Interactive primer · local LLM hardware

THE HARDWARE
BEHIND THE TOKENS

Everything that happens between you hitting enter and the next word appearing — TOPS, precision, VRAM, bandwidth, and why your GPU is built the way it is. Each idea unlocks the next. Play with every widget.

TOPS & precision Quantization CPU · GPU · NPU VRAM & bandwidth Will-it-run calc
01 · The atom

What "operation" even means

Before we count operations per second, we need to know what one operation is. Strip a neural network down and it's just one move repeated at unimaginable scale: multiply two numbers, add the result to a running total.

02 · Counting speed

TOPS, FLOPS & the precision asterisk

TOPS = Tera (trillion) Operations Per Second — how many of those multiply-adds a chip can do. One number, one giant caveat: it's meaningless without the data type it was measured at. 50 TOPS should always be read as 50 TOPS @ INT8.

TOPS — inference, integer math

The headline spec for NPUs and edge AI chips (Apple Neural Engine, Snapdragon, the 40+ TOPS "Copilot+ PC" bar). Almost always quoted at INT8 — halve the bit-width and the number roughly doubles. It's a theoretical peak: real workloads rarely hit it because memory can't feed the units fast enough.

FLOPS — training, floating-point

The same idea for floating-point math, the language of training. Quoted in TFLOPS / PFLOPS. Rule of thumb: training talks FLOPS, local inference talks TOPS — and, as you'll see, often talks bandwidth even more.

03 · Shrinking the numbers

Precision & quantization

Lower precision is the lever that makes local AI possible. Fewer bits per weight means a smaller model, more TOPS on the same chip, and a little less quality. This is what GGUF quant levels are.

04 · Where the work happens

CPU, GPU, Tensor cores & NPUs

All those operations have to run on something. The reason a GPU eats LLMs for breakfast comes down to one idea: massive parallelism beats raw single-core speed when the work is millions of independent multiply-adds.

05 · The real bottleneck

VRAM, bandwidth & the memory wall

Here's the plot twist that surprises most people: for local chat, your speed usually isn't capped by TOPS at all — it's capped by memory bandwidth. Every token streams the whole model out of VRAM. And VRAM capacity is a separate fight, made harder by the KV cache.

06 · Putting it together

Will it run on my rig?

Every concept so far collapses into one practical question. Model size comes from params × precision. Whether it fits is size + KV cache vs VRAM. How fast it runs is bandwidth ÷ size. Turn the dials.

07 · Make it tangible

What that speed actually feels like

The calculator hands you a tok/stok/s — tokens per second, the user-facing speed metric. Everything in this primer rolls up into this number. number, but a number is abstract. This types out at the real rate, so you can feel the gulf between a model crawling on CPU offload and one living comfortably in VRAM.

08 · The full picture

The silicon family: CPU → GPU → NPU → TPU

You've met all the pieces now — multiply-adds, precision, parallelism, bandwidth. Zoom all the way out and they assemble into a family of chips, each making a different bet on power, price, and who controls the silicon. Same job, very different homes.

09 · Reference

Glossary

Every abbreviation used in this primer, in one place. Hover or tap the dotted terms above for the same definitions inline — nothing here goes undefined.

Every abbreviation, defined
Want help choosing or building hardware for local AI? Grid City does in-home & remote builds and setup across NYC and Long Island.
Get in touch