Apple is attaching a dual neural engine from the m6, but it's a shame about the memory bandwidth.
Everyone has been waiting for the m5u Mac Studio.
Currently, with the highest specs, 256g RAM and a 4t SSD, it costs around 21 million won.
If you upgrade the chip, a 32-core neural engine is attached. The RAM bandwidth is 1.2t.
Looking at the specs, it's a workstation, right? But strangely, it seems a bit lacking in LLMs??? The speed isn't as fast compared to CUDA.
Still, it's an amazing spec that can run deepseek-v4-flash quantization 4-bit version alone on a single device.
If I had won the lottery... I would have bought it right away and tried it out.
I want to buy three and try clustering them to run glm5.2 at home.
