  That's not good, because a lot of performance depends on getting data into large on-chip caches.
  
  AMD has already tackled this with its 5800X3D, which stacks a separate SRAM chip on top of the CPU, with thousands of interconnects, and with the just released Radeon 7000 range, which separates out the cache and memory controllers onto their own chips on a multi-chip module.
  
  Expect more of that in the future, and in particular more vertical stacking.
  
  
  
• The iKOOLCORE R1 is another of those weird little Chinese micro-PCs. (Liliputing)
  
  This one has four 2.5Gb Ethernet ports, aimed at the small router/firewall market. It also has a single HDMI port, a USB-C port, and two USB-A. At three inches square and two inches high, it's even smaller than a NUC.
  
  Prices start at $145 - in China.
  
  
  

posted by Pixy Misa at 04:00 AM

| Access Comments