Modern technology in nostalgic hardware: AI in Pentium II

Forum

Modern technology in nostalgic hardware: AI in Pentium II

By admin, Ocak 1, 2025

Modern technology in nostalgic hardware: AI in Pentium II





Modern technology in nostalgic hardware: AI in Pentium II



See Full Size


Consisting of researchers and engineers from Oxford University EXO Labs team, 1998 An old one released in Windows 98 managed to run a powerful language model on the computer. And just 128MB RAMIt has a 350MHz speed Pentium II with processor!

EXO Labs shared this experience with a video on the social media platform X. In the video, it is installed on a dusty Elonex Pentium II system. Llama2.c It was observed that a model based on the model successfully fulfilled the command to create a story. Moreover, it showed a satisfactory performance in terms of speed.

Success is big, goal is different

One of the biggest challenges the team faced was compiling and running a modern AI model on a 1998 operating system. However, overcoming these difficulties 260,000 parameters a LLaMA model 39.31 tokens per second They were able to run it at processing speed. There was a loss of performance in larger models; for example, a model with 1 billion parameters was only able to achieve a speed of 0.0093 tokens per second.




Modern technology in nostalgic hardware: AI in Pentium II



See Full Size


So, what was the purpose of this experiment? EXO Labs makes it possible to run artificial intelligence models on even the most modest hardware. aims to universalize technology. Today, AI technology often depends on powerful server infrastructures and expensive GPUs. However, EXO Labs is working to turn this situation around.

EXO Labs, “BitNetThey are working on a transformer architecture called “. This technology uses ternary weights to reduce the model size. and a model with 7 billion parameters in just 1.38GB of storage. makes it possible to operate. What’s even more impressive is that BitNet is designed to run on CPU only. This architecture can run a model with 100 billion parameters on a single CPU at a processing speed of 5-7 tokens per second.

























Etiketler: , , , ,