AUTOMATIC1111 states lowvram aggressively moves modules in/out of VRAM and is devastating for performance. More VRAM is the cleanest upgrade for interactive generation.
Users moving beyond basic 512px SD 1.x experiments.Stable Diffusion Hardware Requirements: VRAM, RAM & ZimaOS GPU
Learn ZimaOS Stable Diffusion WebUI VRAM, RAM, NVIDIA, low-VRAM and storage requirements, including current container compatibility limits.
Stable Diffusion requirements at a glance
The current ZimaOS Stable Diffusion package points to AUTOMATIC1111 Stable Diffusion WebUI and is configured for the NVIDIA runtime. AUTOMATIC1111 officially documents 4 GB VRAM as a low-VRAM operating tier with medvram/lowvram optimizations, but speed and maximum resolution suffer. System RAM, model family, image resolution, batch size and extensions also affect memory.
- GPU
- The ZimaOS package is configured with NVIDIA runtime and NVIDIA_VISIBLE_DEVICES=all. CPU fallback exists in the package metadata, but CPU-only generation is generally not a practical recommendation for interactive use.
- VRAM
- AUTOMATIC1111 documents 4 GB cards as low-VRAM-capable with medvram/lowvram modes. More VRAM gives much better room for SDXL-class models, higher resolutions, ControlNet, hires fix and batches.
- System RAM
- No universal official system-RAM minimum. Low-VRAM modes offload model components to CPU RAM, so insufficient system RAM can still cause failures even when VRAM optimizations are enabled.
- Storage
- Models, VAEs, LoRAs, ControlNet weights and generated images can consume tens or hundreds of gigabytes. ZimaOS maps separate model/output/config paths.
- Compatibility
- GPU generation matters as much as VRAM. In 2026 ZimaSpace users documented RTX 50-series failures because the current one-click container's PyTorch/CUDA stack did not include the new GPU architecture.
- Best Zima path
- A dedicated NVIDIA GPU is the meaningful path. ZimaCube 2 Creator Pack is the most direct preconfigured Zima GPU option, but verify exact VRAM plus CUDA/PyTorch compatibility before treating it as supported by the current one-click Stable Diffusion image.
From official requirements to the right setup
Stable Diffusion sizing starts with GPU compatibility and VRAM, then model/resolution and system RAM.
-
Official requirements
Confirm the ZimaOS container actually sees the NVIDIA GPU and that its bundled PyTorch/CUDA supports that GPU architecture. A visible GPU in ZimaOS does not guarantee the Stable Diffusion container can execute kernels on it.
-
Confirm your needs
Choose the model/workflow: SD 1.x 512px generation, SDXL, ControlNet, hires fix and large batches have very different VRAM demands.
-
Leave room to grow
Use medvram/lowvram only when needed. AUTOMATIC1111 documents that these modes move model components between VRAM and system RAM, reducing VRAM use at a performance cost.
-
Run it on ZimaOS
Run representative generations at the intended resolution/batch/extensions, watch VRAM and system RAM, then select hardware from measured headroom rather than a generic 'Stable Diffusion minimum'.
Check every playback client
- NVIDIA GPU visible inside the container
- CUDA/PyTorch architecture compatibility
- GPU VRAM
- SD 1.x versus SDXL/other model family
- Image resolution and batch size
- ControlNet/hires fix/extensions
- System RAM for offload
- Model/output storage capacity
Official minimum requirements
AUTOMATIC1111 does not provide one full PC specification, but its troubleshooting/wiki material explicitly supports 4 GB VRAM as a low-VRAM tier with reduced-performance modes.
Treat 4 GB as 'can be made to run some workflows' rather than a comfortable recommendation. The current ZimaOS package also has an independent container-compatibility risk with newer GPU generations.
| Requirement | Official minimum | What this supports |
|---|---|---|
| GPU path in ZimaOS app | NVIDIA runtime | IceWhale package sets NVIDIA runtime and exposes all NVIDIA devices. |
| Low-VRAM tier | 4 GB VRAM | AUTOMATIC1111 documents medvram/lowvram guidance specifically for 4 GB cards. |
| System RAM | No universal official minimum | Offload modes can move model components into CPU RAM. |
| Web UI port | 7860 | Current IceWhale package maps the Stable Diffusion WebUI on 7860. |
| ZimaOS package memory reservation | 512 MB | This is a container reservation in IceWhale compose metadata, not a Stable Diffusion system-RAM requirement. |
| Architecture | amd64 in current IceWhale compose | Current package metadata lists amd64 and points to AUTOMATIC1111. |
When to upgrade your hardware
Upgrade when VRAM or container compatibility blocks the exact image-generation workflow.
