Daily Tech News 5 September 2023

• The EU is preparing to push for a global phase out of fossil fuels at the COP28 climate summit. (Euro News)
  
  I guess LARPing at ruling the world is less effort than actually doing so.
  
  
  

• Sapphire has shown off a Radeon 7800 XT card that isn't an awful blinged-up ultraviolet catastrophe. (WCCFTech)
  
  It's a little chunky - a 2.5 slot model by the look of it - but otherwise unobjectionable.
  
  $499 unless Nvidia does something drastic in the next few hours.
  
  
  
• Performance looks to be pretty good - but definitely wait for reviews. (Tom's Hardware)
  
  The 7700 XT clobbers the 4060 Ti, and the 7800 XT is convincingly faster than the 4070, except for ray tracing benchmarks, and even there both cards actually win some of the comparisons.
  
  The one game where both the AMD cards lose badly is Cyberpunk 2077 with ray tracing set to medium or higher. At the ray tracking "overdrive" setting the game is unplayable on either of the AMD cards - but the 4070 only manages 18 fps, so while much better it's still terrible.
  
  
  
• Writing a C compiler in 500 lines of Python. (Vgel)
  
  Not a complete implementation, but a working compiler nonetheless.
  
  It targets WebAssembly rather than any native instruction set, but the author thinks that actually made it harder, because WebAssembly is a very poor fit for C.
  
  
  
• Why are Germany's wild boars radioactive? (Washington Post / MSN)
  
  Because they eat radioactive truffles.
  
  And why are the truffles radioactive?
  
  Because they grow underground and cesium from nuclear testing in the 50s and 60s is slowly leaching through the soil, where the mushrooms absorb it.
  
  Result: Glowing green eggs and ham.
  
  
  
• Invasive species cost humans $423 billion each year. (The Guardian)
  
  Close the border then.
  
  
  
• Llamas vs. chinchillas. (GitHub)
  
  I'm sure this is very informative if you know what a cosine schedule is, but the article doesn't tell you.
  
  It's about the time spent training LLMs compared against the quality of the results, and it seems that smaller models are better at every point.
  
  

posted by Pixy Misa at 04:00 AM

| Access Comments