Home / Calculators / Home server
Can your machine handle these repos?
Tick the repos you want to host at home and add up the RAM, CPU and disk they ask for according to their official documentation. If the docs do not give a number, I tell you so instead of making one up.
The calculator
Every number comes from the repo's official documentation (source and date in the table below). "No official data" means the docs do not publish it; we do not estimate it.
Documented requirements per repo.
These are the numbers the calculator uses, exactly as each project publishes them. A dash means the documentation does not give the figure.
| Repo | RAM min. · rec. | Cores min. · rec. | Disk min. · rec. | GPU | Source |
|---|---|---|---|---|---|
| Immich | 6 GB · 8 GB | 2 cores · 4 cores | 3 GB · — | Not required | Documentation Checked Oct 29, 2026 · Immich "requirements" documentation (no version number on the page) |
| With only 4 GB it runs only without the machine-learning features. If you limit Docker, Postgres needs at least 2 GB. Your photos and videos go separately: thumbnails and transcoded videos add 10-20% to your library. The database takes 1-3 GB (we use 3) and must live on a local SSD. | |||||
| Jellyfin | 4 GB · 8 GB | — · — | — · 100 GB Recommendation with a shared system: not added |
Optional | Documentation · more Checked Oct 29, 2026 · Jellyfin "Hardware selection" documentation (mentions 10.11 and 10.9) |
| The 4 GB are for a Linux server with no desktop environment ("may be enough"); the general recommendation is 8 GB. Transcoding H.265 or AV1 requires a capable GPU or integrated graphics; the docs give no CPU core counts, only models. The docs DO give a disk figure: they recommend 100 GB on an SSD for the system, Jellyfin and the transcoding cache (it does not count your movies). It is a recommendation that includes the operating system you share with your other repos, so we show it but do not add it as the app's disk. The figures do not include: Your movie and TV library, and the GPU or integrated graphics you use for transcoding. | |||||
| Nextcloud | 128 MB · 512 MB Per process: not added |
— · — | — · — | Not required | Documentation · more Checked Oct 29, 2026 · Nextcloud 35 (administration manual 35; the "latest" path may point to another version) |
| Careful: the official figures (128 MB minimum, 512 MB recommended) are PER PHP PROCESS, not the server total. With several users, a database and apps, the total is quite a bit higher and the docs do not calculate it; that is why we do not add them up and the evaluation stays incomplete. The docs do not give the app size; your files, calendars and contacts grow with use. The figures do not include: The database, the cache, the apps, concurrent users and how many PHP processes run at once: the docs do not model them. | |||||
| Pi-hole | 512 MB · — | — · — | 2 GB · 4 GB | Not required | Documentation Checked Oct 29, 2026 · Pi-hole "prerequisites" documentation (no version number on the page) |
| The docs describe it as very light and say it does not need much processor; they give no core count. Any machine with a supported operating system will do. The docs ask for 2 GB free as a minimum and recommend 4 GB. | |||||
| Stirling-PDF | — · 4 GB | 2 cores · 4 cores | — · 10 GB | Not required | Documentation · more Checked Oct 29, 2026 · Stirling-PDF "Performance" and "Docker" documentation (no version number on the page) |
| It is the "1-10 users" row (2-4 cores, 4 GB, 10 GB), not an absolute minimum. OCR, Office conversion and PDF to image are what need the most memory; if the container runs out of RAM, jobs in progress fail. Figure from the sizing table for 1-10 users; it is space for the app and its temporary files, not for your PDFs. The figures do not include: More users or more load: the table goes up to 8-16 GB for 10-50 users. | |||||
| openmediavault | 1 GB · 8 GB | — · 2 cores | — · — | Not required | Documentation Checked Oct 29, 2026 · openmediavault "stable" documentation (the page does not give a version number) |
| It is a NAS operating system (on top of Debian), not an app you mount on top of another system: if you use it, the whole machine is the NAS. Official table: minimum RAM 1 GiB, "best" 4 GiB, recommendation 8 GiB or more dual channel; recommended CPU is dual core. The docs do not give the system size; they separate the system disk from the data disks, which depend on what you store. The figures do not include: The data disks: they depend on what you store. | |||||
