Run Gemma 3 27B on-premises
○ Memory-fit arithmetic — checkable math, not a benchmark
Gemma 3 27B is Google's efficient mid-size open model — 27 billion dense parameters, multimodal, released under the Gemma license (commercial use permitted, with use terms). At 4-bit it uses a small slice of 128 GB of memory.
Here is the arithmetic against a machine with 128 GB of memory.
Memory fit in 128 GB
| Parameters | 27B (dense), multimodal |
|---|---|
| 4-bit weights | ~15 GB |
| Reference memory | 128 GB (one machine) |
| Left for context + other models | roughly 100 GB |
| Fits in 128 GB | yes, with room for more models |
At ~15 GB the model leaves most of 128 GB free — ideal for co-hosting with other models or serving high concurrency.
What it serves well
- Efficient multimodal chat and analysis on modest memory.
- High-concurrency or multi-model machines.
- A capable mid-size default when a 100B-class model is overkill.
Honest limits
- Memory arithmetic, not a benchmark.
- The Gemma license permits commercial use but carries Google's use terms; we review fit in the assessment.
Frequently asked questions
Does Gemma 3 27B fit in 128 GB of memory?
Easily — ~15 GB at 4-bit against 128 GB leaves roughly 100 GB free for context and other models.
Is it multimodal?
Yes — Gemma 3 27B handles text and images, served on your own hardware.
Can I use it commercially?
The Gemma license permits commercial use subject to Google's terms; we confirm the fit for your case during the assessment.
Find out which model fits your work
An AI diagnostic maps your processes, classes what is sensitive and sizes the model and the hardware against your real tasks.