Immich video performance
Immich Hardware Transcoding Guide: Intel, NVIDIA and Setup
Hardware transcoding can move expensive video encode and decode work away from the CPU, but it should be planned around the actual camera and phone formats in your library. Immich supports several acceleration APIs, and the right choice depends on existing hardware, codec support, power budget, container access and whether a discrete GPU is justified for the number of videos being processed.
Quick answer
What this Immich workload needs
For most home servers, start with an integrated Intel media engine if you are buying a new compact host. Add a discrete GPU only when the workload, codec path or other accelerated services justify the extra price, power and cooling. Immich currently supports NVENC, Quick Sync, RKMPP and VAAPI for transcoding.
Live Amazon hardware
Current products that fit this decision
These listings come from this sprint's dedicated Amazon catalogue. Barebones systems, laptops, enclosures, adapters, unrelated accessories, ambiguous product classes and obvious multipacks are excluded by the normalizer.
Buying decision
Match the hardware to the bottleneck
Choose acceleration by workload rather than brand loyalty. The planner estimates when integrated media hardware is likely enough and when a discrete card provides meaningful concurrency or flexibility.
Interactive decision tool
Immich Transcoding Hardware Planner
Use the workload inputs to identify a practical hardware tier or likely bottleneck before comparing live products. Results are planning guidance; verify current Immich support and the exact hardware configuration before deployment.
Compatibility checklist
Four checks before you purchase
Use a supported deployment
Immich currently favors Linux and Docker Compose for production; full VMs are supported while Docker-in-LXC is not recommended.
Keep PostgreSQL local
Do not treat a faster network as permission to move the standard database data path onto a network share.
Verify hardware access
Quick Sync, NVENC, VAAPI and ML acceleration need the right host, drivers and container device access.
Measure before upgrading
CPU, RAM, SSD, GPU and network bottlenecks create different symptoms. Change the measured bottleneck first.
Transcoding is a compatibility job first
Immich transcodes video so files can be played more reliably across clients and target formats. Performance matters, but the first question is whether the hardware supports the codecs you actually need. H.264 and HEVC are common, while other formats and HDR tone-mapping paths may impose different constraints. Inventory a sample of real videos before buying a GPU solely from benchmark charts.
Immich exposes multiple hardware APIs
The current documentation lists NVENC for NVIDIA, Quick Sync for Intel, RKMPP for Rockchip and VAAPI across supported hardware. These are not interchangeable labels. Each has its own driver, container and codec requirements, and Immich recommends the more specific API when available for a device. Your deployment method should therefore be decided before choosing the exact accelerator.
Integrated Intel graphics are the default value path
For a new home server, Intel integrated media hardware often provides the simplest balance of power, cost and video capability. It avoids the idle power and slot requirements of a discrete card while leaving the CPU free for other work. This is particularly attractive in mini PCs, but it also works in larger desktops where the iGPU remains enabled and exposed to the container.
NVIDIA is useful when you need more than the iGPU
NVENC can be attractive for higher concurrency, existing NVIDIA hardware, or a host that already needs CUDA for other tasks. Immich also notes that two-pass mode is supported for NVENC while other transcoding APIs ignore that setting. A discrete card adds cost, power, drivers and cooling, so its benefit should be tied to a workload that actually needs those resources.
Hardware decode changes end-to-end CPU load
Immich documents hardware encoding as the default accelerated portion and provides a separate hardware-decoding option. If only the encode stage is accelerated, the CPU can still be busy decoding and tone-mapping. When evaluating CPU utilization, confirm which stages are offloaded before deciding that the accelerator or processor is undersized.
Hardware acceleration can trade efficiency for file size
Immich cautions that hardware-transcoded output can be larger than software-transcoded output at similar settings, often with lower quality. The fastest preset is not automatically the best archival choice. For a large video library, output size affects SSD/HDD consumption and backup time, so quality, codec efficiency and generated-media growth belong in the performance decision.
