NodeValid

Published video-model weights, licences and hardware notes

Numeric precision, and the memory it decides

Numeric precision is the format each weight is stored and computed in. Halving it roughly halves the weights in memory, and it also decides whether a particular card supports the run at all. As of 2026-09-22.

Why a memory figure needs its formatA floor stated in one format does not apply in another, and a card that lacks hardware support for the format cannot reach the figure by having enough memory.Full precisionFour bytes per weight,rarely publishedhalve itHalf precisionThe usual publishedconfigurationhalve it againEight-bitA separate build, oncards that support itMemory and eligibility at onceOne repository here names the formats and the card generations together.
Fig. 1 The same weights at three formats, and the card requirement moving with each.
What the repositories actually stateFilled where the repository states something, hollow where it does not. A blank is not an inference.What the repositories actually stateDirectionWeights in memoryWeights in memory — Direction: Down, roughly proportionallyCard eligibilityCard eligibility — Direction: Narrower, not widerOutputOutput — Direction: Changed, by a model-dependent amountParameter countParameter count — Direction: Unchanged
Fig. 2 Filled where the repository states something, hollow where it does not. A blank is not an inference.
What a precision change affects. Recorded 2026-09-22.
EffectDirection
Weights in memoryDown, roughly proportionally
Card eligibilityNarrower, not wider
OutputChanged, by a model-dependent amount
Parameter countUnchanged

Inclusion rule. Effects of changing the numeric format. An effect that does not occur is listed, because it is the one people expect. Order. From the effect on memory outward to the effect on the count.

1A figure without a format is incomplete

One repository in this register states from 4GB in FP16 and from 5GB in BF16 for two of its releases. The formats are part of the values, and dropping them would make both figures unusable.

Most rows here do not name a format. That is recorded as it stands rather than assumed, because assuming half precision would put a convention into the register that the vendor did not state.

2Lower precision narrows the hardware

An fp8 build needs a card whose hardware supports fp8, so the cheaper configuration is available on fewer machines rather than more. One repository states its card generations and formats in the same sentence for this reason.

A card with plenty of memory and an older architecture satisfies the number and not the requirement, which is the failure mode a bare gigabyte figure invites.

3How often the format is stated here

Four families name a numeric format beside a memory figure, and one of those states two different formats for two of its own releases. The rest publish figures with the format left implicit.

A figure whose format is unstated cannot be adjusted, because halving the bytes per parameter is the largest single lever on the weights footprint. The register records the omission rather than filling it.

A reading note, not an entry: no vendor value appears on this page. Where the register records this term for a particular release, it is on the hardware column and on that family's own page. Nearby terms: Model sharding, GPU architecture.