Hardware needed to run ZenithAI
ZenithAI runs its AI models on your own server, not in someone else’s cloud. That needs serious hardware, mostly graphics cards with plenty of memory. You can choose an x86 server or an ARM64 server.
Minimum server (x86)
Section titled “Minimum server (x86)”| Part | Minimum |
|---|---|
| Graphics cards (GPU) | 2 × NVIDIA RTX PRO 5000 Blackwell, 48 GB memory each (96 GB in total) |
| System memory (RAM) | 128 GB |
| Storage | 1 TB SSD |
| Operating system | Rocky Linux 9 or later (x86, 64-bit) |
| Network | Internet during setup, and ports 80 and 443 open |
You install it yourself with the x86 installer. It also runs on Red Hat Enterprise Linux and AlmaLinux, version 9 or later.
ARM64 server
Section titled “ARM64 server”ZenithAI also runs on ARM64 servers. This is an extra choice next to the x86 server above, and you can install it yourself with the ARM64 installer.
| Part | Recommended, with the Genius tier |
|---|---|
| Processor (CPU) | ARM64 (aarch64), 16 or more cores |
| Operating system | Ubuntu 24.04 LTS |
| Graphics memory (GPU) | 96 GB of NVIDIA graphics memory: one 96 GB card, or 2 cards with 48 GB each |
| System memory (RAM) | 128 GB |
| Storage | 1 TB SSD |
| Network | Internet during setup, and ports 80 and 443 open |
Tiers and graphics memory
Section titled “Tiers and graphics memory”On both kinds of server, the installer looks at your graphics cards and sets up the tiers they can hold. With 96 GB, on one card or as 2 cards with 48 GB each, you get all four: Instant, Fast, Smart and Genius. With less graphics memory you get Instant, Fast and Smart, without Genius.
Raising the context window
Section titled “Raising the context window”Each tier has a default context window: 48K tokens on Instant, Fast and Smart and 128K on Genius. The most ZenithAI allows on any tier is 128K (131,072 tokens). A larger window helps long chats and long documents, and it costs graphics memory and time:
- Memory. Every chat being answered at the same moment holds its own window, so memory grows with the window and with the number of simultaneous chats. As a rough guide on a 96 GB server, each extra 16K on Instant or Fast takes about 1.3 GiB of graphics memory, and taking Genius from 48K to 128K took about 5 GiB. Instant, Fast and Smart share one card, so raising them may mean fewer simultaneous chats.
- Speed. A long prompt waits longer for its first word. A prompt near 125K tokens on Genius can take 30 to 40 seconds on its own, and one to two minutes when three such prompts arrive together.
- How. Your administrator asks ZenithAI support to change it. The model service restarts, and
GET /v1/modelsshows the new value afterwards. Do not edit the settings by hand. - Check first. Most prompts are far shorter than the window. Raise it for a real need, such as long contracts or long tool runs.
Also plan for
Section titled “Also plan for”- A web address for your users, such as
ai.yourcompany.in, and a TLS certificate for HTTPS. Sign-in must always happen over HTTPS. - Space for growth. Workspaces, chat files and generated documents are stored on the server.
- Backups. Keep a copy of the application data, its settings and the database on another machine.
- Internet access. The installer downloads packages and AI models, so the server needs internet during setup. After that it is needed only for web search and the daily licence check-in, and ZenithAI keeps working without it.
Your licence and your hardware
Section titled “Your licence and your hardware”Your licence is linked to your server’s hardware. If you change the server or its main hardware later, you will need a new licence key. See Licensing.
Common questions
What is the minimum hardware for ZenithAI?
Two NVIDIA RTX PRO 5000 Blackwell graphics cards with 48 GB of video memory each, so 96 GB in total, plus 128 GB of system RAM and a 1 TB SSD. This is the minimum; more memory and storage give you room to grow.
Can ZenithAI run on a normal office PC or a laptop?
No. ZenithAI runs its AI models on your own server, and those models need large professional graphics cards. Plan for a rack or tower server that meets the minimum list on this page.
Which operating system does ZenithAI use?
ZenithAI is installed on Linux. On an x86 server it uses Rocky Linux 9 or later, and on an ARM64 server it uses Ubuntu 24.04 LTS. Your IT team provides the server, and the ZenithAI installer sets up everything else on it.
Can ZenithAI run on an ARM server?
Yes. ZenithAI runs on ARM64 servers with Ubuntu 24.04 LTS. You can download the ARM64 installer and run it yourself. It sets up the full product with a 30-day free trial and no limit on users.