4 GB VRAM forces lowvram/medvram and generation is too slow
SDXL, ControlNet, hires fix or larger images repeatedly OOM
A newer NVIDIA GPU is unsupported by the one-click container
These workflows can raise VRAM and system-RAM demand well beyond a simple base-model generation. Reddit SDXL discussions show 8 GB can work in some setups but leaves far less margin.
AI art workflows using larger models and extensions.ZimaSpace community reports in 2026 show RTX 5060 Ti visible to ZimaOS but rejected by the Stable Diffusion container because its PyTorch/CUDA build lacked that architecture.
Users with RTX 50-series or newer professional GPUs.Plan hardware growth with confidence
Scale Stable Diffusion by increasing usable VRAM and keeping the software stack current.
Use medvram/lowvram only as a bridge
These options reduce VRAM use by moving model components to system RAM, but trade away speed.
Useful for testing a 4 GB card; not a substitute for adequate VRAM in a daily workflow.Add system RAM when offload is the chosen strategy
Low-VRAM workflows can consume more CPU RAM. Large SDXL/extension workflows may fail from system-memory pressure even when GPU VRAM is optimized.
16 GB or more system RAM provides safer offload headroom than a minimal host, but exact need is workflow-dependent.Keep models and outputs on large SSD/NVMe storage
Checkpoints, LoRAs, VAEs and generated images quickly exceed tiny system disks.
Use SSD/NVMe or ZimaCube storage instead of relying on eMMC for a growing AI-art library.Update the container stack for new GPU generations
A GPU can be physically supported by ZimaOS yet unsupported by an older PyTorch/CUDA image.
Verify torch CUDA capability inside the actual Stable Diffusion container before purchasing around a specific GPU.Can it run on ZimaOS?
The current ZimaOS Stable Diffusion app is an IceWhale-packaged AUTOMATIC1111 WebUI container with an NVIDIA runtime.
Treat the package as AUTOMATIC1111 WebUI
IceWhale's current compose metadata points repo/docs/support to AUTOMATIC1111 and maps model/output/config volumes.
Open the IceWhale Stable Diffusion composeVerify the actual GPU inside the container
Run NVIDIA/CUDA checks in the container before assuming generation works. New GPU architectures may need a newer PyTorch/CUDA base.
Read ZimaSpace RTX 50-series compatibility discussionUse the App Store package only with compatibility caveats
The ZimaOS listing is current, but one-click package age and GPU-generation support are separate from the App Store page being present.
Open Stable Diffusion in the ZimaOS App StoreChoose Zima hardware for Stable Diffusion
Stable Diffusion is a genuine GPU workload. Do not recommend CPU-only Zima hardware as the normal answer just because the container exposes CPU_FALLBACK.
Do you have a compatible NVIDIA GPU path that the current container can actually use?
No verified NVIDIA GPU
CPU fallback is technically possible but is not the recommended interactive Stable Diffusion experience.
- External-GPU experiment base after compatibility validationZimaBoard 2 1664
- Custom GPU/storage expansion baseZimaCube 2 Pro
Verified compatible NVIDIA GPU
Use the GPU tier whose VRAM and software compatibility fit the selected SD workflow.
- Most direct current preconfigured Zima GPU pathZimaCube 2 Creator Pack
No fixed image time or resolution is guaranteed. Model family, GPU VRAM, CUDA/PyTorch support, system RAM, optimization mode, extensions and batch size all matter.
| Zima hardware | Best for | Example workload | Core configuration | Recommended boundary | Next step |
|---|---|---|---|---|---|
| ZimaBoard 2 1664 | An experiment host for a user-added compatible NVIDIA GPU. | AUTOMATIC1111 with external/PCIe GPU, model storage and 16 GB system RAM. |
|
Stock ZimaBoard 2 1664 has no dedicated NVIDIA GPU. Validate GPU dock/PCIe power, CUDA runtime and the exact container before treating this as ready. | Get Now |
| ZimaCube 2 Pro | A storage-rich custom Stable Diffusion host with a user-added compatible GPU. | Large model/output storage plus a custom NVIDIA GPU expansion path. |
|
Stock Pro has no dedicated GPU. CPU/system RAM alone is not the recommended Stable Diffusion path. | Get Now |
| ZimaCube 2 Creator Pack | The most direct current preconfigured Zima GPU platform for Stable Diffusion experiments. | NVIDIA-accelerated image generation plus 64 GB system RAM and 1 TB system storage. |
|
Do not guarantee the one-click Stable Diffusion app works with the installed RTX PRO 2000 until exact GPU generation/VRAM and current container CUDA/PyTorch compatibility are verified. | Get Now |
What the Press Says
Highlights from trusted reviewers worldwide.
