Requirements
Hardware
The generic requirements for hardware to be assigned to the virtual appliance are as follows. The actual requirements may vary due to the wide variety of possible configurations in terms of included technologies, used models, optional GPU usage, etc.
- x86_64 compatible processor
- 60 GB of disk space
- 4 CPU cores
- 32 GB of RAM
- NVIDIA GPU with at least 8 GB VRAM and CUDA compute capability at least 7
The actual requirements depend on chosen set of included technologies and models, on actual processed media files, and also on further configuration according to required processing throughput.
Typically,
- more CPU cores and more RAM are required for parallel processing by multiple technologies
- more GPU VRAM is required for running multiple technologies on GPU
To assess the actual requirements it's necessary to perform own tests using media files corresponding to the planned use case and find the optimal configuration settings.
Software
- hypervisor for running the virtual appliance
- DHCP-assigned IP addresses
- additional requirements apply for GPU support:
- IOMMU (Intel VT-d, AMD-Vi) support in both machine HW and BIOS/UEFI
- Host OS must be able to NOT use the GPU device, so it can be passed to the virtualization platform
- GPU Passthrough support in virtualization platform (PCIe Passthrough) - see table below
GPU
The GPU is really required for Enhanced Speech to Text Built on Whisper
technology to get reasonable performance.
We discourage from using this technology purely on CPU as the transcription
speed is then hardly bearable - e.g., a 1 minute long audio takes about
10 minutes to transcribe - and without extra reconfiguration of numerous
settings leads to various obscure errors.
Other technologies do not require GPU and run reasonably fast on CPU only. However, using GPU significantly increases the processing speed.
| Hypervisor | GPU Passthrough capable | Note |
|---|---|---|
| Oracle VirtualBox | ❌ | |
| VMWare Workstation Pro | ❌ | |
| VMWare ESXi | ✅ | Since version 8.0 U3e available for free (with limitations, e.g. maximum 8 vCPU per VM) |
| Microsoft Hyper-V | ✅ | Only in Windows Server, not possible in Windows 10/11 |
| Proxmox | ✅ | |
| KVM | ✅ |