Skip to content

How-ToSSD

How Much SSD Space Do You Need for Local AI? Model Sizes and Where They're Stored

Local AI models are large: in Ollama, llama3.1:8b is 4.9GB, gpt-oss:20b 14GB, llama3.1:70b 43GB and gpt-oss:120b 65GB, and SDXL's base checkpoint is 6.94GB. Budget about 50GB to experiment and 250GB or more for several models. Ollama, LM Studio, Hugging Face tools and ComfyUI can all store models on a second drive.

By , Editorial team

4 min read

How Much SSD Space Do You Need for Local AI? Model Sizes and Where They're Stored

Most local AI guides focus on VRAM and RAM, but storage fills up fast too. Every model you download is a multi-gigabyte file, and it's easy to end up with several versions of the same model in different apps. This guide uses the download sizes published in Ollama's model library and on Hugging Face, plus each app's own documentation for where files are stored. Forge PC hasn't measured every app's disk use, and sizes change as models are updated, so treat the figures below as examples rather than fixed rules.

How big are local AI models?

Model size depends mainly on the number of parameters and the quantisation (how heavily the weights are compressed). Ollama's library lists these download sizes:

Model (Ollama tag)

Download size

llama3.1:8b (default 4-bit quantisation)

4.9GB

llama3.1:8b-instruct-q8_0 (8-bit)

8.5GB

llama3.1:8b-instruct-fp16 (unquantised)

16GB

gpt-oss:20b

14GB

llama3.1:70b

43GB

gpt-oss:120b

65GB

llama3.1:405b

243GB

Image generation adds more. On Hugging Face, Stability AI's SDXL base checkpoint file (sd_xl_base_1.0.safetensors) is 6.94GB on its own, before you add a refiner, LoRAs, upscalers or other checkpoints.

Two things follow from the table. First, the same model can be three times bigger depending on quantisation: the 8-bit and full-precision versions of Llama 3.1 8B are 8.5GB and 16GB, against 4.9GB for the default. Our LLM quantisation explainer covers the trade-off. Second, the jump from small to large models is steep: a single 70B model is about the size of eight or nine 8B models.

How much space to budget

These are planning suggestions based on the sizes above, not hard limits:

  • Trying local AI: a few small chat models plus one image model fit in roughly 50GB.

  • Regular use: several chat models in two or three sizes, a coding model and a couple of image checkpoints can easily pass 150-250GB.

  • Large models: one 70B-class and one 120B-class model already take over 100GB together, before anything else.

Leave free space on top. SSDs need room for normal housekeeping, and Windows, games and updates need space too. If your system drive is a 1TB SSD that's already half full, a dedicated second drive for models is the tidy solution. See our best 2TB SSD and best 4TB SSD guides.

Does a faster SSD make AI faster?

A faster drive mainly shortens the time it takes to load a model into memory. Once a model is loaded into VRAM or RAM, generation speed depends on your GPU, memory and settings, not on the SSD. A PCIe 4.0 NVMe drive is a sensible choice for a models drive; a hard drive will work but makes loading large models slow. Our NVMe vs SATA SSD guide explains the differences.

Where each app stores models (and how to move them)

Ollama

Ollama's FAQ says models are stored in C:\Users\%username%\.ollama\models on Windows. To use a different folder, set the OLLAMA_MODELS environment variable to the new location. On Windows, quit Ollama from the taskbar first, open Settings, search for "environment variables", choose Edit environment variables for your account, add OLLAMA_MODELS, click OK and restart Ollama. Our guide to managing Ollama models walks through moving existing models.

LM Studio

LM Studio's documentation says you can change the models directory from the My Models tab. Our guide to choosing models in LM Studio explains which downloads to pick.

Hugging Face tools

Python tools that download through the huggingface_hub library cache files in ~/.cache/huggingface by default (the hub cache goes in a hub folder inside it). Hugging Face's documentation says you can change this with the HF_HOME environment variable, or set HF_HUB_CACHE to move only the model cache. Its cache-management guide also explains how to scan and delete cached files.

