# Run Llama 3.3 70B

A capable general assistant that runs entirely on your own hardware. Ready in about 90 seconds on an A100 80GB, from $1.35/hr. No install, no code.

## What it does

Llama 3.3 70B is a large general-purpose language model. It writes, summarizes, answers questions, extracts information from documents, and follows instructions about as well as the commercial assistants most people already use.

The reason to run it here rather than through an API is privacy. Nothing you type leaves the machine we start for you, and we delete the machine and its disk when you stop.

That matters for legal documents, medical notes, HR files, customer records, and anything covered by a contract that forbids sending data to a third party.

## Your setup

We run Llama 3.3 70B on an A100 80GB with 80GB of memory, at $1.35 per hour, billed by the minute. It takes about 34 seconds per thousand words.

Seventy billion parameters is about 140GB at full precision and about 40GB compressed to 4-bit. The compressed version fits on an 80GB card with roughly 40GB left over, and that leftover memory is what holds the conversation and any documents you paste in. Squeezing it onto a 48GB card works until you paste a long document, at which point it runs out of room mid-answer.

## What it costs

100 thousand words written takes about 57 minutes of GPU time, which is about $1.28.

## Two ways to run it

- **Open the app.** Open WebUI in your browser. It looks and works like a normal chat window, and you can upload documents into the conversation.
- **Run a batch job.** Send a spreadsheet where each row is a task. We run all of them and return the answers in a new column.

## Model facts

| Field | Value |
|---|---|
| Parameters | 70B |
| License | Llama 3.3 Community License |
| Memory required | 48GB minimum (4-bit), 80GB recommended |
| Base model | Trained from scratch by Meta |
| Publisher | Meta |
| Recommended hardware | A100 80GB, 80GB |
| Category | [Language models](https://gpuvault.io/working-on/language-models/) |

## Questions people ask

### Is running Llama 3.3 here actually private?

Your session runs on one machine, reachable only through a link tied to your account. We do not read the contents of your session, and the disk is wiped when the session ends. One caveat: the machine belongs to a verified provider in our network, not to us. For regulated data, read our security page and talk to us about a dedicated instance before you start.

### Can I use Llama 3.3 in a commercial product?

Yes, with one condition. The Llama 3.3 Community License permits commercial use unless your product had more than 700 million monthly active users on the day Llama 3.3 was released. Almost no company crosses that line. You also have to display "Built with Llama" if you build a product on top of it.

### Do I need to install anything to use Llama 3.3 70B?

No. We start a machine with Llama 3.3 70B already loaded and hand you a link. Everything runs in your browser, there is nothing to download, and nothing is left on your computer afterwards. It works the same on a Mac, a Windows laptop, or a Chromebook. A workspace is usually ready in about 90 seconds.

### How much does it cost to run 100 thousand words written?

About $1.28. Llama 3.3 70B takes about 34 seconds per thousand words on the A100 80GB we recommend, so 100 thousand words written is roughly 57 minutes of GPU time at $1.35 per hour. Billing is by the minute, and a batch job shuts the GPU off the moment the last item finishes. The calculator above works this out for your own numbers.

### Can I run Llama 3.3 70B on a cheaper card?

Sometimes, and the calculator will not always make it look worth it. Seventy billion parameters is about 140GB at full precision and about 40GB compressed to 4-bit. The compressed version fits on an 80GB card with roughly 40GB left over, and that leftover memory is what holds the conversation and any documents you paste in. Squeezing it onto a 48GB card works until you paste a long document, at which point it runs out of room mid-answer. If you want to try a different card anyway, the advanced catalog lets you pick one and shows the estimated time before you commit.

### What are the license restrictions on Llama 3.3 70B?

Llama 3.3 70B is released under the Llama 3.3 Community License. That is not a permissive license, so read it before you use the output in paid work. We show the license on every model page precisely because this catches people out, and there is usually a permissively licensed alternative in the same category.

---

*Source: https://gpuvault.io/model/llama-3-3-70b/ — GPUVault, GPU rental by the minute.*