| AnythingLLM | 2 GB · — | 2 cores · — | 5 GB · 10 GB | Optional | Documentation · more Checked Oct 29, 2026 · AnythingLLM documentation (container and desktop; no version number on the page) |
| These are the container values using a cloud model; on x86 it requires a CPU with AVX2. If you run the model on the same machine, the model sets the requirement: the desktop docs recommend 16 GB and 8 cores for a basic experience. The 5 GB are for the app (10 GB recommended on a cloud server); any local models you download add to that (their size depends on the model). The figures do not include: The local or built-in language model (if you turn it on): it needs memory of its own depending on the model, and the docs recommend a GPU. | |||||
| Open WebUI | — · — | — · — | 1.62 GB · — | Optional | Documentation Checked Oct 29, 2026 · Open WebUI "Quick start" documentation (image measured by the docs on Oct 16, 2026; Python 3.11 and 3.12) |
| The docs do NOT publish a RAM or CPU minimum. Their test of the slim image measured about 300 MiB at idle, and they clarify that this is not a minimum. The language model you connect (Ollama or another) needs its own resources separately. This is what the :main image weighs (:slim drops to 197 MB); your chats and documents add to it. The figures do not include: The language model you connect (Ollama or another API): it is hosted separately and needs its own resources. | |||||
| InvokeAI | 8 GB · 16 GB | — · — | — · — | Required VRAM 4–8 GB |
Documentation Checked Oct 29, 2026 · InvokeAI "System requirements" documentation (no version number on the page) |
| Minimum = Stable Diffusion 1.5 (8 GB of RAM, 4 GB of VRAM); recommended = SDXL (16 GB, 8 GB). FLUX.1 needs 10 GB of VRAM and 32 GB of RAM. On Linux ARM64 (a Raspberry Pi 5, for example) it only runs on CPU, slowly; it works with API models. The docs do not give the app size; the models you download weigh several GB each. The figures do not include: The models you download (several GB each). | |||||
| ComfyUI | 8 GB · — | — · — | — · — | Optional VRAM min 4 GB |
Documentation Checked Oct 29, 2026 · Repo README (main branch; no version number) |
| The README says it runs even the largest models with "4 GB of VRAM and 8 GB of RAM" thanks to its weight offloading, and that the portable version works on CPU only. Its figures are with a GPU: without one, that promise does not apply. The docs do not give the app size; the models you download weigh several GB each. The figures do not include: The models you download (several GB each). | |||||
| Fooocus | 8 GB · — | — · — | — · — | Required VRAM min 4 GB |
Documentation Checked Oct 29, 2026 · Repo README (project in limited support; no version number) |
| Official table: with an Nvidia RTX 2000/3000/4000 GPU it asks for 4 GB of VRAM and 8 GB of RAM, with system virtual memory (swap) mandatory; with a GTX 1000 it goes up to 8 GB of VRAM. CPU only: 32 GB of RAM. The project is in limited support (bug fixes only). Its troubleshooting guide asks for at least 40 GB free on each disk if the "CPUAllocator" error shows up; it does not give the app size. The figures do not include: The models you download. | |||||
| Sunshine | 4 GB · — | — · — | — · — | Required | Documentation Checked Oct 29, 2026 · Repo README (the docs warn that the tables are under construction) |
| It asks for 4 GB of RAM or more, a Ryzen 3 / Core i3 CPU or better, and a GPU with a hardware encoder (NVENC, VCE, VAAPI or QuickSync). The docs themselves warn that those tables are under construction and that you should not buy hardware based on them. For 4K it suggests Ryzen 5 / Core i5. The docs do not give the app size. | |||||
| Frigate | 4 GB · 16 GB | — · — | — · — | Optional | Documentation Checked Oct 29, 2026 · Frigate "Planning a setup" documentation (no version number on the page) |
| 4 GB is the basic minimum (few cameras, with a detection accelerator and no extra features); 8 GB if you use face or license plate recognition; 16 GB recommended with 8+ cameras. It requires a CPU with AVX and AVX2 (missing on older Celeron, Pentium and Atom) and recommends an accelerator or GPU for detection: the CPU detector is not the recommended one. It depends on how many cameras you have, their resolution and how many days you keep; the docs recommend using an IP camera storage calculator. The figures do not include: Recording storage and the detection accelerator (Coral, GPU or another). | |||||
