Daily Tech News 30 January 2025

• Nvidia's RTX 5080 is here and while it is tempting to blame it for the company's $600 billion share price free fall that was probably just a jumpy and irrational stock market and not directly related to the profound ordinariness of this card. (Tom's Hardware)
  
  It's slightly faster than last year's 4080 Super, which was slightly faster than the previous year's 4080.
  
  It does best at 4K resolutions, where it's 9% faster, probably thanks to the much faster GDDR7 memory. At lower resolutions where memory bandwidth doesn't matter so much, performance gains average just 3%.
  
  And with average gains that low, yes, it is sometimes slower than the previous model. I'm not sure how, because by all the numbers it should be at least a little better.
  
  All the benchmarks seem to pit it against AMD's 7900 XTX, which comes out looking pretty good aside from ray tracing, since it is cheaper, includes 24GB of RAM compared to the 5080's 16GB, and is just as fast for anything except ray tracing.
  
  This review bizarrely suggests skipping the 5080 for the 5090, which delivers 50% better performance for twice the price. Yes, if a $1000 video card isn't fast enough to justify the price, you should spend twice as much for something that is even worse value.
  
  We're going to see a repeat of this next month when the cheaper models - the 5070 Ti and 5070 - arrive. In March we might see something interesting as AMD's new cards launch.
  
  But probably not.
  
  

• So Chinese AI platform DeepSeek reportedly ripped off OpenAI, but OpenAI ripped off everyone on the internet, so nobody cared.
  
  And it reportedly was collecting massive amounts of data from its users, but so is everyone on the internet, so nobody cared.
  
  But it was also leaking that data everywhere through misconfigured logging servers that were open to everyone on the internet. (Wiz)
  
  I don't think that part was intentional.
  
  
  
• Mark Zuckerberg says Meta is going to spend tens of billions of dollars on AI this year regardless of the DeepSeek nonsense. (Tech Crunch)
  
  The company is planning to develop - and release as almost-open source - two new models in the Llama series to compete head on with the closed source ChatGPT.
  
  (The Llama license has a provision that you can't use it for free if you're a trillion-dollar tech giant, so technically it's not open source, but most of us don't need to worry about that.)
  
  
  
• AI-created artwork can receive copyright protection if it's not entirely AI-created. (Associated Press)
  
  Which was already the case.
  
  
  
• Federal government workers unions, which rely for their funding on the number of federal government workers, are against plans to reduce the number of federal government workers. (The Register)
  
  That follows.
  
  
  

posted by Pixy Misa at 04:00 AM

| Access Comments