cpsro

White POTUS on last night's SNL was spot on.

The Russian asset in the Oval Office has an established mechanism for bribing him to reduce any tariffs of interest to you: his $TRUMP cryptocurrency that already made him billions of dollars just before inauguration. Attorney General Pam Bondi made sure your nefarious act won't be discovered, as within hours of being sworn into office, she cancelled federal investigative programs into foreign corruption and unregistered foreign agents.

What's the largest, full precision Deepseek r-1 model that will run on a 512GB Studio? And how does it compare in speed and results to the reduced precision 670b parameter Q4-K-M model?

For general purpose computing, the M4 Max is barely faster than the M3 Max, and the M3 Ultra is looking like it's only a hair faster than the M4 Max. The next significant speed bump will likely arrive with the M5 family which should be implemented in the next generation ASML process.

Where does 2971 single-core score come from for the M3 Max? Geekbench.com indicates over 3100 for the various Mac platforms with M3 Max... and that's an average of a lot of results that includes some people who didn't run a careful benchmark. My own M3 Max results are over 3200.
As for multi-core, Geekbench.com reports ~20,900 average, while my careful test results were over 21,600.