WSL2 changes the hardware matrix
The current Immich transcoding guidance supports Linux and Windows through WSL2 but explicitly notes that Quick Sync is not supported in WSL2. That means two otherwise identical Intel systems can have different practical acceleration options depending on how Immich is deployed. Do not separate the hardware recommendation from the host operating system and container path.
Raspberry Pi is not a transcoding shortcut
Immich currently states that Raspberry Pi is not supported for its hardware-transcoding feature. A Pi can still be useful in other roles, but it should not be selected because you expect the documented Immich hardware-transcoding path to work like an Intel or NVIDIA host. For video-heavy libraries, this limitation materially changes the server choice.
VRAM matters more for some ML work than simple transcodes
A graphics card can serve both video and machine-learning jobs, but those workloads stress hardware differently. Video encode throughput is not the same thing as model memory. If the same NVIDIA or Intel GPU is expected to accelerate machine learning, budget headroom for the relevant backend and concurrency instead of choosing a card only from a transcoding chart.
PCIe and power supply constraints can eliminate GPUs
Compact servers may lack a full-height slot, auxiliary power connector or thermal capacity for a large graphics card. Low-profile cards can fit some small-form-factor systems, but slot width and cooler clearance still matter. Check the exact chassis and PSU before buying a GPU from a live listing; product suitability depends on the complete host, not just software support.
Use a queue-time target, not a benchmark obsession
Most home users do not need every uploaded video processed instantly. A practical sizing target is that background jobs clear within an acceptable time without making the server feel unresponsive. If occasional overnight processing is fine, integrated graphics may be enough. If multiple users upload large 4K videos all day, more acceleration headroom becomes easier to justify.
Verify the exact device after installation
After configuring acceleration, confirm the hardware is actually being used. Immich recommends checking utilization tools and logs rather than assuming success from a saved setting. This step should happen before spending more money: a configuration error can make a capable GPU look slow because the work is still falling back to software processing.
Questions people ask
Immich Hardware Transcoding questions
What hardware transcoding APIs does Immich support?
Immich currently documents NVENC, Quick Sync, RKMPP and VAAPI for video transcoding.
Is Intel or NVIDIA better for Immich transcoding?
Intel integrated graphics are often the value choice for a new home server. NVIDIA can make sense for heavier concurrency, existing CUDA workloads or specific features.
Does Immich hardware transcoding use less CPU?
Yes for the stages that are offloaded. CPU use can remain significant if decoding or other processing still happens in software.
Can Immich use an AMD GPU?
Immich documents VAAPI as a supported transcoding path across AMD, NVIDIA and Intel devices, subject to hardware and driver support.
Does Immich support Raspberry Pi hardware transcoding?
The current Immich hardware-transcoding documentation says Raspberry Pi is not supported for this feature.
Does hardware transcoding reduce video quality?
It can trade some quality and compression efficiency for speed. Immich notes hardware output may be larger than software output at comparable settings.
How much VRAM do I need?
Simple transcoding and machine learning have different memory needs. Size VRAM for the combined workload if the same card will do both.
Do I need hardware decoding too?
Not always, but enabling supported hardware decoding can reduce CPU load further. Verify the codec path and results on your system.
Can I use Quick Sync in WSL2?
Immich currently says Quick Sync is not supported in WSL2 for its hardware-transcoding setup.
How do I verify NVENC or QSV is working?
Check GPU utilization during active transcodes and review Immich/container logs for errors or software fallback.
Official references and methodology
Verify current Immich support before deployment
Cloudzat’s planner is a workload-sizing aid, not a replacement for Immich’s current hardware support matrix. Live products are refreshed from a dedicated performance catalogue and classified as complete mini PCs or discrete GPUs. Final suitability depends on exact codec support, operating system, drivers, container access, chassis and power limits.
- Immich - Requirements
- Immich - Hardware Transcoding
- Immich - Hardware-Accelerated Machine Learning
- Immich - Remote Machine Learning
- Immich - Docker Compose Installation
- Immich - System Settings
As an Amazon Associate, Cloudzat may earn from qualifying purchases. Prices, seller terms, exact configurations, firmware and software support can change.