Immich GPU buying guide
Best GPU for Immich: Transcoding and Machine-Learning Hardware
A GPU can accelerate specific Immich workloads, but buying a discrete card is not automatically the best answer. Integrated media engines and supported accelerators may deliver better efficiency for many home servers.
Quick answer
What to buy for this Immich workload
Choose acceleration by the job: video transcoding, machine-learning inference, or both. First verify the API supported by your Immich deployment and operating system, then buy the smallest hardware tier that meets your concurrency needs.
Live Amazon hardware
Current products that fit the categories in this guide
Listings come from this Immich sprint’s dedicated Amazon catalogue. Accessories, barebones mini PCs, ambiguous NAS bundles, wrong-capacity storage, and mismatched GPU models are excluded by the normalizer.
Buying decision
Choose the system architecture before chasing specifications
Integrated Intel graphics can be compelling for efficient transcoding, while supported discrete GPUs make more sense when machine-learning or concurrent video work justifies the extra power, cooling, and chassis space.
Interactive sizing
Immich GPU Decision Tool
Use your workload to get a hardware tier before comparing products. The result is planning guidance, not a substitute for checking the current Immich compatibility documentation.
Compatibility checklist
Four checks before you purchase
Workload first
Size users, library growth, video share, and background jobs before choosing hardware.
Fast application storage
Keep the database and generated application data on reliable SSD storage where practical.
Protected media capacity
Plan usable capacity after redundancy, reserve space, growth, and backup copies.
Verified acceleration
Confirm the exact hardware, drivers, operating system, and Immich acceleration path before purchase.
Decide whether you need a GPU at all
A discrete GPU is optional for many Immich installations. Before shopping, identify the actual bottleneck: video conversion, machine-learning inference, or slow CPU-only processing during a large import. If the queue already completes fast enough, adding a card may only increase cost and power.
GPU selection begins by naming the accelerated job. Choose acceleration by supported workload, not gaming benchmarks. Verify the current Immich hardware path, drivers, host operating system, container access, power, cooling, and whether a discrete card is actually necessary. In “Decide whether you need a GPU at all,” verify the exact API and device path Immich will use; a card that is fast in games may provide no useful advantage if the software stack cannot reach the relevant engine.
Transcoding versus machine learning
Video transcoding and machine learning use different acceleration stacks. A card that is attractive for one job may be unnecessary for the other. Match the device to the Immich feature you intend to accelerate, then verify driver and container support for that exact path.
Server acceleration has an ownership cost beyond purchase price. Choose acceleration by supported workload, not gaming benchmarks. Verify the current Immich hardware path, drivers, host operating system, container access, power, cooling, and whether a discrete card is actually necessary. “Transcoding versus machine learning” should include slot space, power supply, cooling, driver maintenance, and idle consumption as part of the comparison.
Intel integrated graphics as a baseline
Modern Intel integrated graphics can provide efficient media processing without a separate card. For a compact family server this may be the best balance of idle power, heat, and capability, particularly when video transcoding is more important than heavy machine-learning throughput.
Integrated hardware deserves to be tested before a discrete card is added. Choose acceleration by supported workload, not gaming benchmarks. Verify the current Immich hardware path, drivers, host operating system, container access, power, cooling, and whether a discrete card is actually necessary. For “Intel integrated graphics as a baseline,” start with the supported capability already present in the CPU or iGPU and upgrade only when measured queues or concurrency justify it.
Intel Arc discrete options
Intel Arc cards can add dedicated media and compute resources to systems that lack a suitable iGPU. They still require a compatible PCIe slot, driver stack, adequate airflow, and a host that can expose the device to the Immich containers you actually run.
A separate worker can be more elegant than changing the primary storage host. Choose acceleration by supported workload, not gaming benchmarks. Verify the current Immich hardware path, drivers, host operating system, container access, power, cooling, and whether a discrete card is actually necessary. Use “Intel Arc discrete options” to decide whether acceleration belongs inside the Immich server or on another machine that can be upgraded independently.
NVIDIA CUDA options
NVIDIA cards are relevant where the supported CUDA path aligns with your machine-learning plan or where their media engines fit your transcoding workload. Select the card after checking current Immich requirements rather than assuming every GeForce generation behaves identically.