โZimaCube 2: Not just another NAS, tested with 25TB storage, local AI agents, 4K transcoding, and real homelab workflows.โRead full review
โThe ZimaBoard 2 is a compact x86 server board that can be turned into a mini NAS, home server, media box, or self-hosting hub.โRead full review
โZimaCube 2: A modern, high-performance NAS with plenty of room to growโbuilt for users who want more than basic storage.โRead full review
โCoverage focused on ZimaCube 2's open hardware foundation, no monthly fee, and self-hosting flexibility.โRead full review
Loved by the Community
Stories and reviews from people who build with Zima every day.
โ
โ
โ
โ
โ
Zima Blade Little yet Powerful
Maybe I am not digital natives but I live with PCs since 12 years old in 1984 when IBM PC clone come to my home. Many years have passed and many operating system I've tried. For me Zima blade and CasaOS was a quantum leap for home PC enthusiast and server lab machine to make me stay curious and relevant for this era.
โ
โ
โ
โ
โ
Very good!!
I use ZimaCube Pro as 5th Proxmox cluster node. It runs several VMs and containers, including a VM with GPU passthrough to run a self-hosted LLM. A specific LXC container runs a Samba server for NAS capabilities using four of six RAID 6 SATA HDDs with ZFS.
โ
โ
โ
โ
โ
Great innovation for mini server!
It is very useful and makes a powerful mini server for many purposes, including university and college students in engineering and electronics. Thank you so much for making this server.
โ
โ
โ
โ
โ
Avaliaรงรฃo ZimaBoard 2
Construรญ um servidor de uso pessoal. O desempenho estรก muito bom e funciona perfeitamente onde quer que eu esteja. A surpresa รฉ nรฃo dependermos de grandes estruturas para termos nosso prรณprio servidor de dados. Como iniciante, estou gostando bastante do ZimaOS, pois ele รฉ simples e eficiente.
Frequently asked questions
FAQ topics follow query fan-out around 4 GB/8 GB VRAM, SDXL, system RAM, CPU-only generation, medvram/lowvram and current ZimaOS GPU-container compatibility. Reddit is used to surface real failure modes, not to define official limits.
Is 4 GB VRAM enough for Stable Diffusion WebUI?
AUTOMATIC1111 explicitly supports 4 GB cards through medvram/lowvram optimizations, especially for smaller 512px workflows. It is a constrained tier and those options sacrifice speed.
Is 8 GB VRAM enough for SDXL?
It can work in some optimized workflows, but it is not a universal guarantee. Reddit SDXL users report both successful 8 GB setups and OOMs depending on UI, system RAM, resolution and extensions.
How much system RAM does Stable Diffusion need?
AUTOMATIC1111 does not publish one universal number. System RAM matters because medvram/lowvram move model components out of VRAM, and SDXL/extensions can create additional host-memory pressure.
Can Stable Diffusion run on CPU only?
Technically yes, and the ZimaOS package includes CPU_FALLBACK, but CPU-only diffusion is usually far slower than a compatible GPU and is not the normal product recommendation for interactive image generation.
What do medvram and lowvram actually do?
AUTOMATIC1111 splits model components and moves them between GPU VRAM and CPU RAM. medvram reduces VRAM with a smaller performance penalty; lowvram is more aggressive and much slower.
Why can Stable Diffusion run out of VRAM even when the GPU seems large enough?
Resolution, batch size, VAE, SDXL, hires fix, ControlNet, extensions and memory fragmentation all affect peak allocation. A nominal VRAM number is not the whole workload.
Why can a new RTX 50-series GPU fail in the ZimaOS Stable Diffusion app?
ZimaSpace users documented GPUs that were visible to ZimaOS but unsupported by the container's older PyTorch/CUDA architecture list. This requires a newer image/software stack, not more VRAM.
Can ZimaCube 2 Creator Pack run the ZimaOS Stable Diffusion app?
It is the most direct current Zima NVIDIA hardware path, but do not promise plug-and-play compatibility. Verify the exact RTX PRO 2000 VRAM/generation plus PyTorch CUDA support inside the current app container first.
What sources and further reading informed this Stable Diffusion hardware guide?
IceWhale's compose file establishes the actual ZimaOS package: AUTOMATIC1111-linked, NVIDIA runtime, amd64, port 7860 and persistent model/output paths. AUTOMATIC1111 troubleshooting/optimization docs define the 4 GB low-VRAM behavior. ZimaSpace's 2026 RTX 50-series thread supplies a current package-compatibility warning; Reddit was used for SDXL/VRAM fan-out.
