A 15GB model loaded fully onto the GPU and generated at 5.6 tokens per second. The theoretical ceiling was 8. Capacity and throughput are set by different resources, and the obvious fix for a tight fit — shrink the model — barely moves the one that matters. One division tells you which lever works.
Originally published on hexisteme notes.
"Will it fit on our hardware?" is the wrong first question. It's the one everyone asks, because
it's free to answer — the thing either loads or it doesn't.
Throughput costs you a measurement. So the capacity gate passes, and it feels like the
decision is made.
In 2026, developers are running 80B-parameter models on Macs and 35B models on iPhones. Discover the quantizing, pruning, and…
Which hardware should run this model? Most teams answer that question with habit instead of arithmetic. The model card says GPU.…
DeepSeek-V4-Flash has 284 billion parameters and takes up about 160GB on disk. My laptop does not...
An upper bound on CPU generation speed derived from memory bandwidth and model file size alone, with the arithmetic shown and the…
Did you know that a 35-billion-parameter model can generate tokens at the same compute cost as a 4B...
Mac Studio M4 Max vs Mac Mini M4 Pro for Local AI in 2026: Is the $600 Upgrade to 546 GB/s Worth It?