GPU selection begins by naming the accelerated job. Choose acceleration by supported workload, not gaming benchmarks. Verify the current Immich hardware path, drivers, host operating system, container access, power, cooling, and whether a discrete card is actually necessary. In “NVIDIA CUDA options,” verify the exact API and device path Immich will use; a card that is fast in games may provide no useful advantage if the software stack cannot reach the relevant engine.
VRAM and inference headroom
VRAM matters more as models, batches, or concurrent inference work grow, but buying the largest memory pool is not automatically efficient. For a home photo server, the goal is to clear the job queue quickly enough without turning an occasional workload into a high-power workstation.
Server acceleration has an ownership cost beyond purchase price. Choose acceleration by supported workload, not gaming benchmarks. Verify the current Immich hardware path, drivers, host operating system, container access, power, cooling, and whether a discrete card is actually necessary. “VRAM and inference headroom” should include slot space, power supply, cooling, driver maintenance, and idle consumption as part of the comparison.
Codec support matters more than gaming FPS
Gaming frame rates do not predict Immich value. Codec encode/decode support, hardware API compatibility, inference acceleration, driver stability, and simultaneous-session behavior are more important than raster performance. Compare feature support before comparing benchmark charts.
Integrated hardware deserves to be tested before a discrete card is added. Choose acceleration by supported workload, not gaming benchmarks. Verify the current Immich hardware path, drivers, host operating system, container access, power, cooling, and whether a discrete card is actually necessary. For “Codec support matters more than gaming FPS,” start with the supported capability already present in the CPU or iGPU and upgrade only when measured queues or concurrency justify it.
Driver and container access
Containers need device access. A supported GPU can still be useless if the host OS, VM, or container configuration cannot expose it correctly. Verify the intended Docker or virtualization path before buying the card, especially when the server also runs other services.
A separate worker can be more elegant than changing the primary storage host. Choose acceleration by supported workload, not gaming benchmarks. Verify the current Immich hardware path, drivers, host operating system, container access, power, cooling, and whether a discrete card is actually necessary. Use “Driver and container access” to decide whether acceleration belongs inside the Immich server or on another machine that can be upgraded independently.
PCIe slot, PSU, and chassis constraints
Discrete cards consume physical and electrical resources: a PCIe slot, chassis volume, power connectors, PSU capacity, and cooling. Small NAS enclosures and mini PCs may have no practical upgrade path, which is why architecture should be chosen before the GPU.
GPU selection begins by naming the accelerated job. Choose acceleration by supported workload, not gaming benchmarks. Verify the current Immich hardware path, drivers, host operating system, container access, power, cooling, and whether a discrete card is actually necessary. In “PCIe slot, PSU, and chassis constraints,” verify the exact API and device path Immich will use; a card that is fast in games may provide no useful advantage if the software stack cannot reach the relevant engine.
Idle power and annual running cost
Acceleration hardware often sits idle after a large import finishes. Estimate idle power as part of total ownership cost. A lower-tier card that completes background jobs slightly slower can be the better server component if it saves energy and heat every day.
Server acceleration has an ownership cost beyond purchase price. Choose acceleration by supported workload, not gaming benchmarks. Verify the current Immich hardware path, drivers, host operating system, container access, power, cooling, and whether a discrete card is actually necessary. “Idle power and annual running cost” should include slot space, power supply, cooling, driver maintenance, and idle consumption as part of the comparison.
Used GPU buying risks
Used GPUs can reduce purchase cost, but condition, fan wear, seller support, and uncertain history matter in an always-on server. Compare a used discrete GPU not only with a new card, but also with the cost of moving to a newer mini PC with a capable integrated accelerator.
Integrated hardware deserves to be tested before a discrete card is added. Choose acceleration by supported workload, not gaming benchmarks. Verify the current Immich hardware path, drivers, host operating system, container access, power, cooling, and whether a discrete card is actually necessary. For “Used GPU buying risks,” start with the supported capability already present in the CPU or iGPU and upgrade only when measured queues or concurrency justify it.
When a separate ML worker is cleaner
Immich can place machine-learning services on a separate machine. That can be cleaner when the storage server has no expansion slot or when a powerful GPU is only needed occasionally. A remote worker also lets you upgrade inference hardware without touching the primary media host.