| whisper.cpp | 273 MB · 3.9 GB | — · — | 75 MB · 2.9 GB | Not required | Documentation Checked Oct 29, 2026 · Repo README, "Memory usage" section (no version number) |
| The official table is per model: minimum = tiny (~273 MB), recommended here = large (~3.9 GB, the one that needs the most). In between: small ~0.85 GB, medium ~2.1 GB. The bigger the model, the better it transcribes and the longer it takes. It is the model size on disk: tiny 75 MiB up to large 2.9 GiB. | |||||
| Bark | — · — | — · — | — · — | Optional VRAM 2–12 GB |
Documentation Checked Oct 29, 2026 · Repo README (no version number) |
| The full version needs about 12 GB of VRAM; the small version of the models fits in 8 GB, and with tweaks (CPU offloading and small models) it gets down to ~2 GB. On a CPU or an old GPU the docs warn it is much slower. They give no RAM figures. The docs do not give the size of the models on disk. | |||||
How it is calculated.
- Available. From your machine's RAM and cores we subtract a reserve for the system and Docker: 1 GB and 0.5 core. It is our assumption, not a figure from any repo. From the disk we subtract "Your data".
- Sum of minimums and of recommended figures. We add up the minimum RAM of each ticked repo and, separately, the recommended one. If a repo publishes only one of the two, the other is taken as equal; if there is only a recommended figure, the minimum is left without data and we tell you.
- Why it is an approximation. Several apps can share a database or a proxy and their usage peaks almost never coincide, so adding up can come out short or long; also, the docs' minimums are usually for starting up under a light load. It is a good first filter, not a guarantee.
- The CPU is shared in turns. Unlike RAM, which is occupied all the time, cores are shared out. We only mark "not enough" if a single repo asks for more cores than you have; if the sum of recommended figures exceeds them, it comes out "tight".
- GPU. Repos that require one (Sunshine or InvokeAI, for example) warn you if your machine has no dedicated GPU. If you use integrated graphics with a hardware encoder, check the repo's docs.
- No official data. If the documentation does not publish a figure, it does not enter the sum and never counts as zero: that part stays unevaluated.
- "Incomplete evaluation". We separate two questions: do the figures we do know fit on your machine? and do we have RAM, CPU and disk figures for all the repos you ticked? Only with both in order do we say "Enough" or "Tight". If either is missing, the verdict is "Incomplete evaluation", with the list of what is missing and of what the figures do not include (for example, the language model you connect to Open WebUI is hosted separately). Adding a repo with missing data never improves the result.
- A known failure wins. If what is documented already does not fit, the result is "Not enough" even if data for other repos is missing.
- Suggested machine. A machine is only called "comfortable" when the figures are complete. With missing data we show the smallest one where nothing known fails, making clear it is not a recommendation, and if neither RAM nor CPU has a figure we suggest nothing.
- Scope of each figure. Not every official figure describes the whole service. Those that do (or that depend on a model or a load the docs name) are added up; the others are shown but not added:
- Whole service. Figure for the entire service, not counting anything hosted separately. (Added.)
- Depends on the model. Depends on the model you pick: the minimum is the smallest model and the recommended figure is the largest one on the list. (Added.)
- Depends on the load. Applies to the load the scenario names (for example 1-10 users); with more load you need more. (Added.)
- Per process. Per-process figure, not the total: the total depends on how many processes run and on the database, the cache and the apps. It cannot be added up without those assumptions. (Shown, not added.)
- Per user. Per-user figure: it needs to know how many users there will be. (Shown, not added.)
- Deployment example. Figure from one specific installation (with extra services included), not a minimum for the program. (Shown, not added.)
- Recommendation with a shared system. It includes the operating system and more, which you share with your other repos; adding it for each app would count that space several times. (Shown, not added.)
Minimum versus recommended.
