Buyer's guide / AI PCs
LLM VRAM calculator
Enter a model size, quantization and context length. Get an estimate of the memory needed to run it locally — and see exactly which 2026 machines can hold it.
By Evan Cole · Last updated July 17, 2026
Whether a model runs on your hardware comes down to one number: the memory it needs to load. This calculator estimates it from the model's parameter count, your quantization, and your context length — then tells you which 2026 machines can hold it. For the full explanation of the maths and the hardware, see the how to run LLMs locally guide.
How the estimate works
Weights dominate the footprint: parameters × bytes-per-weight (FP16 = 2.0, Q8 ≈ 1.06, Q4 ≈ 0.55). On top we add the KV cache, which grows with context length, plus ~1.5 GB runtime overhead. Unified-memory machines can't give 100% of their RAM to the model — the calculator uses a conservative usable figure per machine. It's a sizing guide, not a guarantee; real usage varies with the runtime and model architecture. The full breakdown is in the hardware guide, and head-to-head numbers are in the AI PC benchmark.
Frequently asked questions
How is the memory estimate calculated?
Weights = parameters x bytes-per-weight (FP16 = 2.0, Q8 ~= 1.06, Q4 ~= 0.55). On top we add the KV cache (which grows with context length) and ~1.5 GB of runtime overhead. The KV cache is estimated from a per-token cost that scales with model size — it's an approximation, so treat the total as a sizing guide, not a guarantee.
Why does long context need so much more memory?
The KV cache stores the attention state for every token in the context window, so it grows roughly linearly with context length. For a 70B model it can go from ~2 GB at 8K tokens to tens of GB at 128K tokens — at the extreme it rivals the weights. Quantizing the KV cache to 8-bit roughly halves it.
Can I really use all of a machine's memory for the model?
Not quite. Unified-memory machines (Apple, GB10, Ryzen AI Max) reserve some memory for the OS — Apple defaults to ~75% usable for the GPU, though it's tunable. Discrete GPUs give almost all their VRAM to the model. This calculator uses a conservative usable figure per machine.
Sources
The bytes-per-weight and KV-cache math follow the standard VRAM rule-of-thumb (see this VRAM explainer); the KV-cache anchor (~0.3 MB/token for a 70B FP16 model) and the ~75% Apple usable-memory default are drawn from the same technical sources cited in our hardware guide. Machine memory capacities are manufacturer-confirmed. Estimates are approximate and provided as a sizing guide.
Related buyer's guides
All reviews →Buyer's guide / AI PCs
How to run LLMs locally: the 2026 hardware guide
How much VRAM or unified memory you need to run 7B-120B models locally, the capacity-vs-bandwidth tradeoff, and the 2026 hardware that hits each point on the curve — GB10, Ryzen AI Max, Apple Silicon and RTX GPUs.
Read the guide →
Buyer's guide / AI PCs
Mac vs PC for running local LLMs (2026)
It's not Mac vs PC — it's capacity vs speed. A Mac's unified memory fits models a PC GPU can't hold; a PC GPU runs the models that fit far faster. A sourced head-to-head plus which to buy.
Read the comparison →
Buyer's guide / AI PCs
The best AI PC for running LLMs locally in 2026
Six machines for running large language models on your own hardware — from a $799 Mac mini on-ramp to an NVIDIA GB10 supercomputer — ranked by what actually matters: unified-memory capacity and bandwidth, not NPU TOPS.
6 picks →
Buyer's guide / AI PCs
Best CPU for AI and local LLMs (2026)
The honest version: for pure GPU inference the CPU barely matters — so this guide is organised by when it does. Consumer value, the Intel option, unified-memory APUs that hold models a GPU can't, and the workstation platform you only need for multi-GPU rigs.
6 picks →
Benchmark / AI PCs
AI PC benchmark: local-LLM speed and memory (2026)
How the 2026 AI PCs compare for running local LLMs: unified memory, memory bandwidth, NPU TOPS, the biggest model each can hold, and measured tokens per second.
Read the benchmark →
Buyer's guide / PC builds
Best gaming PC under $1,500 in 2026
The honest $1,500 build: a 16GB RX 9060 XT, Ryzen 5 and 32GB DDR5 for strong 1080p-ultra and entry-1440p gaming — with the real shortage-adjusted total, where every dollar goes, and whether a prebuilt beats it this year.
See the build →