About VRAM Fit
What VRAM Fit is, why it exists, how its content is produced and checked, who operates it, and how to report an error.
About VRAM Fit
What this site is
VRAM Fit is an independent website focused on a single subject: working out, before you download, whether a model will actually fit on your GPU or unified memory. Rather than covering a wide topic superficially, the site goes deep on a narrow set of questions. Each page is written to fully answer one question so you do not have to open five more tabs to get a usable answer.
Why it exists
People who run models locally hit a predictable wall. A file downloads, the loader reports out of memory, and the user has no idea whether the problem is the quantization, the context length, or the card itself. The same confusion repeats with unified-memory Macs, multi-GPU splits, and mixture-of-experts models whose headline parameter count hides the real memory cost. VRAM Fit turns those failures into numbers you can check before spending an hour on a download that will not load.
Who runs it
VRAM Fit is operated by an independent editorial team under the evankao.fun domain. It is not affiliated with, endorsed by, or sponsored by any of the companies whose products are discussed here, unless a page explicitly says otherwise. The site is organized as a brand publication rather than a personal blog, which is why pages carry the “VRAM Fit editorial” byline instead of an individual name.
How content is produced and checked
- Formulas are shown, not hidden. The memory math on each page is written out so you can reproduce it. The flagship VRAM calculator prints its formulas and includes hand-verified worked examples.
- Numbers come from stated sources. Hardware sizes use manufacturer nominal values. Quantization sizes use widely published bits-per-parameter approximations, with the caveat that real files vary a few percent. Nothing is presented as a measured figure unless it is.
- Pages are corrected, not duplicated. When a fact changes, the existing page is updated and its revision date moves. You will not find three near-identical pages competing with each other.
- Limitations are stated. Where something depends on a driver, engine, or batch size, the page says so and gives a range rather than a false precision.
- Errors are welcome. If you find one, the contact page reaches a person and the correction goes into the page.
How the site is funded
The site is free to read. Operating costs are covered by advertising, including Google AdSense, which may appear on pages. Advertising has no influence on what is written: no advertiser reviews, approves, or edits editorial content. Any affiliate placement would be disclosed on the page that contains it.
What we do not cover
This site deliberately avoids model leaderboards, benchmark rankings, and “best model” opinion pieces. Those require constant maintenance and go stale or wrong quickly. The memory and quantization material here changes far more slowly, which is what makes it worth publishing as reference.
Contact
Questions, corrections, and partnership enquiries can be sent to contact@vram.evankao.fun.
Frequently asked questions
What is VRAM Fit?
It is an independent reference site focused on one question: will a given model fit the memory you have, and at what quantization. It covers tools and explanations, not model leaderboards or benchmark opinions.
Who operates the site?
VRAM Fit is operated by an independent editorial team under the evankao.fun domain. It is not affiliated with or endorsed by any hardware or model vendor unless a page says so explicitly.
How is the content produced and checked?
Each page states the formulas and assumptions it relies on, cites vendor nominal specs rather than invented figures, and is updated in place when a fact changes. The calculator is verified by hand-worked examples printed on the page.
Why does the site exist?
Local model users repeatedly hit the same wall: a download that will not load, or a context length that crashes mid-session. The site turns those failures into numbers you can check before downloading.
How do I report an error?
Email contact@vram.evankao.fun with the page and the problem. Corrections are applied to the page itself and the revision date is updated, rather than left as a comment.