The minimum is what the project says is needed for it to work; the recommended figure is for it to work smoothly, with more users or with more features turned on. Some documented examples:
- Immich: 6 GB minimum and 8 GB recommended; the docs talk about a smooth experience, especially when uploading files.
- Frigate: 4 GB is the basic minimum (few cameras, with an accelerator and no extra features), 8 GB if you turn on face or license plate recognition, and 16 GB with 8 or more cameras.
- Stirling-PDF: its table goes by number of users: 1 to 10 users, 4 GB; 10 to 50, 8 to 16 GB.
- Jellyfin: 8 GB for average use; on a Linux server with no desktop environment, 4 GB may be enough.
When the verdict is "tight", you usually meet the minimums but not the recommended figures: it will work, with less room to spare.
Frequently asked questions.
- How much RAM does Immich need?According to its documentation, 6 GB minimum and 8 GB recommended, with 2 cores minimum and 4 recommended. With only 4 GB it can run if you turn off the machine-learning features. If you put limits on Docker, the Postgres database needs at least 2 GB by itself.
- Can I run everything on a Raspberry Pi?It depends on what "everything" is. With the documentation's data: an 8 GB Raspberry Pi 5 leaves about 7 GB after the system reserve (our assumption of 1 GB), and Immich already asks for 6 GB as a minimum; if you add Jellyfin (4 GB minimum, 8 recommended) you go over those 7 GB. Pi-hole, on the other hand, asks for 0.5 GB. Tick your repos above and see the verdict with your Pi model. And on ARM64, like the Pi, InvokeAI only runs on CPU, and slowly.
- SD card or SSD?In the repos we checked there is no general rule about SD cards. There are hints, though: Immich says the database must live on a local SSD and never on a network folder, and Jellyfin recommends 100 GB on an SSD for the system, its files and the transcoding cache. If you use a Pi, a USB SSD is usually the most comfortable option for those databases.
- VPS or a machine at home?RAM and cores are calculated the same way for both. What changes is where your data lives and how much disk you can have. A 2 GB VPS, for example, falls below Immich's 6 GB minimum. Choose based on how much control you want over your files and how much you want to maintain yourself.
- Does Docker use a lot?We have no official Docker measurement, so we use an assumption: reserving 1 GB of RAM and half a core for the system and Docker (it is edited in data/reqs.php). What the repos do document: Stirling-PDF adjusts Java's memory up to 70% of the container limit, and if the server runs out of memory, the container stops and jobs in progress fail.
- And my photos, how much disk do I need?Immich says thumbnails and transcoded videos increase the library size by 10 to 20% on average, plus 1 to 3 GB for the database. For example, 500 GB of photos and videos would be 550 to 600 GB, plus the database. Type that figure into "Your data" and the calculator subtracts it from the disk.
Methodology.
We checked each repo's official documentation on Oct 29, 2026 and link the exact page each number comes from. Versions change: what is the minimum today may go up in the next update, so check the source before buying hardware.
This is documentary research, not a lab test: we did not install or measure anything on hardware. Each repo keeps its figures in a single record (data/reqs.php) with their scope, scenario, source and the version of the documentation checked; the repo's fact sheet and this calculator read that same record.
We do not include in the calculator repos whose documentation does not publish a minimum figure that applies to a home server:
- n8n. The n8n docs do not publish general hardware requirements. Their only figure (4 GB of RAM and 2 vCPU) comes from the Docker Compose guide that includes an isolated environment (sandbox) for the AI assistant, and the page does not say whether it applies without that environment or with other install methods. We show it on its guide page as a deployment example, but we do not use it as a minimum for the program. See its fact sheet.
- Vaultwarden. We did not find an official minimum figure that applies to a home server. See its fact sheet.
- Syncthing. We did not find an official minimum figure that applies to a home server. See its fact sheet.
- Paperless-ngx. We did not find an official minimum figure that applies to a home server.
- Penpot. We did not find an official minimum figure that applies to a home server.
The example machines use RAM and cores from the manufacturer's spec sheet when there is a link; the disk and the VPS are round example sizes.
Spotted outdated data? Write to me with the link to the docs and I will fix it.