VRAM Calculator
Calculate densely stored model-weight bytes and GiB from parameter count and precision, then add an explicitly entered memory allowance. No GPU performance prediction.
Interactive calculator · Method VRAM-2.1 · Last reviewed: 2026-11-04
Calculate your scenario
Stored weight bytes = billion parameters × 10⁹ × bits/8. GiB = bytes/2³⁰. Add only the additional memory allowance entered.
Your inputs
Default worked example. Edit the inputs to calculate your scenario.
Default worked example
7 billion parameters × 16 bits ÷ 8 = 14,000,000,000 bytes of densely stored weights. Divide by 2³⁰ for GiB and add only the extra memory you entered.
How to use the calculator
Enter the model size in billions of parameters and a stored precision of 4, 8, 16 or 32 bits per parameter. Enter additional memory explicitly in binary GiB; zero means no additional memory is included in this budget.
What the result means
The budget totals only dense weights and the extra memory you entered. A tiny positive allowance remains positive in scientific notation. Stored precision is a bit budget and does not by itself model a particular quantization format or runtime.
Formula and units
Theoretical dense-weight bytes = billion parameters × 10^9 × stored bits / 8. Weight GiB = those bytes / 2^30. Entered model memory budget = weight GiB + additional GiB. The byte card rounds the theoretical byte equivalent; exact entered values are retained before GiB display rounding.
Worked examples
Model parameters: 7 billion parameters; Additional entered memory allowance: 2 GiB: Model parameters: 7 billion parameters; Additional entered memory allowance: 2 GiB; Stored weight precision: 16.
Example result: Entered model memory budget: 15.0385 GiB; Model weight storage only: 13.0385 GiB; Entered additional memory: 2 GiB; Stored precision: 16 bits/parameter; Weight storage bytes: 14,000,000,000 bytes.
Model parameters: 10.5 billion parameters; Additional entered memory allowance: 3 GiB: Model parameters: 10.5 billion parameters; Additional entered memory allowance: 3 GiB; Stored weight precision: 32.
Example result: Entered model memory budget: 42.1155 GiB; Model weight storage only: 39.1155 GiB; Entered additional memory: 3 GiB; Stored precision: 32 bits/parameter; Weight storage bytes: 42,000,000,000 bytes.
Limitations and common mistakes
Quantization metadata, padding, KV cache, activations, batches, workspace, runtime, sharding and device allocation can add memory. A theoretical byte equivalent is not a measured allocated buffer, a fit verdict, a GPU recommendation or a gaming-performance estimate.
Keep billions of parameters distinct from individual parameters, bits distinct from bytes, and decimal GB distinct from binary GiB. Do not count unentered runtime memory as included.
Sources and assumptions
Bits, bytes and binary units validate storage arithmetic only. KV cache, metadata, runtime workspace, activations and performance are not inferred.
Method VRAM-2.1; last formula review 2026-11-04. Sources and their supported claims. Calculation methodology.
NIST byte, bit and binary-prefix definitions support the declared unit conversions.