A separate worker can be more elegant than changing the primary storage host. Choose acceleration by supported workload, not gaming benchmarks. Verify the current Immich hardware path, drivers, host operating system, container access, power, cooling, and whether a discrete card is actually necessary. Use “When a separate ML worker is cleaner” to decide whether acceleration belongs inside the Immich server or on another machine that can be upgraded independently.
Questions people ask
Immich hardware questions for this workload
Does Immich need a dedicated GPU?
No. Many Immich servers can use CPU processing or supported integrated acceleration. A discrete GPU is justified when measured transcoding or machine-learning queues need more throughput than the existing hardware can provide. For acceleration hardware, confirm the supported API, device generation, driver path, container access, chassis space, PSU capacity, and idle-power cost before purchase.
Is Intel Quick Sync enough for Immich?
Supported Intel media acceleration can be enough for many home Immich workloads, especially when the main need is efficient transcoding. Verify the exact generation, codecs, drivers, and container access. For acceleration hardware, confirm the supported API, device generation, driver path, container access, chassis space, PSU capacity, and idle-power cost before purchase.
Is Intel Arc good for Immich?
Intel Arc can be useful when its supported media or machine-learning path matches the deployment, but a discrete card is unnecessary if integrated hardware already clears the workload fast enough. For acceleration hardware, confirm the supported API, device generation, driver path, container access, chassis space, PSU capacity, and idle-power cost before purchase.
Is NVIDIA CUDA useful for Immich?
Immich documents a CUDA machine-learning acceleration path for supported NVIDIA hardware. Check the current driver and compute requirements before selecting an exact card. For acceleration hardware, confirm the supported API, device generation, driver path, container access, chassis space, PSU capacity, and idle-power cost before purchase.
How much VRAM does an Immich GPU need?
VRAM needs depend on the supported model, batch behavior, and concurrent work. Home users should choose enough memory for the documented acceleration path rather than buying the largest gaming card by default. For acceleration hardware, confirm the supported API, device generation, driver path, container access, chassis space, PSU capacity, and idle-power cost before purchase.
Does gaming GPU performance matter for Immich?
No. Immich value is driven by supported codec engines, inference APIs, drivers, stability, power, and concurrent workload behavior rather than gaming frame-rate benchmarks. For acceleration hardware, confirm the supported API, device generation, driver path, container access, chassis space, PSU capacity, and idle-power cost before purchase.
Can Immich use a GPU inside Docker?
Yes, when the host exposes the supported accelerator device and drivers to the Immich containers correctly. Hardware support on the host does not automatically mean the container can use it. For acceleration hardware, confirm the supported API, device generation, driver path, container access, chassis space, PSU capacity, and idle-power cost before purchase.
Can I pass a GPU through to an Immich VM?
The right answer depends on library size, video share, users, background-job urgency, storage architecture, and the exact hardware configuration. For acceleration hardware, confirm the supported API, device generation, driver path, container access, chassis space, PSU capacity, and idle-power cost before purchase.
How much power does a discrete GPU add to an always-on server?
The right answer depends on library size, video share, users, background-job urgency, storage architecture, and the exact hardware configuration. For acceleration hardware, confirm the supported API, device generation, driver path, container access, chassis space, PSU capacity, and idle-power cost before purchase.
Can Immich machine learning run on another computer?
Yes. Immich can place machine-learning work on a separate machine, which can be useful when the primary media server is compact, storage-focused, or intentionally low power. For acceleration hardware, confirm the supported API, device generation, driver path, container access, chassis space, PSU capacity, and idle-power cost before purchase.
References and methodology
Verify changing requirements before you buy
Cloudzat separates product classes, rejects accessories and ambiguous configurations, and uses the sprint’s dedicated Amazon catalogue. Prices are displayed only when a current featured offer is returned; otherwise the page links to Amazon buying options without inventing a price.
- Immich requirements
- Immich hardware transcoding
- Immich machine-learning acceleration
- Immich backup and restore
As an Amazon Associate, Cloudzat may earn from qualifying purchases. Product availability, prices, seller terms, and compatibility can change.