ComfyUI

ComfyUI's documentation describes an extra_model_paths.yaml file (there's an extra_model_paths.yaml.example in the ComfyUI folder to copy). You set a base_path pointing at a folder of models, so ComfyUI can use models stored elsewhere without copying them. The desktop app uses an extra_models_config.yaml file in your AppData\Roaming\ComfyUI folder instead. Our ComfyUI install guide covers setting it up.

Tips to save space

  • Delete models you don't use. Our Ollama models guide shows how to list and remove them; LM Studio's My Models tab and Hugging Face's cache tools do the same job for those apps.

  • Avoid duplicates. Ollama, LM Studio and Hugging Face tools each keep their own copies, so the same model can end up on your drive two or three times.

  • Pick one quantisation. For most chat use on a gaming PC, a 4-bit or 5-bit version is the practical choice.

Our verdict

Budget around 50GB to try local AI, and 250GB or more if you plan to keep several models or any 70B-class model. Large models are the real space hogs: llama3.1:70b is 43GB and gpt-oss:120b is 65GB in Ollama. A second NVMe SSD used only for models keeps your system drive free, and every major tool (Ollama, LM Studio, Hugging Face and ComfyUI) lets you point it at another folder.

The bottom line

Budget around 50GB to try local AI, and 250GB or more if you plan to keep several models or any 70B-class model. Large models are the real space hogs: llama3.1:70b is 43GB and gpt-oss:120b is 65GB in Ollama. A second NVMe SSD used only for models keeps your system drive free, and every major tool (Ollama, LM Studio, Hugging Face and ComfyUI) lets you point it at another folder.

In this article

Sources

  1. 01Ollama, llama3.1 model tags
  2. 02Ollama, gpt-oss model tags
  3. 03Hugging Face, stabilityai/stable-diffusion-xl-base-1.0 sd_xl_base_1.0.safetensors
  4. 04Ollama, FAQ
  5. 05LM Studio, Download an LLM
  6. 06Hugging Face, Environment variables (huggingface_hub)
  7. 07Hugging Face, Manage the huggingface_hub cache-system
  8. 08ComfyUI, Models

About the author

Forge PC Staff

Editorial team

The shared byline for articles researched, written and edited by the Forge PC editorial team. How we source, check and correct our work is set out in our editorial policy.

More from Forge PC Staff

Read next

How to Enable Wake-on-LAN in Windows 11 (and Why It Stops Working)

How-To

How to Enable Wake-on-LAN in Windows 11 (and Why It Stops Working)

To use Wake-on-LAN, connect the PC by Ethernet, enable wake in the BIOS, turn on Wake on Magic Packet and 'Allow this device to wake the computer' in Device Manager, and wake the PC from sleep. Microsoft says Windows doesn't arm the network adapter after a Fast Startup shutdown, so use sleep or turn Fast Startup off.

Forge PC Staff4 min

Windows 11 Update Stuck or Failing? 9 Fixes to Try

How-To

Windows 11 Update Stuck or Failing? 9 Fixes to Try

If Windows Update is stuck or failing, back up your files and run the Windows Update troubleshooter in the Get Help app. Then check power and internet, use an admin account, unplug extra hardware, clear the update cache, fix the date and time, update drivers and free at least 20GB of space before trying again and restarting.

Forge PC Staff3 min

How to Use Speculative Decoding in LM Studio to Speed Up Local LLMs

How-To

How to Use Speculative Decoding in LM Studio to Speed Up Local LLMs

Speculative decoding in LM Studio pairs your main model with a much smaller draft model from the same family to speed up generation without changing output quality. Turn on Developer Mode, load the main model and pick a draft model in the chat sidebar. LM Studio says speed-ups vary by prompt and can vanish if the draft model is too big.

Forge PC Staff3 min

Best M.2 2230 SSD 2026: Upgrades for Steam Deck, ROG Ally and Small Laptops

Best Picks

Best M.2 2230 SSD 2026: Upgrades for Steam Deck, ROG Ally and Small Laptops

The Crucial P310 2230 is our pick for Steam Deck, ROG Ally and MSI Claw upgrades: Crucial rates it at up to 7,100MB/s reads. WD's SN770M publishes full specs (5,150MB/s reads, 600TBW at 1TB, 1,200TBW at 2TB, five-year warranty), and Corsair's MP600 MINI is worth comparing. Check your device takes a 2230 drive first.

Forge PC Staff3 min