Get Nook
  1. Nook
  2. Labs
  3. nl.MM
  4. NL-2610.004

NL-2610.004Result[nl.MM]

An "8 GB" graphics card reports 7.99 GiB, and the desktop may be using half of it

The Nook team

Submitted 5 October 2026 · 2 min read

Abstract

When we added more video models to Nook, each listed with the video memory (VRAM) it needs, none of the 8 GB models was marked as fitting our 8 GB card. The RTX 4060 reports 8188 MiB, which is 7.99 GiB, so a strict comparison failed by a few megabytes. Nook now allows half a gigabyte of slack when it compares a model's needs with the card. While testing, we also found about 4.7 GB of the card already in use by desktop apps before any model loaded. The number on the box says little about what a model will actually find free.

Comments
RTX 4060 8 GB, Windows 11. Two readings from one machine, not a survey.
Subjects
Memory (nl.MM); Images and video (nl.VI)
Cite as
NL-2610.004 · usenook.ai/labs/what-8-gb-means.html

Full text

Result

An “8 GB” card reports slightly less than 8 GiB, so checks need some slack; Nook allows half a gigabyte. What matters more is what is free: on our card, desktop apps were once holding 4.7 GB of the 8.

What we found

Nook's Video page lists each model with the memory it needs and marks it “Fits your GPU” or “Needs a 16 GB GPU”. After we added Wan 2.2 TI2V 5B, listed as needing 8 GB, it showed as not fitting our 8 GB RTX 4060. The card was not too small; the check was too strict.

ReadingValue
On the box8 GB
Reported by the driver8188 MiB = 7.99 GiB
8 GiB exactly8192 MiB
In use by desktop apps, one afternoonabout 4.7 GB
Left for a model thenabout 3.3 GB

The card reports 4 MiB less than 8 GiB, so “needs 8” compared with “has 7.99” said no. Model makers round in their own ways too: a model listed for 8 GB is meant for 8 GB cards.

What we changed

Nook now counts a model as fitting when the memory it needs is no more than the card's memory plus half a gigabyte. The same rule is used in the app's engine and in its interface, so the mark and the setup agree.

What it means on your machine

  • Look at what is free, not at the label. Task Manager's Performance tab shows it under GPU, or ask the driver:
nvidia-smi --query-gpu=memory.total,memory.used,memory.free --format=csv
  • Close what holds video memory before a big job: browsers with hardware acceleration, games, video calls, screen recorders.
  • If what is left is less than the model needs, on Windows the job will most likely crawl rather than fail. NL-2610.002 explains why, and how to tell.

Limits

How sure we are. One card. The 4.7 GB is a single reading on an afternoon with ordinary apps open; how much the desktop takes depends entirely on what is running, and we did not note which apps held what. Cards of other makes and sizes report their own totals.

Run it on your own GPU.

Create AI videos and images on your own computer with Nook. Open models, local generation, no generation credits.